rocksdb/db/merge_helper.cc

306 lines
12 KiB
C++
Raw Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree. An additional grant
// of patent rights can be found in the PATENTS file in the same directory.
#include "db/merge_helper.h"
#include <stdio.h>
#include <string>
#include "db/dbformat.h"
#include "rocksdb/comparator.h"
#include "rocksdb/db.h"
#include "rocksdb/merge_operator.h"
#include "table/internal_iterator.h"
#include "util/perf_context_imp.h"
#include "util/statistics.h"
namespace rocksdb {
Status MergeHelper::TimedFullMerge(const MergeOperator* merge_operator,
const Slice& key, const Slice* value,
const std::deque<std::string>& operands,
std::string* result, Logger* logger,
Statistics* statistics, Env* env) {
assert(merge_operator != nullptr);
if (operands.size() == 0) {
result->assign(value->data(), value->size());
return Status::OK();
}
bool success;
{
// Setup to time the merge
StopWatchNano timer(env, statistics != nullptr);
PERF_TIMER_GUARD(merge_operator_time_nanos);
// Do the merge
success = merge_operator->FullMerge(key, value, operands, result, logger);
Remove wasteful instrumentation in FullMerge (stacked on D59577) Summary: [ This diff is stacked on top of D59577 ] We keep calling timer.ElapsedNanos() on every call to MergeOperator::FullMerge even when statistics are disabled, this is wasteful. I run the readseq benchmark on a DB containing 100K merge operands for 100K keys (1 operand per key) with 1GB block cache I see slight performance improvment Original results ``` $ ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=100000 --num=100000 --db="/dev/shm/100K_merge_compacted/" --cache_size=1073741824 --use_existing_db --disable_auto_compactions ------------------------------------------------ DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.498 micros/op 2006597 ops/sec; 222.0 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.295 micros/op 3393627 ops/sec; 375.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.285 micros/op 3511155 ops/sec; 388.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.286 micros/op 3500470 ops/sec; 387.2 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.283 micros/op 3530751 ops/sec; 390.6 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.289 micros/op 3464811 ops/sec; 383.3 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.277 micros/op 3612814 ops/sec; 399.7 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.283 micros/op 3539640 ops/sec; 391.6 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.285 micros/op 3503766 ops/sec; 387.6 MB/s ``` After patch ``` $ ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=100000 --num=100000 --db="/dev/shm/100K_merge_compacted/" --cache_size=1073741824 --use_existing_db --disable_auto_compactions ------------------------------------------------ DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.476 micros/op 2100119 ops/sec; 232.3 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.278 micros/op 3600887 ops/sec; 398.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.275 micros/op 3636698 ops/sec; 402.3 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.271 micros/op 3691661 ops/sec; 408.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.273 micros/op 3661534 ops/sec; 405.1 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.276 micros/op 3627106 ops/sec; 401.3 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.272 micros/op 3682635 ops/sec; 407.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.266 micros/op 3758331 ops/sec; 415.8 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.266 micros/op 3761907 ops/sec; 416.2 MB/s ``` Test Plan: make check -j64 Reviewers: yhchiang, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D59583
2016-06-14 01:22:14 +02:00
RecordTick(statistics, MERGE_OPERATION_TOTAL_TIME,
statistics ? timer.ElapsedNanos() : 0);
}
if (!success) {
RecordTick(statistics, NUMBER_MERGE_FAILURES);
return Status::Corruption("Error: Could not perform merge.");
}
return Status::OK();
}
// PRE: iter points to the first merge type entry
// POST: iter points to the first entry beyond the merge process (or the end)
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// keys_, operands_ are updated to reflect the merge result.
// keys_ stores the list of keys encountered while merging.
// operands_ stores the list of merge operands encountered while merging.
// keys_[i] corresponds to operands_[i] for each i.
Status MergeHelper::MergeUntil(InternalIterator* iter,
const SequenceNumber stop_before,
const bool at_bottom) {
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// Get a copy of the internal key, before it's invalidated by iter->Next()
// Also maintain the list of merge operands seen.
assert(HasOperator());
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
keys_.clear();
operands_.clear();
assert(user_merge_operator_);
bool first_key = true;
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// We need to parse the internal key again as the parsed key is
// backed by the internal key!
