rocksdb/db/version_builder.cc

1302 lines
41 KiB
C++
Raw Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "db/version_builder.h"
#include <algorithm>
#include <atomic>
#include <cinttypes>
#include <functional>
#include <map>
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
#include <memory>
#include <set>
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
#include <sstream>
#include <thread>
#include <unordered_map>
#include <unordered_set>
#include <utility>
#include <vector>
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
#include "db/blob/blob_file_meta.h"
#include "db/dbformat.h"
#include "db/internal_stats.h"
#include "db/table_cache.h"
#include "db/version_set.h"
#include "port/port.h"
#include "table/table_reader.h"
#include "util/string_util.h"
namespace ROCKSDB_NAMESPACE {
class VersionBuilder::Rep {
class NewestFirstBySeqNo {
public:
bool operator()(const FileMetaData* lhs, const FileMetaData* rhs) const {
assert(lhs);
assert(rhs);
if (lhs->fd.largest_seqno != rhs->fd.largest_seqno) {
return lhs->fd.largest_seqno > rhs->fd.largest_seqno;
}
if (lhs->fd.smallest_seqno != rhs->fd.smallest_seqno) {
return lhs->fd.smallest_seqno > rhs->fd.smallest_seqno;
}
// Break ties by file number
return lhs->fd.GetNumber() > rhs->fd.GetNumber();
}
};
class BySmallestKey {
public:
explicit BySmallestKey(const InternalKeyComparator* cmp) : cmp_(cmp) {}
bool operator()(const FileMetaData* lhs, const FileMetaData* rhs) const {
assert(lhs);
assert(rhs);
assert(cmp_);
const int r = cmp_->Compare(lhs->smallest, rhs->smallest);
if (r != 0) {
return (r < 0);
}
// Break ties by file number
return (lhs->fd.GetNumber() < rhs->fd.GetNumber());
}
private:
const InternalKeyComparator* cmp_;
};
struct LevelState {
std::unordered_set<uint64_t> deleted_files;
// Map from file number to file meta data.
std::unordered_map<uint64_t, FileMetaData*> added_files;
};
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
// A class that represents the accumulated changes (like additional garbage or
// newly linked/unlinked SST files) for a given blob file after applying a
// series of VersionEdits.
class BlobFileMetaDataDelta {
public:
bool IsEmpty() const {
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
return !additional_garbage_count_ && !additional_garbage_bytes_ &&
newly_linked_ssts_.empty() && newly_unlinked_ssts_.empty();
}
uint64_t GetAdditionalGarbageCount() const {
return additional_garbage_count_;
}
uint64_t GetAdditionalGarbageBytes() const {
return additional_garbage_bytes_;
}
const std::unordered_set<uint64_t>& GetNewlyLinkedSsts() const {
return newly_linked_ssts_;
}
const std::unordered_set<uint64_t>& GetNewlyUnlinkedSsts() const {
return newly_unlinked_ssts_;
}
void AddGarbage(uint64_t count, uint64_t bytes) {
additional_garbage_count_ += count;
additional_garbage_bytes_ += bytes;
}
void LinkSst(uint64_t sst_file_number) {
assert(newly_linked_ssts_.find(sst_file_number) ==
newly_linked_ssts_.end());
// Reconcile with newly unlinked SSTs on the fly. (Note: an SST can be
// linked to and unlinked from the same blob file in the case of a trivial
// move.)
auto it = newly_unlinked_ssts_.find(sst_file_number);
if (it != newly_unlinked_ssts_.end()) {
newly_unlinked_ssts_.erase(it);
} else {
newly_linked_ssts_.emplace(sst_file_number);
}
}
void UnlinkSst(uint64_t sst_file_number) {
assert(newly_unlinked_ssts_.find(sst_file_number) ==
newly_unlinked_ssts_.end());
// Reconcile with newly linked SSTs on the fly. (Note: an SST can be
// linked to and unlinked from the same blob file in the case of a trivial
// move.)
auto it = newly_linked_ssts_.find(sst_file_number);
if (it != newly_linked_ssts_.end()) {
newly_linked_ssts_.erase(it);
} else {
newly_unlinked_ssts_.emplace(sst_file_number);
}
}
private:
uint64_t additional_garbage_count_ = 0;
uint64_t additional_garbage_bytes_ = 0;
std::unordered_set<uint64_t> newly_linked_ssts_;
std::unordered_set<uint64_t> newly_unlinked_ssts_;
};
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
// A class that represents the state of a blob file after applying a series of
// VersionEdits. In addition to the resulting state, it also contains the
// delta (see BlobFileMetaDataDelta above). The resulting state can be used to
// identify obsolete blob files, while the delta makes it possible to
// efficiently detect trivial moves.
class MutableBlobFileMetaData {
public:
// To be used for brand new blob files
explicit MutableBlobFileMetaData(
std::shared_ptr<SharedBlobFileMetaData>&& shared_meta)
: shared_meta_(std::move(shared_meta)) {}
// To be used for pre-existing blob files
explicit MutableBlobFileMetaData(
const std::shared_ptr<BlobFileMetaData>& meta)
: shared_meta_(meta->GetSharedMeta()),
linked_ssts_(meta->GetLinkedSsts()),
garbage_blob_count_(meta->GetGarbageBlobCount()),
garbage_blob_bytes_(meta->GetGarbageBlobBytes()) {}
const std::shared_ptr<SharedBlobFileMetaData>& GetSharedMeta() const {
return shared_meta_;
}
uint64_t GetBlobFileNumber() const {
assert(shared_meta_);
return shared_meta_->GetBlobFileNumber();
}
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
bool HasDelta() const { return !delta_.IsEmpty(); }
const std::unordered_set<uint64_t>& GetLinkedSsts() const {
return linked_ssts_;
}
uint64_t GetGarbageBlobCount() const { return garbage_blob_count_; }
uint64_t GetGarbageBlobBytes() const { return garbage_blob_bytes_; }
bool AddGarbage(uint64_t count, uint64_t bytes) {
assert(shared_meta_);
if (garbage_blob_count_ + count > shared_meta_->GetTotalBlobCount() ||
garbage_blob_bytes_ + bytes > shared_meta_->GetTotalBlobBytes()) {
return false;
}
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
delta_.AddGarbage(count, bytes);
garbage_blob_count_ += count;
garbage_blob_bytes_ += bytes;
return true;
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
}
void LinkSst(uint64_t sst_file_number) {
delta_.LinkSst(sst_file_number);
assert(linked_ssts_.find(sst_file_number) == linked_ssts_.end());
linked_ssts_.emplace(sst_file_number);
}
void UnlinkSst(uint64_t sst_file_number) {
delta_.UnlinkSst(sst_file_number);
assert(linked_ssts_.find(sst_file_number) != linked_ssts_.end());
linked_ssts_.erase(sst_file_number);
}
private:
std::shared_ptr<SharedBlobFileMetaData> shared_meta_;
// Accumulated changes
BlobFileMetaDataDelta delta_;
// Resulting state after applying the changes
BlobFileMetaData::LinkedSsts linked_ssts_;
uint64_t garbage_blob_count_ = 0;
uint64_t garbage_blob_bytes_ = 0;
};
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
const FileOptions& file_options_;
const ImmutableCFOptions* const ioptions_;
TableCache* table_cache_;
VersionStorageInfo* base_vstorage_;
VersionSet* version_set_;
int num_levels_;
LevelState* levels_;
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
// Store sizes of levels larger than num_levels_. We do this instead of
// storing them in levels_ to avoid regression in case there are no files
// on invalid levels. The version is not consistent if in the end the files
// on invalid levels don't cancel out.
