2016-02-10 00:12:00 +01:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
2017-07-16 01:03:42 +02:00
|
|
|
// 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).
|
2013-10-16 23:59:46 +02:00
|
|
|
//
|
2011-03-18 23:37:00 +01:00
|
|
|
// 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.
|
|
|
|
|
2013-10-05 07:32:05 +02:00
|
|
|
#pragma once
|
2011-03-18 23:37:00 +01:00
|
|
|
#include <stdio.h>
|
Introduce a new MultiGet batching implementation (#5011)
Summary:
This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching.
Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to -
1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch()
2. Bloom filter cachelines can be prefetched, hiding the cache miss latency
The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress.
Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32).
Batch Sizes
1 | 2 | 4 | 8 | 16 | 32
Random pattern (Stride length 0)
4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get
4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching)
4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching)
Good locality (Stride length 16)
4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753
4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781
4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135
Good locality (Stride length 256)
4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232
4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268
4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62
Medium locality (Stride length 4096)
4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555
4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465
4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891
dbbench command used (on a DB with 4 levels, 12 million keys)-
TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011
Differential Revision: D14348703
Pulled By: anand1976
fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
|
|
|
#include <memory>
|
2014-07-23 21:31:11 +02:00
|
|
|
#include <string>
|
2016-08-16 17:16:04 +02:00
|
|
|
#include <utility>
|
Introduce a new MultiGet batching implementation (#5011)
Summary:
This PR introduces a new MultiGet() API, with the underlying implementation grouping keys based on SST file and batching lookups in a file. The reason for the new API is twofold - the definition allows callers to allocate storage for status and values on stack instead of std::vector, as well as return values as PinnableSlices in order to avoid copying, and it keeps the original MultiGet() implementation intact while we experiment with batching.
Batching is useful when there is some spatial locality to the keys being queries, as well as larger batch sizes. The main benefits are due to -
1. Fewer function calls, especially to BlockBasedTableReader::MultiGet() and FullFilterBlockReader::KeysMayMatch()
2. Bloom filter cachelines can be prefetched, hiding the cache miss latency
The next step is to optimize the binary searches in the level_storage_info, index blocks and data blocks, since we could reduce the number of key comparisons if the keys are relatively close to each other. The batching optimizations also need to be extended to other formats, such as PlainTable and filter formats. This also needs to be added to db_stress.
Benchmark results from db_bench for various batch size/locality of reference combinations are given below. Locality was simulated by offsetting the keys in a batch by a stride length. Each SST file is about 8.6MB uncompressed and key/value size is 16/100 uncompressed. To focus on the cpu benefit of batching, the runs were single threaded and bound to the same cpu to eliminate interference from other system events. The results show a 10-25% improvement in micros/op from smaller to larger batch sizes (4 - 32).
Batch Sizes
1 | 2 | 4 | 8 | 16 | 32
Random pattern (Stride length 0)
4.158 | 4.109 | 4.026 | 4.05 | 4.1 | 4.074 - Get
4.438 | 4.302 | 4.165 | 4.122 | 4.096 | 4.075 - MultiGet (no batching)
4.461 | 4.256 | 4.277 | 4.11 | 4.182 | 4.14 - MultiGet (w/ batching)
Good locality (Stride length 16)
4.048 | 3.659 | 3.248 | 2.99 | 2.84 | 2.753
4.429 | 3.728 | 3.406 | 3.053 | 2.911 | 2.781
4.452 | 3.45 | 2.833 | 2.451 | 2.233 | 2.135
Good locality (Stride length 256)
4.066 | 3.786 | 3.581 | 3.447 | 3.415 | 3.232
4.406 | 4.005 | 3.644 | 3.49 | 3.381 | 3.268
4.393 | 3.649 | 3.186 | 2.882 | 2.676 | 2.62
Medium locality (Stride length 4096)
4.012 | 3.922 | 3.768 | 3.61 | 3.582 | 3.555
4.364 | 4.057 | 3.791 | 3.65 | 3.57 | 3.465
4.479 | 3.758 | 3.316 | 3.077 | 2.959 | 2.891
dbbench command used (on a DB with 4 levels, 12 million keys)-
TEST_TMPDIR=/dev/shm numactl -C 10 ./db_bench.tmp -use_existing_db=true -benchmarks="readseq,multireadrandom" -write_buffer_size=4194304 -target_file_size_base=4194304 -max_bytes_for_level_base=16777216 -num=12000000 -reads=12000000 -duration=90 -threads=1 -compression_type=none -cache_size=4194304000 -batch_size=32 -disable_auto_compactions=true -bloom_bits=10 -cache_index_and_filter_blocks=true -pin_l0_filter_and_index_blocks_in_cache=true -multiread_batched=true -multiread_stride=4
Pull Request resolved: https://github.com/facebook/rocksdb/pull/5011
Differential Revision: D14348703
Pulled By: anand1976
fbshipit-source-id: 774406dab3776d979c809522a67bedac6c17f84b
2019-04-11 23:24:09 +02:00
|
|
|
#include "db/lookup_key.h"
|
|
|
|
#include "db/merge_context.h"
|
2019-06-01 02:19:43 +02:00
|
|
|
#include "logging/logging.h"
|
2018-03-23 21:16:37 +01:00
|
|
|
#include "monitoring/perf_context_imp.h"
|
2013-08-23 17:38:13 +02:00
|
|
|
#include "rocksdb/comparator.h"
|
|
|
|
#include "rocksdb/db.h"
|
|
|
|
#include "rocksdb/filter_policy.h"
|
|
|
|
#include "rocksdb/slice.h"
|
2014-04-10 23:19:43 +02:00
|
|
|
#include "rocksdb/slice_transform.h"
|
2013-10-29 01:54:09 +01:00
|
|
|
#include "rocksdb/table.h"
|
2013-08-23 17:38:13 +02:00
|
|
|
#include "rocksdb/types.h"
|
2011-03-18 23:37:00 +01:00
|
|
|
#include "util/coding.h"
|
2019-03-27 18:24:16 +01:00
|
|
|
#include "util/user_comparator_wrapper.h"
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2013-10-04 06:49:15 +02:00
|
|
|
namespace rocksdb {
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2019-05-24 01:16:38 +02:00
|
|
|
// The file declares data structures and functions that deal with internal
|
|
|
|
// keys.
|
|
|
|
// Each internal key contains a user key, a sequence number (SequenceNumber)
|
|
|
|
// and a type (ValueType), and they are usually encoded together.
|
|
|
|
// There are some related helper classes here.
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
class InternalKey;
|
|
|
|
|
|
|
|
// Value types encoded as the last component of internal keys.
|
|
|
|
// DO NOT CHANGE THESE ENUM VALUES: they are embedded in the on-disk
|
|
|
|
// data structures.
|
2014-01-27 22:53:22 +01:00
|
|
|
// The highest bit of the value type needs to be reserved to SST tables
|
|
|
|
// for them to do more flexible encoding.
|
|
|
|
enum ValueType : unsigned char {
|
2011-03-18 23:37:00 +01:00
|
|
|
kTypeDeletion = 0x0,
|
2013-03-21 23:59:47 +01:00
|
|
|
kTypeValue = 0x1,
|
2013-08-15 01:32:46 +02:00
|
|
|
kTypeMerge = 0x2,
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
kTypeLogData = 0x3, // WAL only.
|
|
|
|
kTypeColumnFamilyDeletion = 0x4, // WAL only.
|
|
|
|
kTypeColumnFamilyValue = 0x5, // WAL only.
|
|
|
|
kTypeColumnFamilyMerge = 0x6, // WAL only.
|
|
|
|
kTypeSingleDeletion = 0x7,
|
|
|
|
kTypeColumnFamilySingleDeletion = 0x8, // WAL only.
|
Modification of WriteBatch to support two phase commit
Summary: Adds three new WriteBatch data types: Prepare(xid), Commit(xid), Rollback(xid). Prepare(xid) should precede the (single) operation to which is applies. There can obviously be multiple Prepare(xid) markers. There should only be one Rollback(xid) or Commit(xid) marker yet not both. None of this logic is currently enforced and will most likely be implemented further up such as in the memtableinserter. All three markers are similar to PutLogData in that they are writebatch meta-data, ie stored but not counted. All three markers differ from PutLogData in that they will actually be written to disk. As for WriteBatchWithIndex, Prepare, Commit, Rollback are all implemented just as PutLogData and none are tested just as PutLogData.