// Assume no internal key corruption as it has been successfully parsed
// by the caller.
// original_key_is_iter variable is just caching the information:
// original_key_is_iter == (iter->key().ToString() == original_key)
bool original_key_is_iter = true;
std::string original_key = iter->key().ToString();
// Important:
// orig_ikey is backed by original_key if keys_.empty()
// orig_ikey is backed by keys_.back() if !keys_.empty()
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
ParsedInternalKey orig_ikey;
ParseInternalKey(original_key, &orig_ikey);
Status s;
bool hit_the_next_user_key = false;
for (; iter->Valid(); iter->Next(), original_key_is_iter = false) {
ParsedInternalKey ikey;
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
assert(keys_.size() == operands_.size());
if (!ParseInternalKey(iter->key(), &ikey)) {
// stop at corrupted key
if (assert_valid_internal_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
assert(!"Corrupted internal key not expected.");
return Status::Corruption("Corrupted internal key not expected.");
}
break;
} else if (first_key) {
assert(user_comparator_->Equal(ikey.user_key, orig_ikey.user_key));
first_key = false;
} else if (!user_comparator_->Equal(ikey.user_key, orig_ikey.user_key)) {
// hit a different user key, stop right here
hit_the_next_user_key = true;
break;
} else if (stop_before && ikey.sequence <= stop_before) {
// hit an entry that's visible by the previous snapshot, can't touch that
break;
}
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// At this point we are guaranteed that we need to process this 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
assert(IsValueType(ikey.type));
if (ikey.type != kTypeMerge) {
if (ikey.type != kTypeValue && ikey.type != kTypeDeletion) {
// Merges operands can only be used with puts and deletions, single
// deletions are not supported.
assert(false);
// release build doesn't have asserts, so we return error status
return Status::InvalidArgument(
" Merges operands can only be used with puts and deletions, single "
"deletions are not supported.");
}
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
// hit a put/delete
// => merge the put value or a nullptr with operands_
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// => store result in operands_.back() (and update keys_.back())
// => change the entry type to kTypeValue for keys_.back()
// We are done! Success!
// If there are no operands, just return the Status::OK(). That will cause
// the compaction iterator to write out the key we're currently at, which
// is the put/delete we just encountered.
if (keys_.empty()) {
return Status::OK();
}
// TODO(noetzli) If the merge operator returns false, we are currently
// (almost) silently dropping the put/delete. That's probably not what we
// want.
const Slice val = iter->value();
const Slice* val_ptr = (kTypeValue == ikey.type) ? &val : nullptr;
std::string merge_result;
s = TimedFullMerge(user_merge_operator_, ikey.user_key, val_ptr,
operands_, &merge_result, logger_, stats_, env_);
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// We store the result in keys_.back() and operands_.back()
// if nothing went wrong (i.e.: no operand corruption on disk)
if (s.ok()) {
// The original key encountered
original_key = std::move(keys_.back());
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
orig_ikey.type = kTypeValue;
UpdateInternalKey(&original_key, orig_ikey.sequence, orig_ikey.type);
keys_.clear();
operands_.clear();
keys_.emplace_front(std::move(original_key));
operands_.emplace_front(std::move(merge_result));
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
}
// move iter to the next entry
iter->Next();
return s;
} else {
// hit a merge
// => if there is a compaction filter, apply it.
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// => merge the operand into the front of the operands_ list
// if not filtered
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// => then continue because we haven't yet seen a Put/Delete.
//
// Keep queuing keys and operands until we either meet a put / delete
// request or later did a partial merge.
Slice value_slice = iter->value();
// add an operand to the list if:
// 1) it's included in one of the snapshots. in that case we *must* write
// it out, no matter what compaction filter says
// 2) it's not filtered by a compaction filter
if (ikey.sequence <= latest_snapshot_ ||
!FilterMerge(orig_ikey.user_key, value_slice)) {
if (original_key_is_iter) {
// this is just an optimization that saves us one memcpy
keys_.push_front(std::move(original_key));
} else {
keys_.push_front(iter->key().ToString());
}
if (keys_.size() == 1) {
// we need to re-anchor the orig_ikey because it was anchored by
// original_key before
ParseInternalKey(keys_.back(), &orig_ikey);
}
operands_.push_front(value_slice.ToString());
}
}
}
if (operands_.size() == 0) {
// we filtered out all the merge operands
return Status::OK();
}
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// We are sure we have seen this key's entire history if we are at the
// last level and exhausted all internal keys of this user key.