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
std::unordered_map<int, size_t> invalid_level_sizes_;
// Whether there are invalid new files or invalid deletion on levels larger
// than num_levels_.
bool has_invalid_levels_;
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
// Current levels of table files affected by additions/deletions.
std::unordered_map<uint64_t, int> table_file_levels_;
NewestFirstBySeqNo level_zero_cmp_;
BySmallestKey level_nonzero_cmp_;
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
// Mutable metadata objects for all blob files affected by the series of
// version edits.
std::map<uint64_t, MutableBlobFileMetaData> mutable_blob_file_metas_;
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
public:
Rep(const FileOptions& file_options, const ImmutableCFOptions* ioptions,
TableCache* table_cache, VersionStorageInfo* base_vstorage,
VersionSet* version_set)
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
: file_options_(file_options),
ioptions_(ioptions),
table_cache_(table_cache),
base_vstorage_(base_vstorage),
version_set_(version_set),
num_levels_(base_vstorage->num_levels()),
has_invalid_levels_(false),
level_nonzero_cmp_(base_vstorage_->InternalComparator()) {
assert(ioptions_);
levels_ = new LevelState[num_levels_];
}
~Rep() {
for (int level = 0; level < num_levels_; level++) {
const auto& added = levels_[level].added_files;
for (auto& pair : added) {
UnrefFile(pair.second);
}
}
delete[] levels_;
}
void UnrefFile(FileMetaData* f) {
f->refs--;
if (f->refs <= 0) {
if (f->table_reader_handle) {
assert(table_cache_ != nullptr);
table_cache_->ReleaseHandle(f->table_reader_handle);
f->table_reader_handle = nullptr;
}
delete f;
}
}
// Mapping used for checking the consistency of links between SST files and
// blob files. It is built using the forward links (table file -> blob file),
// and is subsequently compared with the inverse mapping stored in the
// BlobFileMetaData objects.
using ExpectedLinkedSsts =
std::unordered_map<uint64_t, BlobFileMetaData::LinkedSsts>;
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
static void UpdateExpectedLinkedSsts(
uint64_t table_file_number, uint64_t blob_file_number,
ExpectedLinkedSsts* expected_linked_ssts) {
assert(expected_linked_ssts);
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
if (blob_file_number == kInvalidBlobFileNumber) {
return;
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
}
(*expected_linked_ssts)[blob_file_number].emplace(table_file_number);
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
}
template <typename Checker>
Status CheckConsistencyDetailsForLevel(
const VersionStorageInfo* vstorage, int level, Checker checker,
const std::string& sync_point,
ExpectedLinkedSsts* expected_linked_ssts) const {
#ifdef NDEBUG
(void)sync_point;
#endif
assert(vstorage);
assert(level >= 0 && level < num_levels_);
assert(expected_linked_ssts);
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
const auto& level_files = vstorage->LevelFiles(level);
if (level_files.empty()) {
return Status::OK();
}
assert(level_files[0]);
UpdateExpectedLinkedSsts(level_files[0]->fd.GetNumber(),
level_files[0]->oldest_blob_file_number,
expected_linked_ssts);
for (size_t i = 1; i < level_files.size(); ++i) {
assert(level_files[i]);
UpdateExpectedLinkedSsts(level_files[i]->fd.GetNumber(),
level_files[i]->oldest_blob_file_number,
expected_linked_ssts);
auto lhs = level_files[i - 1];
auto rhs = level_files[i];
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
#ifndef NDEBUG
auto pair = std::make_pair(&lhs, &rhs);
TEST_SYNC_POINT_CALLBACK(sync_point, &pair);
#endif
const Status s = checker(lhs, rhs);
if (!s.ok()) {
return s;
}
}
return Status::OK();
}
// Make sure table files are sorted correctly and that the links between
// table files and blob files are consistent.
Status CheckConsistencyDetails(const VersionStorageInfo* vstorage) const {
assert(vstorage);
ExpectedLinkedSsts expected_linked_ssts;
if (num_levels_ > 0) {
// Check L0
{
auto l0_checker = [this](const FileMetaData* lhs,
const FileMetaData* rhs) {
assert(lhs);
assert(rhs);
if (!level_zero_cmp_(lhs, rhs)) {
std::ostringstream oss;
oss << "L0 files are not sorted properly: files #"
<< lhs->fd.GetNumber() << ", #" << rhs->fd.GetNumber();
return Status::Corruption("VersionBuilder", oss.str());
}
if (rhs->fd.smallest_seqno == rhs->fd.largest_seqno) {
// This is an external file that we ingested
const SequenceNumber external_file_seqno = rhs->fd.smallest_seqno;
if (!(external_file_seqno < lhs->fd.largest_seqno ||
external_file_seqno == 0)) {
std::ostringstream oss;
oss << "L0 file #" << lhs->fd.GetNumber() << " with seqno "
<< lhs->fd.smallest_seqno << ' ' << lhs->fd.largest_seqno
<< " vs. file #" << rhs->fd.GetNumber()
<< " with global_seqno " << external_file_seqno;
return Status::Corruption("VersionBuilder", oss.str());
}
} else if (lhs->fd.smallest_seqno <= rhs->fd.smallest_seqno) {
std::ostringstream oss;
oss << "L0 file #" << lhs->fd.GetNumber() << " with seqno "
<< lhs->fd.smallest_seqno << ' ' << lhs->fd.largest_seqno
<< " vs. file #" << rhs->fd.GetNumber() << " with seqno "
<< rhs->fd.smallest_seqno << ' ' << rhs->fd.largest_seqno;
return Status::Corruption("VersionBuilder", oss.str());
}
return Status::OK();
};
const Status s = CheckConsistencyDetailsForLevel(
vstorage, /* level */ 0, l0_checker,
"VersionBuilder::CheckConsistency0", &expected_linked_ssts);
if (!s.ok()) {
return s;
}
}
// Check L1 and up
const InternalKeyComparator* const icmp = vstorage->InternalComparator();
assert(icmp);
for (int level = 1; level < num_levels_; ++level) {
auto checker = [this, level, icmp](const FileMetaData* lhs,
const FileMetaData* rhs) {
assert(lhs);
assert(rhs);
if (!level_nonzero_cmp_(lhs, rhs)) {
std::ostringstream oss;
oss << 'L' << level << " files are not sorted properly: files #"
<< lhs->fd.GetNumber() << ", #" << rhs->fd.GetNumber();
return Status::Corruption("VersionBuilder", oss.str());
}
// Make sure there is no overlap in level
if (icmp->Compare(lhs->largest, rhs->smallest) >= 0) {
std::ostringstream oss;
oss << 'L' << level << " has overlapping ranges: file #"
<< lhs->fd.GetNumber()
<< " largest key: " << lhs->largest.DebugString(true)
<< " vs. file #" << rhs->fd.GetNumber()
<< " smallest key: " << rhs->smallest.DebugString(true);
return Status::Corruption("VersionBuilder", oss.str());
}
return Status::OK();
};
const Status s = CheckConsistencyDetailsForLevel(
vstorage, level, checker, "VersionBuilder::CheckConsistency1",
&expected_linked_ssts);
if (!s.ok()) {
return s;
}
}
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
// Make sure that all blob files in the version have non-garbage data and
// the links between them and the table files are consistent.