Test Plan: single unit test in write_batch_test.
Reviewers: hermanlee4, sdong, anthony
Subscribers: leveldb, dhruba, vasilep, andrewkr
Differential Revision: https://reviews.facebook.net/D57867
2016-04-08 08:35:51 +02:00
|
|
|
kTypeBeginPrepareXID = 0x9, // WAL only.
|
|
|
|
kTypeEndPrepareXID = 0xA, // WAL only.
|
|
|
|
kTypeCommitXID = 0xB, // WAL only.
|
|
|
|
kTypeRollbackXID = 0xC, // WAL only.
|
|
|
|
kTypeNoop = 0xD, // WAL only.
|
2016-08-16 17:16:04 +02:00
|
|
|
kTypeColumnFamilyRangeDeletion = 0xE, // WAL only.
|
|
|
|
kTypeRangeDeletion = 0xF, // meta block
|
2017-10-03 18:08:07 +02:00
|
|
|
kTypeColumnFamilyBlobIndex = 0x10, // Blob DB only
|
|
|
|
kTypeBlobIndex = 0x11, // Blob DB only
|
2017-11-11 20:23:43 +01:00
|
|
|
// When the prepared record is also persisted in db, we use a different
|
|
|
|
// record. This is to ensure that the WAL that is generated by a WritePolicy
|
|
|
|
// is not mistakenly read by another, which would result into data
|
|
|
|
// inconsistency.
|
|
|
|
kTypeBeginPersistedPrepareXID = 0x12, // WAL only.
|
2018-06-29 03:46:39 +02:00
|
|
|
// Similar to kTypeBeginPersistedPrepareXID, this is to ensure that WAL
|
|
|
|
// generated by WriteUnprepared write policy is not mistakenly read by
|
|
|
|
// another.
|
2018-07-07 02:17:36 +02:00
|
|
|
kTypeBeginUnprepareXID = 0x13, // WAL only.
|
|
|
|
kMaxValue = 0x7F // Not used for storing records.
|
2011-03-18 23:37:00 +01:00
|
|
|
};
|
2014-01-27 22:53:22 +01:00
|
|
|
|
2016-09-28 03:20:57 +02:00
|
|
|
// Defined in dbformat.cc
|
|
|
|
extern const ValueType kValueTypeForSeek;
|
|
|
|
extern const ValueType kValueTypeForSeekForPrev;
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
|
2016-08-16 17:16:04 +02:00
|
|
|
// Checks whether a type is an inline value type
|
|
|
|
// (i.e. a type used in memtable skiplist and sst file datablock).
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
inline bool IsValueType(ValueType t) {
|
2017-10-03 18:08:07 +02:00
|
|
|
return t <= kTypeMerge || t == kTypeSingleDeletion || t == kTypeBlobIndex;
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2016-08-16 17:16:04 +02:00
|
|
|
// Checks whether a type is from user operation
|
|
|
|
// kTypeRangeDeletion is in meta block so this API is separated from above
|
|
|
|
inline bool IsExtendedValueType(ValueType t) {
|
2016-08-20 00:10:31 +02:00
|
|
|
return IsValueType(t) || t == kTypeRangeDeletion;
|
2016-08-16 17:16:04 +02:00
|
|
|
}
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
// We leave eight bits empty at the bottom so a type and sequence#
|
|
|
|
// can be packed together into 64-bits.
|
2019-03-28 00:13:08 +01:00
|
|
|
static const SequenceNumber kMaxSequenceNumber = ((0x1ull << 56) - 1);
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2016-10-19 01:59:37 +02:00
|
|
|
static const SequenceNumber kDisableGlobalSequenceNumber = port::kMaxUint64;
|
|
|
|
|
2019-05-24 01:16:38 +02:00
|
|
|
// The data structure that represents an internal key in the way that user_key,
|
|
|
|
// sequence number and type are stored in separated forms.
|
2011-03-18 23:37:00 +01:00
|
|
|
struct ParsedInternalKey {
|
|
|
|
Slice user_key;
|
|
|
|
SequenceNumber sequence;
|
|
|
|
ValueType type;
|
|
|
|
|
2017-06-29 00:36:11 +02:00
|
|
|
ParsedInternalKey()
|
|
|
|
: sequence(kMaxSequenceNumber) // Make code analyzer happy
|
2019-03-28 00:13:08 +01:00
|
|
|
{} // Intentionally left uninitialized (for speed)
|
2011-03-18 23:37:00 +01:00
|
|
|
ParsedInternalKey(const Slice& u, const SequenceNumber& seq, ValueType t)
|
2019-03-28 00:13:08 +01:00
|
|
|
: user_key(u), sequence(seq), type(t) {}
|
2012-12-16 03:28:36 +01:00
|
|
|
std::string DebugString(bool hex = false) const;
|
2017-07-24 20:28:20 +02:00
|
|
|
|
|
|
|
void clear() {
|
|
|
|
user_key.clear();
|
|
|
|
sequence = 0;
|
|
|
|
type = kTypeDeletion;
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
};
|
|
|
|
|
|
|
|
// Return the length of the encoding of "key".
|
|
|
|
inline size_t InternalKeyEncodingLength(const ParsedInternalKey& key) {
|
|
|
|
return key.user_key.size() + 8;
|
|
|
|
}
|
|
|
|
|
2015-05-29 23:36:35 +02:00
|
|
|
// Pack a sequence number and a ValueType into a uint64_t
|
2014-04-01 23:45:30 +02:00
|
|
|
extern uint64_t PackSequenceAndType(uint64_t seq, ValueType t);
|
|
|
|
|
2015-05-29 23:36:35 +02:00
|
|
|
// Given the result of PackSequenceAndType, store the sequence number in *seq
|
|
|
|
// and the ValueType in *t.
|
|
|
|
extern void UnPackSequenceAndType(uint64_t packed, uint64_t* seq, ValueType* t);
|
|
|
|
|
2017-11-02 06:52:17 +01:00
|
|
|
EntryType GetEntryType(ValueType value_type);
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
// Append the serialization of "key" to *result.
|
|
|
|
extern void AppendInternalKey(std::string* result,
|
|
|
|
const ParsedInternalKey& key);
|
2016-12-01 16:00:17 +01:00
|
|
|
// Serialized internal key consists of user key followed by footer.
|
|
|
|
// This function appends the footer to *result, assuming that *result already
|
|
|
|
// contains the user key at the end.
|
|
|
|
extern void AppendInternalKeyFooter(std::string* result, SequenceNumber s,
|
|
|
|
ValueType t);
|
2011-03-18 23:37:00 +01:00
|
|
|
|
|
|
|
// Attempt to parse an internal key from "internal_key". On success,
|
|
|
|
// stores the parsed data in "*result", and returns true.
|
|
|
|
//
|
|
|
|
// On error, returns false, leaves "*result" in an undefined state.
|
|
|
|
extern bool ParseInternalKey(const Slice& internal_key,
|
|
|
|
ParsedInternalKey* result);
|
|
|
|
|
|
|
|
// Returns the user key portion of an internal key.
|
|
|
|
inline Slice ExtractUserKey(const Slice& internal_key) {
|
|
|
|
assert(internal_key.size() >= 8);
|
|
|
|
return Slice(internal_key.data(), internal_key.size() - 8);
|
|
|
|
}
|
|
|
|
|
2019-06-06 08:07:28 +02:00
|
|
|
inline Slice ExtractUserKeyAndStripTimestamp(const Slice& internal_key,
|
|
|
|
size_t ts_sz) {
|
|
|
|
assert(internal_key.size() >= 8 + ts_sz);
|
|
|
|
return Slice(internal_key.data(), internal_key.size() - 8 - ts_sz);
|
|
|
|
}
|
|
|
|
|
|
|
|
inline Slice StripTimestampFromUserKey(const Slice& user_key, size_t ts_sz) {
|
|
|
|
assert(user_key.size() >= ts_sz);
|
|
|
|
return Slice(user_key.data(), user_key.size() - ts_sz);
|
|
|
|
}
|
|
|
|
|
2018-07-14 02:34:54 +02:00
|
|
|
inline uint64_t ExtractInternalKeyFooter(const Slice& internal_key) {
|
2011-03-18 23:37:00 +01:00
|
|
|
assert(internal_key.size() >= 8);
|
|
|
|
const size_t n = internal_key.size();
|
2018-07-14 02:34:54 +02:00
|
|
|
return DecodeFixed64(internal_key.data() + n - 8);
|
|
|
|
}
|
|
|
|
|
|
|
|
inline ValueType ExtractValueType(const Slice& internal_key) {
|
|
|
|
uint64_t num = ExtractInternalKeyFooter(internal_key);
|
2011-03-18 23:37:00 +01:00
|
|
|
unsigned char c = num & 0xff;
|
|
|
|
return static_cast<ValueType>(c);
|
|
|
|
}
|
|
|
|
|
|
|
|
// A comparator for internal keys that uses a specified comparator for
|
|
|
|
// the user key portion and breaks ties by decreasing sequence number.
|
2017-09-11 20:53:22 +02:00
|
|
|
class InternalKeyComparator
|
|
|
|
#ifdef NDEBUG
|
|
|
|
final
|
|
|
|
#endif
|
|
|
|
: public Comparator {
|
2011-03-18 23:37:00 +01:00
|
|
|
private:
|
2019-03-27 18:24:16 +01:00
|
|
|
UserComparatorWrapper user_comparator_;
|
2013-06-10 22:28:58 +02:00
|
|
|
std::string name_;
|
2019-03-28 00:13:08 +01:00
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
public:
|
2019-03-27 18:24:16 +01:00
|
|
|
explicit InternalKeyComparator(const Comparator* c)
|
|
|
|
: user_comparator_(c),
|
|
|
|
name_("rocksdb.InternalKeyComparator:" +
|
|
|
|
std::string(user_comparator_.Name())) {}
|
2014-01-27 22:53:22 +01:00
|
|
|
virtual ~InternalKeyComparator() {}
|
2013-06-10 22:28:58 +02:00
|
|
|
|
2015-02-26 20:28:41 +01:00
|
|
|
virtual const char* Name() const override;
|
|
|
|
virtual int Compare(const Slice& a, const Slice& b) const override;
|
2018-02-01 03:45:49 +01:00
|
|
|
// Same as Compare except that it excludes the value type from comparison
|
|
|
|
virtual int CompareKeySeq(const Slice& a, const Slice& b) const;
|
2015-02-26 20:28:41 +01:00
|
|
|
virtual void FindShortestSeparator(std::string* start,
|
|
|
|
const Slice& limit) const override;
|
|
|
|
virtual void FindShortSuccessor(std::string* key) const override;
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2019-03-27 18:24:16 +01:00
|
|
|
const Comparator* user_comparator() const {
|
|
|
|
return user_comparator_.user_comparator();
|
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
|
|
|
|
int Compare(const InternalKey& a, const InternalKey& b) const;
|
2014-01-27 22:53:22 +01:00
|
|
|
int Compare(const ParsedInternalKey& a, const ParsedInternalKey& b) const;
|
2017-07-29 01:23:50 +02:00
|
|
|
virtual const Comparator* GetRootComparator() const override {
|
2019-03-27 18:24:16 +01:00
|
|
|
return user_comparator_.GetRootComparator();
|
2017-07-29 01:23:50 +02:00
|
|
|
}
|
2011-03-18 23:37:00 +01:00
|
|
|
};
|
|
|
|
|
2019-05-24 01:16:38 +02:00
|
|
|
// The class represent the internal key in encoded form.
|
2011-03-18 23:37:00 +01:00
|
|
|
class InternalKey {
|
|
|
|
private:
|
|
|
|
std::string rep_;
|
2019-03-28 00:13:08 +01:00
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
public:
|
2019-03-28 00:13:08 +01:00
|
|
|
InternalKey() {} // Leave rep_ as empty to indicate it is invalid
|
2014-11-06 20:14:28 +01:00
|
|
|
InternalKey(const Slice& _user_key, SequenceNumber s, ValueType t) {
|
|
|
|
AppendInternalKey(&rep_, ParsedInternalKey(_user_key, s, t));
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2015-04-24 03:08:37 +02:00
|
|
|
// sets the internal key to be bigger or equal to all internal keys with this
|
|
|
|
// user key
|
|
|
|
void SetMaxPossibleForUserKey(const Slice& _user_key) {
|
2017-09-13 02:16:44 +02:00
|
|
|
AppendInternalKey(
|
|
|
|
&rep_, ParsedInternalKey(_user_key, 0, static_cast<ValueType>(0)));
|
2015-04-24 03:08:37 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
// sets the internal key to be smaller or equal to all internal keys with this
|
|
|
|
// user key
|
|
|
|
void SetMinPossibleForUserKey(const Slice& _user_key) {
|
2017-09-13 02:16:44 +02:00
|
|
|
AppendInternalKey(&rep_, ParsedInternalKey(_user_key, kMaxSequenceNumber,
|
|
|
|
kValueTypeForSeek));
|
2015-04-24 03:08:37 +02:00
|
|
|
}
|
|
|
|
|
[fix] SIGSEGV when VersionEdit in MANIFEST is corrupted
Summary:
This was reported by our customers in task #4295529.
Cause:
* MANIFEST file contains a VersionEdit, which contains file entries whose 'smallest' and 'largest' internal keys are empty. String with zero characters. Root cause of corruption was not investigated. We should report corruption when this happens. However, we currently SIGSEGV.
Here's what happens:
* VersionEdit encodes zero-strings happily and stores them in smallest and largest InternalKeys. InternalKey::Encode() does assert when `rep_.empty()`, but we don't assert in production environemnts. Also, we should never assert as a result of DB corruption.
* As part of our ConsistencyCheck, we call GetLiveFilesMetaData()
* GetLiveFilesMetadata() calls `file->largest.user_key().ToString()`
* user_key() function does: 1. assert(size > 8) (ooops, no assert), 2. returns `Slice(internal_key.data(), internal_key.size() - 8)`
* since `internal_key.size()` is unsigned int, this call translates to `Slice(whatever, 1298471928561892576182756)`. Bazinga.
Fix:
* VersionEdit checks if InternalKey is valid in `VersionEdit::GetInternalKey()`. If it's invalid, returns corruption.
Lessons learned:
* Always keep in mind that even if you `assert()`, production code will continue execution even if assert fails.
* Never `assert` based on DB corruption. Assert only if the code should guarantee that assert can't fail.