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
const auto& blob_files = vstorage->GetBlobFiles();
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
for (const auto& blob_file_meta : blob_files) {
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
assert(blob_file_meta);
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
const uint64_t blob_file_number = blob_file_meta->GetBlobFileNumber();
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
if (blob_file_meta->GetGarbageBlobCount() >=
blob_file_meta->GetTotalBlobCount()) {
std::ostringstream oss;
oss << "Blob file #" << blob_file_number
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
<< " consists entirely of garbage";
return Status::Corruption("VersionBuilder", oss.str());
}
if (blob_file_meta->GetLinkedSsts() !=
expected_linked_ssts[blob_file_number]) {
std::ostringstream oss;
oss << "Links are inconsistent between table files and blob file #"
<< blob_file_number;
return Status::Corruption("VersionBuilder", oss.str());
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
}
Status ret_s;
TEST_SYNC_POINT_CALLBACK("VersionBuilder::CheckConsistencyBeforeReturn",
&ret_s);
return ret_s;
}
Status CheckConsistency(const VersionStorageInfo* vstorage) const {
assert(vstorage);
// Always run consistency checks in debug build
#ifdef NDEBUG
if (!vstorage->force_consistency_checks()) {
return Status::OK();
}
#endif
Status s = CheckConsistencyDetails(vstorage);
if (s.IsCorruption() && s.getState()) {
// Make it clear the error is due to force_consistency_checks = 1 or
// debug build
#ifdef NDEBUG
auto prefix = "force_consistency_checks";
#else
auto prefix = "force_consistency_checks(DEBUG)";
#endif
s = Status::Corruption(prefix, s.getState());
} else {
// was only expecting corruption with message, or OK
assert(s.ok());
}
return s;
}
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
bool CheckConsistencyForNumLevels() const {
// Make sure there are no files on or beyond num_levels().
if (has_invalid_levels_) {
return false;
}
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
for (const auto& pair : invalid_level_sizes_) {
const size_t level_size = pair.second;
if (level_size != 0) {
return false;
}
}
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
return true;
}
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
bool IsBlobFileInVersion(uint64_t blob_file_number) const {
auto mutable_it = mutable_blob_file_metas_.find(blob_file_number);
if (mutable_it != mutable_blob_file_metas_.end()) {
return true;
}
assert(base_vstorage_);
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
const auto meta = base_vstorage_->GetBlobFileMetaData(blob_file_number);
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
return !!meta;
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
}
MutableBlobFileMetaData* GetOrCreateMutableBlobFileMetaData(
uint64_t blob_file_number) {
auto mutable_it = mutable_blob_file_metas_.find(blob_file_number);
if (mutable_it != mutable_blob_file_metas_.end()) {
return &mutable_it->second;
}
assert(base_vstorage_);
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
const auto meta = base_vstorage_->GetBlobFileMetaData(blob_file_number);
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
if (meta) {
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
mutable_it = mutable_blob_file_metas_
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
.emplace(blob_file_number, MutableBlobFileMetaData(meta))
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
.first;
return &mutable_it->second;
}
return nullptr;
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
Status ApplyBlobFileAddition(const BlobFileAddition& blob_file_addition) {
const uint64_t blob_file_number = blob_file_addition.GetBlobFileNumber();
if (IsBlobFileInVersion(blob_file_number)) {
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
std::ostringstream oss;
oss << "Blob file #" << blob_file_number << " already added";
return Status::Corruption("VersionBuilder", oss.str());
}
// Note: we use C++11 for now but in C++14, this could be done in a more
// elegant way using generalized lambda capture.
VersionSet* const vs = version_set_;
const ImmutableCFOptions* const ioptions = ioptions_;
auto deleter = [vs, ioptions](SharedBlobFileMetaData* shared_meta) {
if (vs) {
assert(ioptions);
assert(!ioptions->cf_paths.empty());
assert(shared_meta);
vs->AddObsoleteBlobFile(shared_meta->GetBlobFileNumber(),
ioptions->cf_paths.front().path);
}
delete shared_meta;
};
auto shared_meta = SharedBlobFileMetaData::Create(
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
blob_file_number, blob_file_addition.GetTotalBlobCount(),
blob_file_addition.GetTotalBlobBytes(),
blob_file_addition.GetChecksumMethod(),
blob_file_addition.GetChecksumValue(), deleter);
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
mutable_blob_file_metas_.emplace(
blob_file_number, MutableBlobFileMetaData(std::move(shared_meta)));
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
return Status::OK();
}
Status ApplyBlobFileGarbage(const BlobFileGarbage& blob_file_garbage) {
const uint64_t blob_file_number = blob_file_garbage.GetBlobFileNumber();
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
MutableBlobFileMetaData* const mutable_meta =
GetOrCreateMutableBlobFileMetaData(blob_file_number);
if (!mutable_meta) {
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
std::ostringstream oss;
oss << "Blob file #" << blob_file_number << " not found";
return Status::Corruption("VersionBuilder", oss.str());
}
if (!mutable_meta->AddGarbage(blob_file_garbage.GetGarbageBlobCount(),
blob_file_garbage.GetGarbageBlobBytes())) {
std::ostringstream oss;
oss << "Garbage overflow for blob file #" << blob_file_number;
return Status::Corruption("VersionBuilder", oss.str());
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
return Status::OK();
}
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
int GetCurrentLevelForTableFile(uint64_t file_number) const {
auto it = table_file_levels_.find(file_number);
if (it != table_file_levels_.end()) {
return it->second;
}
assert(base_vstorage_);
return base_vstorage_->GetFileLocation(file_number).GetLevel();
}
uint64_t GetOldestBlobFileNumberForTableFile(int level,
uint64_t file_number) const {
assert(level < num_levels_);
const auto& added_files = levels_[level].added_files;
auto it = added_files.find(file_number);
if (it != added_files.end()) {
const FileMetaData* const meta = it->second;
assert(meta);
return meta->oldest_blob_file_number;
}
assert(base_vstorage_);
const FileMetaData* const meta =
base_vstorage_->GetFileMetaDataByNumber(file_number);
assert(meta);
return meta->oldest_blob_file_number;
}
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
Status ApplyFileDeletion(int level, uint64_t file_number) {
assert(level != VersionStorageInfo::FileLocation::Invalid().GetLevel());
const int current_level = GetCurrentLevelForTableFile(file_number);
if (level != current_level) {
if (level >= num_levels_) {
has_invalid_levels_ = true;
}
std::ostringstream oss;
oss << "Cannot delete table file #" << file_number << " from level "
<< level << " since it is ";
if (current_level ==
VersionStorageInfo::FileLocation::Invalid().GetLevel()) {
oss << "not in the LSM tree";
} else {
oss << "on level " << current_level;
}
return Status::Corruption("VersionBuilder", oss.str());
}
if (level >= num_levels_) {
assert(invalid_level_sizes_[level] > 0);
--invalid_level_sizes_[level];
table_file_levels_[file_number] =
VersionStorageInfo::FileLocation::Invalid().GetLevel();
return Status::OK();
}
const uint64_t blob_file_number =
GetOldestBlobFileNumberForTableFile(level, file_number);
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
if (blob_file_number != kInvalidBlobFileNumber) {
MutableBlobFileMetaData* const mutable_meta =
GetOrCreateMutableBlobFileMetaData(blob_file_number);
if (mutable_meta) {
mutable_meta->UnlinkSst(file_number);
}
}
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
auto& level_state = levels_[level];
auto& add_files = level_state.added_files;
auto add_it = add_files.find(file_number);
if (add_it != add_files.end()) {
UnrefFile(add_it->second);
add_files.erase(add_it);
}
auto& del_files = level_state.deleted_files;
assert(del_files.find(file_number) == del_files.end());
del_files.emplace(file_number);
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
table_file_levels_[file_number] =
VersionStorageInfo::FileLocation::Invalid().GetLevel();
return Status::OK();
}
Status ApplyFileAddition(int level, const FileMetaData& meta) {
assert(level != VersionStorageInfo::FileLocation::Invalid().GetLevel());
const uint64_t file_number = meta.fd.GetNumber();
const int current_level = GetCurrentLevelForTableFile(file_number);
if (current_level !=
VersionStorageInfo::FileLocation::Invalid().GetLevel()) {
if (level >= num_levels_) {
has_invalid_levels_ = true;
}
std::ostringstream oss;
oss << "Cannot add table file #" << file_number << " to level " << level
<< " since it is already in the LSM tree on level " << current_level;
return Status::Corruption("VersionBuilder", oss.str());
}
if (level >= num_levels_) {
++invalid_level_sizes_[level];
table_file_levels_[file_number] = level;
return Status::OK();
}
auto& level_state = levels_[level];
auto& del_files = level_state.deleted_files;
auto del_it = del_files.find(file_number);
if (del_it != del_files.end()) {
del_files.erase(del_it);
}
FileMetaData* const f = new FileMetaData(meta);
f->refs = 1;
auto& add_files = level_state.added_files;
assert(add_files.find(file_number) == add_files.end());
add_files.emplace(file_number, f);
const uint64_t blob_file_number = f->oldest_blob_file_number;
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
if (blob_file_number != kInvalidBlobFileNumber) {
MutableBlobFileMetaData* const mutable_meta =
GetOrCreateMutableBlobFileMetaData(blob_file_number);
if (mutable_meta) {
mutable_meta->LinkSst(file_number);
}
}
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
table_file_levels_[file_number] = level;
return Status::OK();
}
// Apply all of the edits in *edit to the current state.