Test Plan: dumped offending manifest. Before: assert. Now: corruption
Reviewers: dhruba, haobo, sdong
Reviewed By: dhruba
CC: leveldb
Differential Revision: https://reviews.facebook.net/D18507
2014-05-08 01:52:12 +02:00
|
|
|
bool Valid() const {
|
|
|
|
ParsedInternalKey parsed;
|
|
|
|
return ParseInternalKey(Slice(rep_), &parsed);
|
|
|
|
}
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
void DecodeFrom(const Slice& s) { rep_.assign(s.data(), s.size()); }
|
|
|
|
Slice Encode() const {
|
|
|
|
assert(!rep_.empty());
|
|
|
|
return rep_;
|
|
|
|
}
|
|
|
|
|
|
|
|
Slice user_key() const { return ExtractUserKey(rep_); }
|
2015-09-10 23:35:25 +02:00
|
|
|
size_t size() { return rep_.size(); }
|
2011-03-18 23:37:00 +01:00
|
|
|
|
2016-06-13 18:57:43 +02:00
|
|
|
void Set(const Slice& _user_key, SequenceNumber s, ValueType t) {
|
|
|
|
SetFrom(ParsedInternalKey(_user_key, s, t));
|
|
|
|
}
|
|
|
|
|
2011-03-18 23:37:00 +01:00
|
|
|
void SetFrom(const ParsedInternalKey& p) {
|
|
|
|
rep_.clear();
|
|
|
|
AppendInternalKey(&rep_, p);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Clear() { rep_.clear(); }
|
2011-10-06 01:30:28 +02:00
|
|
|
|
2016-12-01 16:00:17 +01:00
|
|
|
// The underlying representation.
|
|
|
|
// Intended only to be used together with ConvertFromUserKey().
|
|
|
|
std::string* rep() { return &rep_; }
|
|
|
|
|
|
|
|
// Assuming that *rep() contains a user key, this method makes internal key
|
|
|
|
// out of it in-place. This saves a memcpy compared to Set()/SetFrom().
|
|
|
|
void ConvertFromUserKey(SequenceNumber s, ValueType t) {
|
|
|
|
AppendInternalKeyFooter(&rep_, s, t);
|
|
|
|
}
|
|
|
|
|
2012-12-16 03:28:36 +01:00
|
|
|
std::string DebugString(bool hex = false) const;
|
2011-03-18 23:37:00 +01:00
|
|
|
};
|
|
|
|
|
2019-03-28 00:13:08 +01:00
|
|
|
inline int InternalKeyComparator::Compare(const InternalKey& a,
|
|
|
|
const InternalKey& b) const {
|
2011-03-18 23:37:00 +01:00
|
|
|
return Compare(a.Encode(), b.Encode());
|
|
|
|
}
|
|
|
|
|
|
|
|
inline bool ParseInternalKey(const Slice& internal_key,
|
|
|
|
ParsedInternalKey* result) {
|
|
|
|
const size_t n = internal_key.size();
|
|
|
|
if (n < 8) return false;
|
|
|
|
uint64_t num = DecodeFixed64(internal_key.data() + n - 8);
|
|
|
|
unsigned char c = num & 0xff;
|
|
|
|
result->sequence = num >> 8;
|
|
|
|
result->type = static_cast<ValueType>(c);
|
2014-01-27 22:53:22 +01:00
|
|
|
assert(result->type <= ValueType::kMaxValue);
|
2011-03-18 23:37:00 +01:00
|
|
|
result->user_key = Slice(internal_key.data(), n - 8);
|
2016-08-16 17:16:04 +02:00
|
|
|
return IsExtendedValueType(result->type);
|
2011-03-18 23:37:00 +01:00
|
|
|
}
|
|
|
|
|
2015-07-14 09:21:41 +02:00
|
|
|
// Update the sequence number in the internal key.
|
|
|
|
// Guarantees not to invalidate ikey.data().
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
inline void UpdateInternalKey(std::string* ikey, uint64_t seq, ValueType t) {
|
2015-07-14 09:21:41 +02:00
|
|
|
size_t ikey_sz = ikey->size();
|
|
|
|
assert(ikey_sz >= 8);
|
2013-02-15 23:31:24 +01:00
|
|
|
uint64_t newval = (seq << 8) | t;
|
2015-07-14 09:21:41 +02:00
|
|
|
|
|
|
|
// Note: Since C++11, strings are guaranteed to be stored contiguously and
|
|
|
|
// string::operator[]() is guaranteed not to change ikey.data().
|
|
|
|
EncodeFixed64(&(*ikey)[ikey_sz - 8], newval);
|
2013-02-15 23:31:24 +01:00
|
|
|
}
|
|
|
|
|
2013-06-14 07:09:08 +02:00
|
|
|
// Get the sequence number from the internal key
|
|
|
|
inline uint64_t GetInternalKeySeqno(const Slice& internal_key) {
|
|
|
|
const size_t n = internal_key.size();
|
|
|
|
assert(n >= 8);
|
|
|
|
uint64_t num = DecodeFixed64(internal_key.data() + n - 8);
|
|
|
|
return num >> 8;
|
|
|
|
}
|
|
|
|
|
2019-05-24 01:16:38 +02:00
|
|
|
// The class to store keys in an efficient way. It allows:
|
|
|
|
// 1. Users can either copy the key into it, or have it point to an unowned
|
|
|
|
// address.
|
|
|
|
// 2. For copied key, a short inline buffer is kept to reduce memory
|
|
|
|
// allocation for smaller keys.
|
|
|
|
// 3. It tracks user key or internal key, and allow conversion between them.
|
2014-04-08 01:56:26 +02:00
|
|
|
class IterKey {
|
|
|
|
public:
|
2015-12-16 21:08:30 +01:00
|
|
|
IterKey()
|
2017-04-04 23:17:16 +02:00
|
|
|
: buf_(space_),
|
|
|
|
key_(buf_),
|
|
|
|
key_size_(0),
|
2019-03-21 17:51:29 +01:00
|
|
|
buf_size_(sizeof(space_)),
|
2017-04-04 23:17:16 +02:00
|
|
|
is_user_key_(true) {}
|
2019-09-12 03:07:12 +02:00
|
|
|
// No copying allowed
|
|
|
|
IterKey(const IterKey&) = delete;
|
|
|
|
void operator=(const IterKey&) = delete;
|
2014-04-08 01:56:26 +02:00
|
|
|
|
2014-04-09 02:30:45 +02:00
|
|
|
~IterKey() { ResetBuffer(); }
|
2014-04-08 01:56:26 +02:00
|
|
|
|
2018-07-17 02:10:44 +02:00
|
|
|
// The bool will be picked up by the next calls to SetKey
|
|
|
|
void SetIsUserKey(bool is_user_key) { is_user_key_ = is_user_key; }
|
|
|
|
|
|
|
|
// Returns the key in whichever format that was provided to KeyIter
|
|
|
|
Slice GetKey() const { return Slice(key_, key_size_); }
|
|
|
|
|
2017-04-04 23:17:16 +02:00
|
|
|
Slice GetInternalKey() const {
|
|
|
|
assert(!IsUserKey());
|
|
|
|
return Slice(key_, key_size_);
|
|
|
|
}
|
2014-04-08 01:56:26 +02:00
|
|
|
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
Slice GetUserKey() const {
|
2017-04-04 23:17:16 +02:00
|
|
|
if (IsUserKey()) {
|
|
|
|
return Slice(key_, key_size_);
|
|
|
|
} else {
|
|
|
|
assert(key_size_ >= 8);
|
|
|
|
return Slice(key_, key_size_ - 8);
|
|
|
|
}
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
}
|
|
|
|
|
2015-09-18 20:10:00 +02:00
|
|
|
size_t Size() const { return key_size_; }
|
2014-07-23 21:31:11 +02:00
|
|
|
|
2014-04-09 02:30:45 +02:00
|
|
|
void Clear() { key_size_ = 0; }
|
2014-04-08 01:56:26 +02:00
|
|
|
|
2014-07-23 21:31:11 +02:00
|
|
|
// Append "non_shared_data" to its back, from "shared_len"
|
|
|
|
// This function is used in Block::Iter::ParseNextKey
|
|
|
|
// shared_len: bytes in [0, shard_len-1] would be remained
|
|
|
|
// non_shared_data: data to be append, its length must be >= non_shared_len
|
|
|
|
void TrimAppend(const size_t shared_len, const char* non_shared_data,
|
|
|
|
const size_t non_shared_len) {
|
|
|
|
assert(shared_len <= key_size_);
|
|
|
|
size_t total_size = shared_len + non_shared_len;
|
2015-12-16 21:08:30 +01:00
|
|
|
|
|
|
|
if (IsKeyPinned() /* key is not in buf_ */) {
|
|
|
|
// Copy the key from external memory to buf_ (copy shared_len bytes)
|
|
|
|
EnlargeBufferIfNeeded(total_size);
|
|
|
|
memcpy(buf_, key_, shared_len);
|
|
|
|
} else if (total_size > buf_size_) {
|
2014-07-23 21:31:11 +02:00
|
|
|
// Need to allocate space, delete previous space
|
|
|
|
char* p = new char[total_size];
|
|
|
|
memcpy(p, key_, shared_len);
|
|
|
|
|
2015-12-16 21:08:30 +01:00
|
|
|
if (buf_ != space_) {
|
|
|
|
delete[] buf_;
|
2014-07-23 21:31:11 +02:00
|
|
|
}
|
|
|
|
|
2015-12-16 21:08:30 +01:00
|
|
|
buf_ = p;
|
2014-07-23 21:31:11 +02:00
|
|
|
buf_size_ = total_size;
|
|
|
|
}
|
|
|
|
|
2015-12-16 21:08:30 +01:00
|
|
|
memcpy(buf_ + shared_len, non_shared_data, non_shared_len);
|
|
|
|
key_ = buf_;
|
|
|
|
key_size_ = total_size;
|
2014-07-23 21:31:11 +02:00
|
|
|
}
|
|
|
|
|
2018-07-17 02:10:44 +02:00
|
|
|
Slice SetKey(const Slice& key, bool copy = true) {
|
|
|
|
// is_user_key_ expected to be set already via SetIsUserKey
|
|
|
|
return SetKeyImpl(key, copy);
|
|
|
|
}
|
|
|
|
|
2017-04-04 23:17:16 +02:00
|
|
|
Slice SetUserKey(const Slice& key, bool copy = true) {
|
|
|
|
is_user_key_ = true;
|
|
|
|
return SetKeyImpl(key, copy);
|
|
|
|
}
|
|
|
|
|
|
|
|
Slice SetInternalKey(const Slice& key, bool copy = true) {
|
|
|
|
is_user_key_ = false;
|
|
|
|
return SetKeyImpl(key, copy);
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
// Copies the content of key, updates the reference to the user key in ikey
|
|
|
|
// and returns a Slice referencing the new copy.
|
2017-04-04 23:17:16 +02:00
|
|
|
Slice SetInternalKey(const Slice& key, ParsedInternalKey* ikey) {
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
size_t key_n = key.size();
|
|
|
|
assert(key_n >= 8);
|
2017-04-04 23:17:16 +02:00
|
|
|
SetInternalKey(key);
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
ikey->user_key = Slice(key_, key_n - 8);
|
|
|
|
return Slice(key_, key_n);
|
|
|
|
}
|
|
|
|
|
2016-10-19 01:59:37 +02:00
|
|
|
// Copy the key into IterKey own buf_
|
|
|
|
void OwnKey() {
|
|
|
|
assert(IsKeyPinned() == true);
|
|
|
|
|
|
|
|
Reserve(key_size_);
|
|
|
|
memcpy(buf_, key_, key_size_);
|
|
|
|
key_ = buf_;
|
|
|
|
}
|
|
|
|
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
// Update the sequence number in the internal key. Guarantees not to
|
|
|
|
// invalidate slices to the key (and the user key).
|
|
|
|
void UpdateInternalKey(uint64_t seq, ValueType t) {
|
2015-12-16 21:08:30 +01:00
|
|
|
assert(!IsKeyPinned());
|
Support for SingleDelete()
Summary:
This patch fixes #7460559. It introduces SingleDelete as a new database
operation. This operation can be used to delete keys that were never
overwritten (no put following another put of the same key). If an overwritten
key is single deleted the behavior is undefined. Single deletion of a
non-existent key has no effect but multiple consecutive single deletions are
not allowed (see limitations).
In contrast to the conventional Delete() operation, the deletion entry is
removed along with the value when the two are lined up in a compaction. Note:
The semantics are similar to @igor's prototype that allowed to have this
behavior on the granularity of a column family (
https://reviews.facebook.net/D42093 ). This new patch, however, is more
aggressive when it comes to removing tombstones: It removes the SingleDelete
together with the value whenever there is no snapshot between them while the
older patch only did this when the sequence number of the deletion was older
than the earliest snapshot.
Most of the complex additions are in the Compaction Iterator, all other changes
should be relatively straightforward. The patch also includes basic support for
single deletions in db_stress and db_bench.
Limitations:
- Not compatible with cuckoo hash tables
- Single deletions cannot be used in combination with merges and normal
deletions on the same key (other keys are not affected by this)
- Consecutive single deletions are currently not allowed (and older version of
this patch supported this so it could be resurrected if needed)
Test Plan: make all check
Reviewers: yhchiang, sdong, rven, anthony, yoshinorim, igor
Reviewed By: igor
Subscribers: maykov, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43179
2015-09-17 20:42:56 +02:00
|
|
|
assert(key_size_ >= 8);
|
|
|
|
uint64_t newval = (seq << 8) | t;
|
2015-12-16 21:08:30 +01:00
|
|
|
EncodeFixed64(&buf_[key_size_ - 8], newval);
|
2014-04-08 01:56:26 +02:00
|
|
|
}
|
|
|
|
|
2015-12-16 21:08:30 +01:00
|
|
|
bool IsKeyPinned() const { return (key_ != buf_); }
|
|
|
|
|
2014-06-19 01:36:48 +02:00
|
|
|
void SetInternalKey(const Slice& key_prefix, const Slice& user_key,
|
|
|
|
SequenceNumber s,
|
2014-04-08 01:56:26 +02:00
|
|
|
ValueType value_type = kValueTypeForSeek) {
|
2014-06-19 01:36:48 +02:00
|
|
|
size_t psize = key_prefix.size();
|
2014-04-08 01:56:26 +02:00
|
|
|
size_t usize = user_key.size();
|
2014-06-19 01:36:48 +02:00
|
|
|
EnlargeBufferIfNeeded(psize + usize + sizeof(uint64_t));
|
|
|
|
if (psize > 0) {
|
2015-12-16 21:08:30 +01:00
|
|
|
memcpy(buf_, key_prefix.data(), psize);
|
2014-06-19 01:36:48 +02:00
|
|
|
}
|
2015-12-16 21:08:30 +01:00
|
|
|
memcpy(buf_ + psize, user_key.data(), usize);
|
|
|
|
EncodeFixed64(buf_ + usize + psize, PackSequenceAndType(s, value_type));
|
|
|
|
|
|
|
|
key_ = buf_;
|
2014-06-19 01:36:48 +02:00
|
|
|
key_size_ = psize + usize + sizeof(uint64_t);
|
2017-04-04 23:17:16 +02:00
|
|
|
is_user_key_ = false;
|
2014-06-19 01:36:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
void SetInternalKey(const Slice& user_key, SequenceNumber s,
|
|
|
|
ValueType value_type = kValueTypeForSeek) {
|
|
|
|
SetInternalKey(Slice(), user_key, s, value_type);
|
|
|
|
}
|
|
|
|
|
|
|
|
void Reserve(size_t size) {
|
|
|
|
EnlargeBufferIfNeeded(size);
|
|
|
|
key_size_ = size;
|