Status Apply(const VersionEdit* edit) {
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
{
const Status s = CheckConsistency(base_vstorage_);
if (!s.ok()) {
return s;
}
}
// Note: we process the blob file related changes first because the
// table file addition/deletion logic depends on the blob files
// already being there.
// Add new blob files
for (const auto& blob_file_addition : edit->GetBlobFileAdditions()) {
const Status s = ApplyBlobFileAddition(blob_file_addition);
if (!s.ok()) {
return s;
}
}
// Increase the amount of garbage for blob files affected by GC
for (const auto& blob_file_garbage : edit->GetBlobFileGarbages()) {
const Status s = ApplyBlobFileGarbage(blob_file_garbage);
if (!s.ok()) {
return s;
}
}
// Delete table files
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
for (const auto& deleted_file : edit->GetDeletedFiles()) {
const int level = deleted_file.first;
const uint64_t file_number = deleted_file.second;
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
const Status s = ApplyFileDeletion(level, file_number);
if (!s.ok()) {
return s;
}
}
// Add new table files
for (const auto& new_file : edit->GetNewFiles()) {
const int level = new_file.first;
Improve consistency checks in VersionBuilder (#6901) Summary: The patch cleans up the code and improves the consistency checks around adding/deleting table files in `VersionBuilder`. Namely, it makes the checks stricter and improves them in the following ways: 1) A table file can now only be deleted from the LSM tree using the level it resides on. Earlier, there was some unnecessary wiggle room for trivially moved files (they could be deleted using a lower level number than the actual one). 2) A table file cannot be added to the tree if it is already present in the tree on any level (not just the target level). The earlier code only had an assertion (which is a no-op in release builds) that the newly added file is not already present on the target level. 3) The above consistency checks around state transitions are now mandatory, as opposed to the earlier `CheckConsistencyForDeletes`, which was a no-op in release mode unless `force_consistency_checks` was set to `true`. The rationale here is that assuming that the initial state is consistent, a valid transition leads to a next state that is also consistent; however, an *invalid* transition offers no such guarantee. Hence it makes sense to validate the transitions unconditionally, and save `force_consistency_checks` for the paranoid checks that re-validate the entire state. 4) The new checks build on the mechanism introduced in https://github.com/facebook/rocksdb/pull/6862, which enables us to efficiently look up the location (level and position within level) of files in a `Version` by file number. This makes the consistency checks much more efficient than the earlier `CheckConsistencyForDeletes`, which essentially performed a linear search. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6901 Test Plan: Extended the unit tests and ran: `make check` `make whitebox_crash_test` Reviewed By: ajkr Differential Revision: D21822714 Pulled By: ltamasi fbshipit-source-id: e2b29c8b6da1bf0f59004acc889e4870b2d18215
2020-06-03 20:21:16 +02:00
const FileMetaData& meta = new_file.second;
const Status s = ApplyFileAddition(level, meta);
if (!s.ok()) {
return s;
}
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
return Status::OK();
}
// Helper function template for merging the blob file metadata from the base
// version with the mutable metadata representing the state after applying the
// edits. The function objects process_base and process_mutable are
// respectively called to handle a base version object when there is no
// matching mutable object, and a mutable object when there is no matching
// base version object. process_both is called to perform the merge when a
// given blob file appears both in the base version and the mutable list. The
// helper stops processing objects if a function object returns false. Blob
// files with a file number below first_blob_file are not processed.