2014-04-08 01:56:26 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
void SetInternalKey(const ParsedInternalKey& parsed_key) {
|
2014-06-19 01:36:48 +02:00
|
|
|
SetInternalKey(Slice(), parsed_key);
|
|
|
|
}
|
|
|
|
|
|
|
|
void SetInternalKey(const Slice& key_prefix,
|
|
|
|
const ParsedInternalKey& parsed_key_suffix) {
|
|
|
|
SetInternalKey(key_prefix, parsed_key_suffix.user_key,
|
|
|
|
parsed_key_suffix.sequence, parsed_key_suffix.type);
|
2014-04-08 01:56:26 +02:00
|
|
|
}
|
|
|
|
|
2014-07-01 20:05:05 +02:00
|
|
|
void EncodeLengthPrefixedKey(const Slice& key) {
|
|
|
|
auto size = key.size();
|
2014-11-11 22:47:22 +01:00
|
|
|
EnlargeBufferIfNeeded(size + static_cast<size_t>(VarintLength(size)));
|
2015-12-16 21:08:30 +01:00
|
|
|
char* ptr = EncodeVarint32(buf_, static_cast<uint32_t>(size));
|
2014-07-01 20:05:05 +02:00
|
|
|
memcpy(ptr, key.data(), size);
|
2015-12-16 21:08:30 +01:00
|
|
|
key_ = buf_;
|
2017-04-04 23:17:16 +02:00
|
|
|
is_user_key_ = true;
|
2014-07-01 20:05:05 +02:00
|
|
|
}
|
|
|
|
|
2017-04-04 23:17:16 +02:00
|
|
|
bool IsUserKey() const { return is_user_key_; }
|
|
|
|
|
2014-04-08 01:56:26 +02:00
|
|
|
private:
|
2015-12-16 21:08:30 +01:00
|
|
|
char* buf_;
|
|
|
|
const char* key_;
|
2014-04-08 01:56:26 +02:00
|
|
|
size_t key_size_;
|
2019-03-21 17:51:29 +01:00
|
|
|
size_t buf_size_;
|
2014-04-08 01:56:26 +02:00
|
|
|
char space_[32]; // Avoid allocation for short keys
|
2017-04-04 23:17:16 +02:00
|
|
|
bool is_user_key_;
|
|
|
|
|
|
|
|
Slice SetKeyImpl(const Slice& key, bool copy) {
|
|
|
|
size_t size = key.size();
|
|
|
|
if (copy) {
|
|
|
|
// Copy key to buf_
|
|
|
|
EnlargeBufferIfNeeded(size);
|
|
|
|
memcpy(buf_, key.data(), size);
|
|
|
|
key_ = buf_;
|
|
|
|
} else {
|
|
|
|
// Update key_ to point to external memory
|
|
|
|
key_ = key.data();
|
|
|
|
}
|
|
|
|
key_size_ = size;
|
|
|
|
return Slice(key_, key_size_);
|
|
|
|
}
|
2014-04-08 01:56:26 +02:00
|
|
|
|
2014-04-09 02:30:45 +02:00
|
|
|
void ResetBuffer() {
|
2015-12-16 21:08:30 +01:00
|
|
|
if (buf_ != space_) {
|
|
|
|
delete[] buf_;
|
|
|
|
buf_ = space_;
|
2014-04-09 02:30:45 +02:00
|
|
|
}
|
2014-04-24 02:51:16 +02:00
|
|
|
buf_size_ = sizeof(space_);
|
2014-04-09 02:30:45 +02:00
|
|
|
key_size_ = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Enlarge the buffer size if needed based on key_size.
|
|
|
|
// By default, static allocated buffer is used. Once there is a key
|
|
|
|
// larger than the static allocated buffer, another buffer is dynamically
|
|
|
|
// allocated, until a larger key buffer is requested. In that case, we
|
|
|
|
// reallocate buffer and delete the old one.
|
|
|
|
void EnlargeBufferIfNeeded(size_t key_size) {
|
|
|
|
// If size is smaller than buffer size, continue using current buffer,
|
|
|
|
// or the static allocated one, as default
|
|
|
|
if (key_size > buf_size_) {
|
2017-09-15 00:41:19 +02:00
|
|
|
EnlargeBuffer(key_size);
|
2014-04-09 02:30:45 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-09-15 00:41:19 +02:00
|
|
|
void EnlargeBuffer(size_t key_size);
|
2014-04-08 01:56:26 +02:00
|
|
|
};
|
|
|
|
|
2019-05-24 01:16:38 +02:00
|
|
|
// Convert from a SliceTranform of user keys, to a SliceTransform of
|
|
|
|
// user keys.
|
2014-04-10 23:19:43 +02:00
|
|
|
class InternalKeySliceTransform : public SliceTransform {
|
|
|
|
public:
|
|
|
|
explicit InternalKeySliceTransform(const SliceTransform* transform)
|
|
|
|
: transform_(transform) {}
|
|
|
|
|
2015-02-26 20:28:41 +01:00
|
|
|
virtual const char* Name() const override { return transform_->Name(); }
|
2014-04-10 23:19:43 +02:00
|
|
|
|
2015-02-26 20:28:41 +01:00
|
|
|
virtual Slice Transform(const Slice& src) const override {
|
2014-04-10 23:19:43 +02:00
|
|
|
auto user_key = ExtractUserKey(src);
|
|
|
|
return transform_->Transform(user_key);
|
|
|
|
}
|
|
|
|
|
2015-02-26 20:28:41 +01:00
|
|
|
virtual bool InDomain(const Slice& src) const override {
|
2014-04-10 23:19:43 +02:00
|
|
|
auto user_key = ExtractUserKey(src);
|
|
|
|
return transform_->InDomain(user_key);
|
|
|
|
}
|
|
|
|
|
2015-02-26 20:28:41 +01:00
|
|
|
virtual bool InRange(const Slice& dst) const override {
|
2014-04-10 23:19:43 +02:00
|
|
|
auto user_key = ExtractUserKey(dst);
|
|
|
|
return transform_->InRange(user_key);
|
|
|
|
}
|
|
|
|
|
|
|
|
const SliceTransform* user_prefix_extractor() const { return transform_; }
|
|
|
|
|
|
|
|
private:
|
|
|
|
// Like comparator, InternalKeySliceTransform will not take care of the
|
|
|
|
// deletion of transform_
|
|
|
|
const SliceTransform* const transform_;
|
|
|
|
};
|
|
|
|
|
2016-04-02 00:23:46 +02:00
|
|
|
// Read the key of a record from a write batch.
|
|
|
|
// if this record represent the default column family then cf_record
|
|
|
|
// must be passed as false, otherwise it must be passed as true.
|
|
|
|
extern bool ReadKeyFromWriteBatchEntry(Slice* input, Slice* key,
|
|
|
|
bool cf_record);
|
|
|
|
|
2014-08-19 00:19:17 +02:00
|
|
|
// Read record from a write batch piece from input.
|
|
|
|
// tag, column_family, key, value and blob are return values. Callers own the
|
|
|
|
// Slice they point to.
|
|
|
|
// Tag is defined as ValueType.
|
|
|
|
// input will be advanced to after the record.
|
|
|
|
extern Status ReadRecordFromWriteBatch(Slice* input, char* tag,
|
|
|
|
uint32_t* column_family, Slice* key,
|
Modification of WriteBatch to support two phase commit
Summary: Adds three new WriteBatch data types: Prepare(xid), Commit(xid), Rollback(xid). Prepare(xid) should precede the (single) operation to which is applies. There can obviously be multiple Prepare(xid) markers. There should only be one Rollback(xid) or Commit(xid) marker yet not both. None of this logic is currently enforced and will most likely be implemented further up such as in the memtableinserter. All three markers are similar to PutLogData in that they are writebatch meta-data, ie stored but not counted. All three markers differ from PutLogData in that they will actually be written to disk. As for WriteBatchWithIndex, Prepare, Commit, Rollback are all implemented just as PutLogData and none are tested just as PutLogData.