template <typename ProcessBase, typename ProcessMutable, typename ProcessBoth>
void MergeBlobFileMetas(uint64_t first_blob_file, ProcessBase process_base,
ProcessMutable process_mutable,
ProcessBoth process_both) const {
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
assert(base_vstorage_);
Clean up blob files based on the linked SST set (#7001) Summary: The earlier `VersionBuilder` code only cleaned up blob files that were marked as entirely consisting of garbage using `VersionEdits` with `BlobFileGarbage`. This covers the cases when table files go through regular compaction, where we iterate through the KVs and thus have an opportunity to calculate the amount of garbage (that is, most cases). However, it does not help when table files are simply dropped (e.g. deletion compactions or the `DeleteFile` API). To deal with such cases, the patch adds logic that cleans up all blob files at the head of the list until the first one with linked SSTs is found. (As an example, let's assume we have blob files with numbers 1..10, and the first one with any linked SSTs is number 8. This means that SSTs in the `Version` only rely on blob files with numbers >= 8, and thus 1..7 are no longer needed.) The code change itself is pretty small; however, changing the logic like this necessitated changes to some tests that have been added recently (namely to the ones that use blob files in isolation, i.e. without any table files referring to them). Some of these cases were fixed by bypassing `VersionBuilder` altogether in order to keep the tests simple (which actually makes them more proper unit tests as well), while the `VersionBuilder` unit tests were fixed by adding dummy table files to the test cases as needed. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7001 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D22119474 Pulled By: ltamasi fbshipit-source-id: c6547141355667d4291d9661d6518eb741e7b54a
2020-07-01 00:30:01 +02:00
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
auto base_it = base_vstorage_->GetBlobFileMetaDataLB(first_blob_file);
const auto base_it_end = base_vstorage_->GetBlobFiles().end();
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
auto mutable_it = mutable_blob_file_metas_.lower_bound(first_blob_file);
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
const auto mutable_it_end = mutable_blob_file_metas_.end();
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
while (base_it != base_it_end && mutable_it != mutable_it_end) {
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
const auto& base_meta = *base_it;
assert(base_meta);
const uint64_t base_blob_file_number = base_meta->GetBlobFileNumber();
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
const uint64_t mutable_blob_file_number = mutable_it->first;
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
if (base_blob_file_number < mutable_blob_file_number) {
if (!process_base(base_meta)) {
return;
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
++base_it;
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
} else if (mutable_blob_file_number < base_blob_file_number) {
const auto& mutable_meta = mutable_it->second;
if (!process_mutable(mutable_meta)) {
return;
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
++mutable_it;
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
} else {
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
assert(base_blob_file_number == mutable_blob_file_number);
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
const auto& mutable_meta = mutable_it->second;
if (!process_both(base_meta, mutable_meta)) {
return;
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
++base_it;
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
++mutable_it;
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
}
}
while (base_it != base_it_end) {
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
const auto& base_meta = *base_it;
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
if (!process_base(base_meta)) {
return;
}
Clean up blob files based on the linked SST set (#7001) Summary: The earlier `VersionBuilder` code only cleaned up blob files that were marked as entirely consisting of garbage using `VersionEdits` with `BlobFileGarbage`. This covers the cases when table files go through regular compaction, where we iterate through the KVs and thus have an opportunity to calculate the amount of garbage (that is, most cases). However, it does not help when table files are simply dropped (e.g. deletion compactions or the `DeleteFile` API). To deal with such cases, the patch adds logic that cleans up all blob files at the head of the list until the first one with linked SSTs is found. (As an example, let's assume we have blob files with numbers 1..10, and the first one with any linked SSTs is number 8. This means that SSTs in the `Version` only rely on blob files with numbers >= 8, and thus 1..7 are no longer needed.) The code change itself is pretty small; however, changing the logic like this necessitated changes to some tests that have been added recently (namely to the ones that use blob files in isolation, i.e. without any table files referring to them). Some of these cases were fixed by bypassing `VersionBuilder` altogether in order to keep the tests simple (which actually makes them more proper unit tests as well), while the `VersionBuilder` unit tests were fixed by adding dummy table files to the test cases as needed. Pull Request resolved: https://github.com/facebook/rocksdb/pull/7001 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D22119474 Pulled By: ltamasi fbshipit-source-id: c6547141355667d4291d9661d6518eb741e7b54a
2020-07-01 00:30:01 +02:00
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
++base_it;
}
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
while (mutable_it != mutable_it_end) {
const auto& mutable_meta = mutable_it->second;
if (!process_mutable(mutable_meta)) {
return;
}
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
Aggregate blob file related changes in VersionBuilder as VersionEdits are applied (#9085) Summary: The current VersionBuilder code on mainline keeps track of blob file related changes ("delta") induced by a series of `VersionEdit`s in the form of `BlobFileMetaDataDelta` objects. Specifically, `BlobFileMetaDataDelta` contains the amount of additional garbage generated by compactions, as well as the set of newly linked/unlinked SSTs. This is very handy for detecting trivial moves, since in that case the newly linked and unlinked SSTs cancel each other out. However, this representation does not allow us to easily tell whether a certain blob file is obsolete after applying a set of `VersionEdit`s or not. In order to solve this issue, the patch introduces `MutableBlobFileMetaData`, which, in addition to the delta, also contains the materialized state after applying a set of version edits (i.e. the total amount of garbage and the resulting set of linked SSTs). This will enable us to add further consistency checks and to improve certain pieces of functionality where knowing up front which blob files get obsoleted is beneficial. (Note: this patch is just the refactoring part; I plan to create separate PRs for the enhancements.) Pull Request resolved: https://github.com/facebook/rocksdb/pull/9085 Test Plan: Ran `make check` and the stress tests in BlobDB mode. Reviewed By: riversand963 Differential Revision: D31980867 Pulled By: ltamasi fbshipit-source-id: cc4286778b10900af720423d6b772c77f28a93e3
2021-10-30 02:45:35 +02:00
++mutable_it;
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
}
}
// Helper function template for finding the first blob file that has linked
// SSTs.
template <typename Meta>
static bool CheckLinkedSsts(const Meta& meta,
uint64_t* min_oldest_blob_file_num) {
assert(min_oldest_blob_file_num);
if (!meta.GetLinkedSsts().empty()) {
assert(*min_oldest_blob_file_num == kInvalidBlobFileNumber);
*min_oldest_blob_file_num = meta.GetBlobFileNumber();
return false;
}
return true;
}
// Find the oldest blob file that has linked SSTs.
uint64_t GetMinOldestBlobFileNumber() const {
uint64_t min_oldest_blob_file_num = kInvalidBlobFileNumber;
auto process_base =
[&min_oldest_blob_file_num](
const std::shared_ptr<BlobFileMetaData>& base_meta) {
assert(base_meta);
return CheckLinkedSsts(*base_meta, &min_oldest_blob_file_num);
};
auto process_mutable = [&min_oldest_blob_file_num](
const MutableBlobFileMetaData& mutable_meta) {
return CheckLinkedSsts(mutable_meta, &min_oldest_blob_file_num);
};
auto process_both = [&min_oldest_blob_file_num](
const std::shared_ptr<BlobFileMetaData>& base_meta,
const MutableBlobFileMetaData& mutable_meta) {
#ifndef NDEBUG
assert(base_meta);
assert(base_meta->GetSharedMeta() == mutable_meta.GetSharedMeta());
#else
(void)base_meta;
#endif
// Look at mutable_meta since it supersedes *base_meta
return CheckLinkedSsts(mutable_meta, &min_oldest_blob_file_num);
};
MergeBlobFileMetas(kInvalidBlobFileNumber, process_base, process_mutable,
process_both);
return min_oldest_blob_file_num;
}
static std::shared_ptr<BlobFileMetaData> CreateBlobFileMetaData(
const MutableBlobFileMetaData& mutable_meta) {
return BlobFileMetaData::Create(
mutable_meta.GetSharedMeta(), mutable_meta.GetLinkedSsts(),
mutable_meta.GetGarbageBlobCount(), mutable_meta.GetGarbageBlobBytes());
}
// Add the blob file specified by meta to *vstorage if it is determined to
// contain valid data (blobs).
template <typename Meta>
static void AddBlobFileIfNeeded(VersionStorageInfo* vstorage, Meta&& meta) {
assert(vstorage);
assert(meta);
if (meta->GetLinkedSsts().empty() &&
meta->GetGarbageBlobCount() >= meta->GetTotalBlobCount()) {
return;
}
vstorage->AddBlobFile(std::forward<Meta>(meta));
}
// Merge the blob file metadata from the base version with the changes (edits)
// applied, and save the result into *vstorage.