Test Plan: single unit test in write_batch_test.
Reviewers: hermanlee4, sdong, anthony
Subscribers: leveldb, dhruba, vasilep, andrewkr
Differential Revision: https://reviews.facebook.net/D57867
2016-04-08 08:35:51 +02:00
|
|
|
Slice* value, Slice* blob, Slice* xid);
|
2016-08-20 00:10:31 +02:00
|
|
|
|
|
|
|
// When user call DeleteRange() to delete a range of keys,
|
|
|
|
// we will store a serialized RangeTombstone in MemTable and SST.
|
|
|
|
// the struct here is a easy-understood form
|
|
|
|
// start/end_key_ is the start/end user key of the range to be deleted
|
|
|
|
struct RangeTombstone {
|
|
|
|
Slice start_key_;
|
|
|
|
Slice end_key_;
|
|
|
|
SequenceNumber seq_;
|
2016-12-20 01:44:30 +01:00
|
|
|
RangeTombstone() = default;
|
|
|
|
RangeTombstone(Slice sk, Slice ek, SequenceNumber sn)
|
2016-08-20 00:10:31 +02:00
|
|
|
: start_key_(sk), end_key_(ek), seq_(sn) {}
|
|
|
|
|
2016-12-20 01:44:30 +01:00
|
|
|
RangeTombstone(ParsedInternalKey parsed_key, Slice value) {
|
Compaction Support for Range Deletion
Summary:
This diff introduces RangeDelAggregator, which takes ownership of iterators
provided to it via AddTombstones(). The tombstones are organized in a two-level
map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data
copy by holding Slices returned by the iterator, which remain valid thanks to pinning.
For compaction, we create a hierarchical range tombstone iterator with structure
matching the iterator over compaction input data. An aggregator based on that
iterator is used by CompactionIterator to determine which keys are covered by
range tombstones. In case of merge operand, the same aggregator is used by
MergeHelper. Upon finishing each file in the compaction, relevant range tombstones
are added to the output file's range tombstone metablock and file boundaries are
updated accordingly.
To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete()
considers tombstones in the key's snapshot stripe. When this function is used outside of
compaction, it also checks newer stripes, which can contain covering tombstones. Currently
the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges
within a stripe such that binary search can be used.
RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range
to a new table's range tombstone meta-block. Since range tombstones may fall in the gap
between files, we may need to extend some files' key-ranges. The strategy is (1) first file
extends as far left as possible and other files do not extend left, (2) all files extend right
until either the start of the next file or the end of the last range tombstone in the gap,
whichever comes first.
One other notable change is adding release/move semantics to ScopedArenaIterator
such that it can be used to transfer ownership of an arena-allocated iterator, similar to
how unique_ptr is used for malloc'd data.
Depends on D61473
Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927
Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark
Reviewed By: lightmark
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
|
|
|
start_key_ = parsed_key.user_key;
|
|
|
|
seq_ = parsed_key.sequence;
|
|
|
|
end_key_ = value;
|
2016-08-20 00:10:31 +02:00
|
|
|
}
|
|
|
|
|
Compaction Support for Range Deletion
Summary:
This diff introduces RangeDelAggregator, which takes ownership of iterators
provided to it via AddTombstones(). The tombstones are organized in a two-level
map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data
copy by holding Slices returned by the iterator, which remain valid thanks to pinning.
For compaction, we create a hierarchical range tombstone iterator with structure
matching the iterator over compaction input data. An aggregator based on that
iterator is used by CompactionIterator to determine which keys are covered by
range tombstones. In case of merge operand, the same aggregator is used by
MergeHelper. Upon finishing each file in the compaction, relevant range tombstones
are added to the output file's range tombstone metablock and file boundaries are
updated accordingly.
To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete()
considers tombstones in the key's snapshot stripe. When this function is used outside of
compaction, it also checks newer stripes, which can contain covering tombstones. Currently
the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges
within a stripe such that binary search can be used.
RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range
to a new table's range tombstone meta-block. Since range tombstones may fall in the gap
between files, we may need to extend some files' key-ranges. The strategy is (1) first file
extends as far left as possible and other files do not extend left, (2) all files extend right
until either the start of the next file or the end of the last range tombstone in the gap,
whichever comes first.
One other notable change is adding release/move semantics to ScopedArenaIterator
such that it can be used to transfer ownership of an arena-allocated iterator, similar to
how unique_ptr is used for malloc'd data.
Depends on D61473
Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927
Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark
Reviewed By: lightmark
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
|
|
|
// be careful to use Serialize(), allocates new memory
|
|
|
|
std::pair<InternalKey, Slice> Serialize() const {
|
2016-08-20 00:10:31 +02:00
|
|
|
auto key = InternalKey(start_key_, seq_, kTypeRangeDeletion);
|
|
|
|
Slice value = end_key_;
|
|
|
|
return std::make_pair(std::move(key), std::move(value));
|
|
|
|
}
|
|
|
|
|
Compaction Support for Range Deletion
Summary:
This diff introduces RangeDelAggregator, which takes ownership of iterators
provided to it via AddTombstones(). The tombstones are organized in a two-level
map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data
copy by holding Slices returned by the iterator, which remain valid thanks to pinning.
For compaction, we create a hierarchical range tombstone iterator with structure
matching the iterator over compaction input data. An aggregator based on that
iterator is used by CompactionIterator to determine which keys are covered by
range tombstones. In case of merge operand, the same aggregator is used by
MergeHelper. Upon finishing each file in the compaction, relevant range tombstones
are added to the output file's range tombstone metablock and file boundaries are
updated accordingly.
To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete()
considers tombstones in the key's snapshot stripe. When this function is used outside of
compaction, it also checks newer stripes, which can contain covering tombstones. Currently
the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges
within a stripe such that binary search can be used.
RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range
to a new table's range tombstone meta-block. Since range tombstones may fall in the gap
between files, we may need to extend some files' key-ranges. The strategy is (1) first file
extends as far left as possible and other files do not extend left, (2) all files extend right
until either the start of the next file or the end of the last range tombstone in the gap,
whichever comes first.
One other notable change is adding release/move semantics to ScopedArenaIterator
such that it can be used to transfer ownership of an arena-allocated iterator, similar to
how unique_ptr is used for malloc'd data.
Depends on D61473
Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927
Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark
Reviewed By: lightmark
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
|
|
|
// be careful to use SerializeKey(), allocates new memory
|
|
|
|
InternalKey SerializeKey() const {
|
2016-08-20 00:10:31 +02:00
|
|
|
return InternalKey(start_key_, seq_, kTypeRangeDeletion);
|
|
|
|
}
|
Compaction Support for Range Deletion
Summary:
This diff introduces RangeDelAggregator, which takes ownership of iterators
provided to it via AddTombstones(). The tombstones are organized in a two-level
map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data
copy by holding Slices returned by the iterator, which remain valid thanks to pinning.
For compaction, we create a hierarchical range tombstone iterator with structure
matching the iterator over compaction input data. An aggregator based on that
iterator is used by CompactionIterator to determine which keys are covered by
range tombstones. In case of merge operand, the same aggregator is used by
MergeHelper. Upon finishing each file in the compaction, relevant range tombstones
are added to the output file's range tombstone metablock and file boundaries are
updated accordingly.
To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete()
considers tombstones in the key's snapshot stripe. When this function is used outside of
compaction, it also checks newer stripes, which can contain covering tombstones. Currently
the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges
within a stripe such that binary search can be used.
RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range
to a new table's range tombstone meta-block. Since range tombstones may fall in the gap
between files, we may need to extend some files' key-ranges. The strategy is (1) first file
extends as far left as possible and other files do not extend left, (2) all files extend right
until either the start of the next file or the end of the last range tombstone in the gap,
whichever comes first.
One other notable change is adding release/move semantics to ScopedArenaIterator
such that it can be used to transfer ownership of an arena-allocated iterator, similar to
how unique_ptr is used for malloc'd data.
Depends on D61473
Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927
Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark
Reviewed By: lightmark
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
|
|
|
|
2018-07-14 02:34:54 +02:00
|
|
|
// The tombstone end-key is exclusive, so we generate an internal-key here
|
|
|
|
// which has a similar property. Using kMaxSequenceNumber guarantees that
|
|
|
|
// the returned internal-key will compare less than any other internal-key
|
|
|
|
// with the same user-key. This in turn guarantees that the serialized
|
|
|
|
// end-key for a tombstone such as [a-b] will compare less than the key "b".
|
|
|
|
//
|
Compaction Support for Range Deletion
Summary:
This diff introduces RangeDelAggregator, which takes ownership of iterators
provided to it via AddTombstones(). The tombstones are organized in a two-level
map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data
copy by holding Slices returned by the iterator, which remain valid thanks to pinning.
For compaction, we create a hierarchical range tombstone iterator with structure
matching the iterator over compaction input data. An aggregator based on that
iterator is used by CompactionIterator to determine which keys are covered by
range tombstones. In case of merge operand, the same aggregator is used by
MergeHelper. Upon finishing each file in the compaction, relevant range tombstones
are added to the output file's range tombstone metablock and file boundaries are
updated accordingly.
To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete()
considers tombstones in the key's snapshot stripe. When this function is used outside of
compaction, it also checks newer stripes, which can contain covering tombstones. Currently
the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges
within a stripe such that binary search can be used.
RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range
to a new table's range tombstone meta-block. Since range tombstones may fall in the gap
between files, we may need to extend some files' key-ranges. The strategy is (1) first file
extends as far left as possible and other files do not extend left, (2) all files extend right
until either the start of the next file or the end of the last range tombstone in the gap,
whichever comes first.
One other notable change is adding release/move semantics to ScopedArenaIterator
such that it can be used to transfer ownership of an arena-allocated iterator, similar to
how unique_ptr is used for malloc'd data.
Depends on D61473
Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927
Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark
Reviewed By: lightmark
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
|
|
|
// be careful to use SerializeEndKey(), allocates new memory
|
|
|
|
InternalKey SerializeEndKey() const {
|
2018-07-14 02:34:54 +02:00
|
|
|
return InternalKey(end_key_, kMaxSequenceNumber, kTypeRangeDeletion);
|
Compaction Support for Range Deletion
Summary:
This diff introduces RangeDelAggregator, which takes ownership of iterators
provided to it via AddTombstones(). The tombstones are organized in a two-level
map (snapshot stripe -> begin key -> tombstone). Tombstone creation avoids data
copy by holding Slices returned by the iterator, which remain valid thanks to pinning.
For compaction, we create a hierarchical range tombstone iterator with structure
matching the iterator over compaction input data. An aggregator based on that
iterator is used by CompactionIterator to determine which keys are covered by
range tombstones. In case of merge operand, the same aggregator is used by
MergeHelper. Upon finishing each file in the compaction, relevant range tombstones
are added to the output file's range tombstone metablock and file boundaries are
updated accordingly.
To check whether a key is covered by range tombstone, RangeDelAggregator::ShouldDelete()
considers tombstones in the key's snapshot stripe. When this function is used outside of
compaction, it also checks newer stripes, which can contain covering tombstones. Currently
the intra-stripe check involves a linear scan; however, in the future we plan to collapse ranges
within a stripe such that binary search can be used.
RangeDelAggregator::AddToBuilder() adds all range tombstones in the table's key-range
to a new table's range tombstone meta-block. Since range tombstones may fall in the gap
between files, we may need to extend some files' key-ranges. The strategy is (1) first file
extends as far left as possible and other files do not extend left, (2) all files extend right
until either the start of the next file or the end of the last range tombstone in the gap,
whichever comes first.
One other notable change is adding release/move semantics to ScopedArenaIterator
such that it can be used to transfer ownership of an arena-allocated iterator, similar to
how unique_ptr is used for malloc'd data.
Depends on D61473
Test Plan: compaction_iterator_test, mock_table, end-to-end tests in D63927
Reviewers: sdong, IslamAbdelRahman, wanning, yhchiang, lightmark
Reviewed By: lightmark
Subscribers: andrewkr, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D62205
2016-10-18 21:04:56 +02:00
|
|
|
}
|
2016-08-20 00:10:31 +02:00
|
|
|
};
|
|
|
|
|
2019-03-28 00:13:08 +01:00
|
|
|
inline int InternalKeyComparator::Compare(const Slice& akey,
|
|
|
|
const Slice& bkey) const {
|
2018-03-23 21:16:37 +01:00
|
|
|
// Order by:
|
|
|
|
// increasing user key (according to user-supplied comparator)
|
|
|
|
// decreasing sequence number
|
|
|
|
// decreasing type (though sequence# should be enough to disambiguate)
|
2019-03-27 18:24:16 +01:00
|
|
|
int r = user_comparator_.Compare(ExtractUserKey(akey), ExtractUserKey(bkey));
|
2018-03-23 21:16:37 +01:00
|
|
|
if (r == 0) {
|
|
|
|
const uint64_t anum = DecodeFixed64(akey.data() + akey.size() - 8);
|
|
|
|
const uint64_t bnum = DecodeFixed64(bkey.data() + bkey.size() - 8);
|
|
|
|
if (anum > bnum) {
|
|
|
|
r = -1;
|
|
|
|
} else if (anum < bnum) {
|
|
|
|
r = +1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2019-03-28 00:13:08 +01:00
|
|
|
inline int InternalKeyComparator::CompareKeySeq(const Slice& akey,
|
|
|
|
const Slice& bkey) const {
|
2018-03-23 21:16:37 +01:00
|
|
|
// Order by:
|
|
|
|
// increasing user key (according to user-supplied comparator)
|
|
|
|
// decreasing sequence number
|
2019-03-27 18:24:16 +01:00
|
|
|
int r = user_comparator_.Compare(ExtractUserKey(akey), ExtractUserKey(bkey));
|
2018-03-23 21:16:37 +01:00
|
|
|
if (r == 0) {
|
|
|
|
// Shift the number to exclude the last byte which contains the value type
|
|
|
|
const uint64_t anum = DecodeFixed64(akey.data() + akey.size() - 8) >> 8;
|
|
|
|
const uint64_t bnum = DecodeFixed64(bkey.data() + bkey.size() - 8) >> 8;
|
|
|
|
if (anum > bnum) {
|
|
|
|
r = -1;
|
|
|
|
} else if (anum < bnum) {
|
|
|
|
r = +1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2019-05-24 01:16:38 +02:00
|
|
|
// Wrap InternalKeyComparator as a comparator class for ParsedInternalKey.
|
2018-10-25 05:13:06 +02:00
|
|
|
struct ParsedInternalKeyComparator {
|
|
|
|
explicit ParsedInternalKeyComparator(const InternalKeyComparator* c)
|
|
|
|
: cmp(c) {}
|
|
|
|
|
|
|
|
bool operator()(const ParsedInternalKey& a,
|
|
|
|
const ParsedInternalKey& b) const {
|
|
|
|
return cmp->Compare(a, b) < 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
const InternalKeyComparator* cmp;
|
|
|
|
};
|
|
|
|
|
2013-10-04 06:49:15 +02:00
|
|
|
} // namespace rocksdb
|