void SaveBlobFilesTo(VersionStorageInfo* vstorage) const {
assert(vstorage);
Use a sorted vector instead of a map to store blob file metadata (#9526) Summary: The patch replaces `std::map` with a sorted `std::vector` for `VersionStorageInfo::blob_files_` and preallocates the space for the `vector` before saving the `BlobFileMetaData` into the new `VersionStorageInfo` in `VersionBuilder::Rep::SaveBlobFilesTo`. These changes reduce the time the DB mutex is held while saving new `Version`s, and using a sorted `vector` also makes lookups faster thanks to better memory locality. In addition, the patch introduces helper methods `VersionStorageInfo::GetBlobFileMetaData` and `VersionStorageInfo::GetBlobFileMetaDataLB` that can be used by clients to perform lookups in the `vector`, and does some general cleanup in the parts of code where blob file metadata are used. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9526 Test Plan: Ran `make check` and the crash test script for a while. Performance was tested using a load-optimized benchmark (`fillseq` with vector memtable, no WAL) and small file sizes so that a significant number of files are produced: ``` numactl --interleave=all ./db_bench --benchmarks=fillseq --allow_concurrent_memtable_write=false --level0_file_num_compaction_trigger=4 --level0_slowdown_writes_trigger=20 --level0_stop_writes_trigger=30 --max_background_jobs=8 --max_write_buffer_number=8 --db=/data/ltamasi-dbbench --wal_dir=/data/ltamasi-dbbench --num=800000000 --num_levels=8 --key_size=20 --value_size=400 --block_size=8192 --cache_size=51539607552 --cache_numshardbits=6 --compression_max_dict_bytes=0 --compression_ratio=0.5 --compression_type=lz4 --bytes_per_sync=8388608 --cache_index_and_filter_blocks=1 --cache_high_pri_pool_ratio=0.5 --benchmark_write_rate_limit=0 --write_buffer_size=16777216 --target_file_size_base=16777216 --max_bytes_for_level_base=67108864 --verify_checksum=1 --delete_obsolete_files_period_micros=62914560 --max_bytes_for_level_multiplier=8 --statistics=0 --stats_per_interval=1 --stats_interval_seconds=20 --histogram=1 --memtablerep=skip_list --bloom_bits=10 --open_files=-1 --subcompactions=1 --compaction_style=0 --min_level_to_compress=3 --level_compaction_dynamic_level_bytes=true --pin_l0_filter_and_index_blocks_in_cache=1 --soft_pending_compaction_bytes_limit=167503724544 --hard_pending_compaction_bytes_limit=335007449088 --min_level_to_compress=0 --use_existing_db=0 --sync=0 --threads=1 --memtablerep=vector --allow_concurrent_memtable_write=false --disable_wal=1 --enable_blob_files=1 --blob_file_size=16777216 --min_blob_size=0 --blob_compression_type=lz4 --enable_blob_garbage_collection=1 --seed=<some value> ``` Final statistics before the patch: ``` Cumulative writes: 0 writes, 700M keys, 0 commit groups, 0.0 writes per commit group, ingest: 284.62 GB, 121.27 MB/s Interval writes: 0 writes, 334K keys, 0 commit groups, 0.0 writes per commit group, ingest: 139.28 MB, 72.46 MB/s ``` With the patch: ``` Cumulative writes: 0 writes, 760M keys, 0 commit groups, 0.0 writes per commit group, ingest: 308.66 GB, 131.52 MB/s Interval writes: 0 writes, 445K keys, 0 commit groups, 0.0 writes per commit group, ingest: 185.35 MB, 93.15 MB/s ``` Total time to complete the benchmark is 2611 seconds with the patch, down from 2986 secs. Reviewed By: riversand963 Differential Revision: D34082728 Pulled By: ltamasi fbshipit-source-id: fc598abf676dce436734d06bb9d2d99a26a004fc
2022-02-09 21:35:39 +01:00
assert(base_vstorage_);
vstorage->ReserveBlob(base_vstorage_->GetBlobFiles().size() +
mutable_blob_file_metas_.size());
const uint64_t oldest_blob_file_with_linked_ssts =
GetMinOldestBlobFileNumber();
auto process_base =
[vstorage](const std::shared_ptr<BlobFileMetaData>& base_meta) {
assert(base_meta);
AddBlobFileIfNeeded(vstorage, base_meta);
return true;
};
auto process_mutable =
[vstorage](const MutableBlobFileMetaData& mutable_meta) {
AddBlobFileIfNeeded(vstorage, CreateBlobFileMetaData(mutable_meta));
return true;
};
auto process_both = [vstorage](
const std::shared_ptr<BlobFileMetaData>& base_meta,
const MutableBlobFileMetaData& mutable_meta) {
assert(base_meta);
assert(base_meta->GetSharedMeta() == mutable_meta.GetSharedMeta());
if (!mutable_meta.HasDelta()) {
assert(base_meta->GetGarbageBlobCount() ==
mutable_meta.GetGarbageBlobCount());
assert(base_meta->GetGarbageBlobBytes() ==
mutable_meta.GetGarbageBlobBytes());
assert(base_meta->GetLinkedSsts() == mutable_meta.GetLinkedSsts());
AddBlobFileIfNeeded(vstorage, base_meta);
return true;
}
AddBlobFileIfNeeded(vstorage, CreateBlobFileMetaData(mutable_meta));
return true;
};
MergeBlobFileMetas(oldest_blob_file_with_linked_ssts, process_base,
process_mutable, process_both);
}
void MaybeAddFile(VersionStorageInfo* vstorage, int level,
FileMetaData* f) const {
const uint64_t file_number = f->fd.GetNumber();
const auto& level_state = levels_[level];
const auto& del_files = level_state.deleted_files;
const auto del_it = del_files.find(file_number);
if (del_it != del_files.end()) {
// f is to-be-deleted table file
vstorage->RemoveCurrentStats(f);
} else {
const auto& add_files = level_state.added_files;
const auto add_it = add_files.find(file_number);
// Note: if the file appears both in the base version and in the added
// list, the added FileMetaData supersedes the one in the base version.
if (add_it != add_files.end() && add_it->second != f) {
vstorage->RemoveCurrentStats(f);
} else {
vstorage->AddFile(level, f);
}
}
}
template <typename Cmp>
void SaveSSTFilesTo(VersionStorageInfo* vstorage, int level, Cmp cmp) const {
// Merge the set of added files with the set of pre-existing files.
// Drop any deleted files. Store the result in *vstorage.
const auto& base_files = base_vstorage_->LevelFiles(level);
const auto& unordered_added_files = levels_[level].added_files;
vstorage->Reserve(level, base_files.size() + unordered_added_files.size());
// Sort added files for the level.
std::vector<FileMetaData*> added_files;
added_files.reserve(unordered_added_files.size());
for (const auto& pair : unordered_added_files) {
added_files.push_back(pair.second);
}
std::sort(added_files.begin(), added_files.end(), cmp);
auto base_iter = base_files.begin();
auto base_end = base_files.end();
auto added_iter = added_files.begin();
auto added_end = added_files.end();
while (added_iter != added_end || base_iter != base_end) {
if (base_iter == base_end ||
(added_iter != added_end && cmp(*added_iter, *base_iter))) {
MaybeAddFile(vstorage, level, *added_iter++);
} else {
MaybeAddFile(vstorage, level, *base_iter++);
}
}
}
void SaveSSTFilesTo(VersionStorageInfo* vstorage) const {
assert(vstorage);
if (!num_levels_) {
return;
}
SaveSSTFilesTo(vstorage, /* level */ 0, level_zero_cmp_);
for (int level = 1; level < num_levels_; ++level) {
SaveSSTFilesTo(vstorage, level, level_nonzero_cmp_);
}
}
// Save the current state in *vstorage.
Status SaveTo(VersionStorageInfo* vstorage) const {
Status s = CheckConsistency(base_vstorage_);
if (!s.ok()) {
return s;
}
s = CheckConsistency(vstorage);
if (!s.ok()) {
return s;
}
SaveSSTFilesTo(vstorage);
Add blob files to VersionStorageInfo/VersionBuilder (#6597) Summary: The patch adds a couple of classes to represent metadata about blob files: `SharedBlobFileMetaData` contains the information elements that are immutable (once the blob file is closed), e.g. blob file number, total number and size of blob files, checksum method/value, while `BlobFileMetaData` contains attributes that can vary across versions like the amount of garbage in the file. There is a single `SharedBlobFileMetaData` for each blob file, which is jointly owned by the `BlobFileMetaData` objects that point to it; `BlobFileMetaData` objects, in turn, are owned by `Version`s and can also be shared if the (immutable _and_ mutable) state of the blob file is the same in two versions. In addition, the patch adds the blob file metadata to `VersionStorageInfo`, and extends `VersionBuilder` so that it can apply blob file related `VersionEdit`s (i.e. those containing `BlobFileAddition`s and/or `BlobFileGarbage`), and save blob file metadata to a new `VersionStorageInfo`. Consistency checks are also extended to ensure that table files point to blob files that are part of the `Version`, and that all blob files that are part of any given `Version` have at least some _non_-garbage data in them. Pull Request resolved: https://github.com/facebook/rocksdb/pull/6597 Test Plan: `make check` Reviewed By: riversand963 Differential Revision: D20656803 Pulled By: ltamasi fbshipit-source-id: f1f74d135045b3b42d0146f03ee576ef0a4bfd80
2020-03-27 02:48:55 +01:00
SaveBlobFilesTo(vstorage);
s = CheckConsistency(vstorage);
return s;
}
Fast path for detecting unchanged prefix_extractor (#9407) Summary: Fixes a major performance regression in 6.26, where extra CPU is spent in SliceTransform::AsString when reads involve a prefix_extractor (Get, MultiGet, Seek). Common case performance is now better than 6.25. This change creates a "fast path" for verifying that the current prefix extractor is unchanged and compatible with what was used to generate a table file. This fast path detects the common case by pointer comparison on the current prefix_extractor and a "known good" prefix extractor (if applicable) that is saved at the time the table reader is opened. The "known good" prefix extractor is saved as another shared_ptr copy (in an existing field, however) to ensure the pointer is not recycled. When the prefix_extractor has changed to a different instance but same compatible configuration (rare, odd), performance is still a regression compared to 6.25, but this is likely acceptable because of the oddity of such a case. The performance of incompatible prefix_extractor is essentially unchanged. Also fixed a minor case (ForwardIterator) where a prefix_extractor could be used via a raw pointer after being freed as a shared_ptr, if replaced via SetOptions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9407 Test Plan: ## Performance Populate DB with `TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=10000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=12` Running head-to-head comparisons simultaneously with `TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -use_existing_db -readonly -benchmarks=seekrandom -num=10000000 -duration=20 -disable_wal=1 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=12` Below each is compared by ops/sec vs. baseline which is version 6.25 (multiple baseline runs because of variable machine load) v6.26: 4833 vs. 6698 (<- major regression!) v6.27: 4737 vs. 6397 (still) New: 6704 vs. 6461 (better than baseline in common case) Disabled fastpath: 4843 vs. 6389 (e.g. if prefix extractor instance changes but is still compatible) Changed prefix size (no usable filter) in new: 787 vs. 5927 Changed prefix size (no usable filter) in new & baseline: 773 vs. 784 Reviewed By: mrambacher Differential Revision: D33677812 Pulled By: pdillinger fbshipit-source-id: 571d9711c461fb97f957378a061b7e7dbc4d6a76
2022-01-21 20:36:36 +01:00
Status LoadTableHandlers(
InternalStats* internal_stats, int max_threads,
bool prefetch_index_and_filter_in_cache, bool is_initial_load,
const std::shared_ptr<const SliceTransform>& prefix_extractor,
size_t max_file_size_for_l0_meta_pin) {
assert(table_cache_ != nullptr);
size_t table_cache_capacity = table_cache_->get_cache()->GetCapacity();
bool always_load = (table_cache_capacity == TableCache::kInfiniteCapacity);
size_t max_load = port::kMaxSizet;
if (!always_load) {
// If it is initial loading and not set to always loading all the
// files, we only load up to kInitialLoadLimit files, to limit the
// time reopening the DB.
const size_t kInitialLoadLimit = 16;
size_t load_limit;
// If the table cache is not 1/4 full, we pin the table handle to
// file metadata to avoid the cache read costs when reading the file.
// The downside of pinning those files is that LRU won't be followed
// for those files. This doesn't matter much because if number of files
// of the DB excceeds table cache capacity, eventually no table reader
// will be pinned and LRU will be followed.
if (is_initial_load) {
load_limit = std::min(kInitialLoadLimit, table_cache_capacity / 4);
} else {
load_limit = table_cache_capacity / 4;
}
size_t table_cache_usage = table_cache_->get_cache()->GetUsage();
if (table_cache_usage >= load_limit) {
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
// TODO (yanqin) find a suitable status code.
return Status::OK();
} else {
max_load = load_limit - table_cache_usage;
}
}
// <file metadata, level>
std::vector<std::pair<FileMetaData*, int>> files_meta;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
std::vector<Status> statuses;
for (int level = 0; level < num_levels_; level++) {
for (auto& file_meta_pair : levels_[level].added_files) {
auto* file_meta = file_meta_pair.second;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
// If the file has been opened before, just skip it.
if (!file_meta->table_reader_handle) {
files_meta.emplace_back(file_meta, level);
statuses.emplace_back(Status::OK());
}
if (files_meta.size() >= max_load) {
break;
}
}
if (files_meta.size() >= max_load) {
break;
}
}
std::atomic<size_t> next_file_meta_idx(0);
std::function<void()> load_handlers_func([&]() {
while (true) {
size_t file_idx = next_file_meta_idx.fetch_add(1);
if (file_idx >= files_meta.size()) {
break;
}
auto* file_meta = files_meta[file_idx].first;
int level = files_meta[file_idx].second;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
statuses[file_idx] = table_cache_->FindTable(
ReadOptions(), file_options_,
*(base_vstorage_->InternalComparator()), file_meta->fd,
&file_meta->table_reader_handle, prefix_extractor, false /*no_io */,
true /* record_read_stats */,
internal_stats->GetFileReadHist(level), false, level,
prefetch_index_and_filter_in_cache, max_file_size_for_l0_meta_pin,
file_meta->temperature);
if (file_meta->table_reader_handle != nullptr) {
// Load table_reader
file_meta->fd.table_reader = table_cache_->GetTableReaderFromHandle(
file_meta->table_reader_handle);
}
}
});
std::vector<port::Thread> threads;
for (int i = 1; i < max_threads; i++) {
threads.emplace_back(load_handlers_func);
}
load_handlers_func();
for (auto& t : threads) {
t.join();
}
Status ret;
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
for (const auto& s : statuses) {
if (!s.ok()) {
if (ret.ok()) {
ret = s;
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
}
}
return ret;
}
};
Introduce a new storage specific Env API (#5761) Summary: The current Env API encompasses both storage/file operations, as well as OS related operations. Most of the APIs return a Status, which does not have enough metadata about an error, such as whether its retry-able or not, scope (i.e fault domain) of the error etc., that may be required in order to properly handle a storage error. The file APIs also do not provide enough control over the IO SLA, such as timeout, prioritization, hinting about placement and redundancy etc. This PR separates out the file/storage APIs from Env into a new FileSystem class. The APIs are updated to return an IOStatus with metadata about the error, as well as to take an IOOptions structure as input in order to allow more control over the IO. The user can set both ```options.env``` and ```options.file_system``` to specify that RocksDB should use the former for OS related operations and the latter for storage operations. Internally, a ```CompositeEnvWrapper``` has been introduced that inherits from ```Env``` and redirects individual methods to either an ```Env``` implementation or the ```FileSystem``` as appropriate. When options are sanitized during ```DB::Open```, ```options.env``` is replaced with a newly allocated ```CompositeEnvWrapper``` instance if both env and file_system have been specified. This way, the rest of the RocksDB code can continue to function as before. This PR also ports PosixEnv to the new API by splitting it into two - PosixEnv and PosixFileSystem. PosixEnv is defined as a sub-class of CompositeEnvWrapper, and threading/time functions are overridden with Posix specific implementations in order to avoid an extra level of indirection. The ```CompositeEnvWrapper``` translates ```IOStatus``` return code to ```Status```, and sets the severity to ```kSoftError``` if the io_status is retryable. The error handling code in RocksDB can then recover the DB automatically. Pull Request resolved: https://github.com/facebook/rocksdb/pull/5761 Differential Revision: D18868376 Pulled By: anand1976 fbshipit-source-id: 39efe18a162ea746fabac6360ff529baba48486f
2019-12-13 23:47:08 +01:00
VersionBuilder::VersionBuilder(const FileOptions& file_options,
const ImmutableCFOptions* ioptions,
TableCache* table_cache,
VersionStorageInfo* base_vstorage,
VersionSet* version_set)
: rep_(new Rep(file_options, ioptions, table_cache, base_vstorage,
version_set)) {}
VersionBuilder::~VersionBuilder() = default;
bool VersionBuilder::CheckConsistencyForNumLevels() {
return rep_->CheckConsistencyForNumLevels();
}
Status VersionBuilder::Apply(const VersionEdit* edit) {
return rep_->Apply(edit);
}
Status VersionBuilder::SaveTo(VersionStorageInfo* vstorage) const {
return rep_->SaveTo(vstorage);
}
Support for single-primary, multi-secondary instances (#4899) Summary: This PR allows RocksDB to run in single-primary, multi-secondary process mode. The writer is a regular RocksDB (e.g. an `DBImpl`) instance playing the role of a primary. Multiple `DBImplSecondary` processes (secondaries) share the same set of SST files, MANIFEST, WAL files with the primary. Secondaries tail the MANIFEST of the primary and apply updates to their own in-memory state of the file system, e.g. `VersionStorageInfo`. This PR has several components: 1. (Originally in #4745). Add a `PathNotFound` subcode to `IOError` to denote the failure when a secondary tries to open a file which has been deleted by the primary. 2. (Similar to #4602). Add `FragmentBufferedReader` to handle partially-read, trailing record at the end of a log from where future read can continue. 3. (Originally in #4710 and #4820). Add implementation of the secondary, i.e. `DBImplSecondary`. 3.1 Tail the primary's MANIFEST during recovery. 3.2 Tail the primary's MANIFEST during normal processing by calling `ReadAndApply`. 3.3 Tailing WAL will be in a future PR. 4. Add an example in 'examples/multi_processes_example.cc' to demonstrate the usage of secondary RocksDB instance in a multi-process setting. Instructions to run the example can be found at the beginning of the source code. Pull Request resolved: https://github.com/facebook/rocksdb/pull/4899 Differential Revision: D14510945 Pulled By: riversand963 fbshipit-source-id: 4ac1c5693e6012ad23f7b4b42d3c374fecbe8886
2019-03-27 00:41:31 +01:00
Status VersionBuilder::LoadTableHandlers(
InternalStats* internal_stats, int max_threads,
bool prefetch_index_and_filter_in_cache, bool is_initial_load,
Fast path for detecting unchanged prefix_extractor (#9407) Summary: Fixes a major performance regression in 6.26, where extra CPU is spent in SliceTransform::AsString when reads involve a prefix_extractor (Get, MultiGet, Seek). Common case performance is now better than 6.25. This change creates a "fast path" for verifying that the current prefix extractor is unchanged and compatible with what was used to generate a table file. This fast path detects the common case by pointer comparison on the current prefix_extractor and a "known good" prefix extractor (if applicable) that is saved at the time the table reader is opened. The "known good" prefix extractor is saved as another shared_ptr copy (in an existing field, however) to ensure the pointer is not recycled. When the prefix_extractor has changed to a different instance but same compatible configuration (rare, odd), performance is still a regression compared to 6.25, but this is likely acceptable because of the oddity of such a case. The performance of incompatible prefix_extractor is essentially unchanged. Also fixed a minor case (ForwardIterator) where a prefix_extractor could be used via a raw pointer after being freed as a shared_ptr, if replaced via SetOptions. Pull Request resolved: https://github.com/facebook/rocksdb/pull/9407 Test Plan: ## Performance Populate DB with `TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -benchmarks=fillrandom -num=10000000 -disable_wal=1 -write_buffer_size=10000000 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=12` Running head-to-head comparisons simultaneously with `TEST_TMPDIR=/dev/shm/rocksdb ./db_bench -use_existing_db -readonly -benchmarks=seekrandom -num=10000000 -duration=20 -disable_wal=1 -bloom_bits=16 -compaction_style=2 -fifo_compaction_max_table_files_size_mb=10000 -fifo_compaction_allow_compaction=0 -prefix_size=12` Below each is compared by ops/sec vs. baseline which is version 6.25 (multiple baseline runs because of variable machine load) v6.26: 4833 vs. 6698 (<- major regression!) v6.27: 4737 vs. 6397 (still) New: 6704 vs. 6461 (better than baseline in common case) Disabled fastpath: 4843 vs. 6389 (e.g. if prefix extractor instance changes but is still compatible) Changed prefix size (no usable filter) in new: 787 vs. 5927 Changed prefix size (no usable filter) in new & baseline: 773 vs. 784 Reviewed By: mrambacher Differential Revision: D33677812 Pulled By: pdillinger fbshipit-source-id: 571d9711c461fb97f957378a061b7e7dbc4d6a76
2022-01-21 20:36:36 +01:00
const std::shared_ptr<const SliceTransform>& prefix_extractor,
size_t max_file_size_for_l0_meta_pin) {
return rep_->LoadTableHandlers(
internal_stats, max_threads, prefetch_index_and_filter_in_cache,
is_initial_load, prefix_extractor, max_file_size_for_l0_meta_pin);
}
uint64_t VersionBuilder::GetMinOldestBlobFileNumber() const {
return rep_->GetMinOldestBlobFileNumber();
}
BaseReferencedVersionBuilder::BaseReferencedVersionBuilder(
ColumnFamilyData* cfd)
: version_builder_(new VersionBuilder(
cfd->current()->version_set()->file_options(), cfd->ioptions(),
cfd->table_cache(), cfd->current()->storage_info(),
cfd->current()->version_set())),
version_(cfd->current()) {
version_->Ref();
}
BaseReferencedVersionBuilder::BaseReferencedVersionBuilder(
ColumnFamilyData* cfd, Version* v)
: version_builder_(new VersionBuilder(
cfd->current()->version_set()->file_options(), cfd->ioptions(),
cfd->table_cache(), v->storage_info(), v->version_set())),
version_(v) {
assert(version_ != cfd->current());
}
BaseReferencedVersionBuilder::~BaseReferencedVersionBuilder() {
version_->Unref();
}
} // namespace ROCKSDB_NAMESPACE