// NOTE: !iter->Valid() does not necessarily mean we hit the
// beginning of a user key, as versions of a user key might be
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
// split into multiple files (even files on the same level)
// and some files might not be included in the compaction/merge.
//
// There are also cases where we have seen the root of history of this
// key without being sure of it. Then, we simply miss the opportunity
// to combine the keys. Since VersionSet::SetupOtherInputs() always makes
// sure that all merge-operands on the same level get compacted together,
// this will simply lead to these merge operands moving to the next level.
//
// So, we only perform the following logic (to merge all operands together
// without a Put/Delete) if we are certain that we have seen the end of key.
bool surely_seen_the_beginning = hit_the_next_user_key && at_bottom;
if (surely_seen_the_beginning) {
// do a final merge with nullptr as the existing value and say
// bye to the merge type (it's now converted to a Put)
assert(kTypeMerge == orig_ikey.type);
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
assert(operands_.size() >= 1);
assert(operands_.size() == keys_.size());
std::string merge_result;
s = TimedFullMerge(user_merge_operator_, orig_ikey.user_key, nullptr,
operands_, &merge_result, logger_, stats_, env_);
if (s.ok()) {
// The original key encountered
// We are certain that keys_ is not empty here (see assertions couple of
// lines before).
original_key = std::move(keys_.back());
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
orig_ikey.type = kTypeValue;
UpdateInternalKey(&original_key, orig_ikey.sequence, orig_ikey.type);
keys_.clear();
operands_.clear();
keys_.emplace_front(std::move(original_key));
operands_.emplace_front(std::move(merge_result));
[RocksDB] [MergeOperator] The new Merge Interface! Uses merge sequences. Summary: Here are the major changes to the Merge Interface. It has been expanded to handle cases where the MergeOperator is not associative. It does so by stacking up merge operations while scanning through the key history (i.e.: during Get() or Compaction), until a valid Put/Delete/end-of-history is encountered; it then applies all of the merge operations in the correct sequence starting with the base/sentinel value. I have also introduced an "AssociativeMerge" function which allows the user to take advantage of associative merge operations (such as in the case of counters). The implementation will always attempt to merge the operations/operands themselves together when they are encountered, and will resort to the "stacking" method if and only if the "associative-merge" fails. This implementation is conjectured to allow MergeOperator to handle the general case, while still providing the user with the ability to take advantage of certain efficiencies in their own merge-operator / data-structure. NOTE: This is a preliminary diff. This must still go through a lot of review, revision, and testing. Feedback welcome! Test Plan: -This is a preliminary diff. I have only just begun testing/debugging it. -I will be testing this with the existing MergeOperator use-cases and unit-tests (counters, string-append, and redis-lists) -I will be "desk-checking" and walking through the code with the help gdb. -I will find a way of stress-testing the new interface / implementation using db_bench, db_test, merge_test, and/or db_stress. -I will ensure that my tests cover all cases: Get-Memtable, Get-Immutable-Memtable, Get-from-Disk, Iterator-Range-Scan, Flush-Memtable-to-L0, Compaction-L0-L1, Compaction-Ln-L(n+1), Put/Delete found, Put/Delete not-found, end-of-history, end-of-file, etc. -A lot of feedback from the reviewers. Reviewers: haobo, dhruba, zshao, emayanke Reviewed By: haobo CC: leveldb Differential Revision: https://reviews.facebook.net/D11499
2013-08-06 05:14:32 +02:00
}
} else {
// We haven't seen the beginning of the key nor a Put/Delete.
// Attempt to use the user's associative merge function to
// merge the stacked merge operands into a single operand.
//
// TODO(noetzli) The docblock of MergeUntil suggests that a successful
// partial merge returns Status::OK(). Should we change the status code
// after a successful partial merge?
s = Status::MergeInProgress();
if (operands_.size() >= 2 &&
operands_.size() >= min_partial_merge_operands_) {
bool merge_success = false;
std::string merge_result;
{
StopWatchNano timer(env_, stats_ != nullptr);
PERF_TIMER_GUARD(merge_operator_time_nanos);
merge_success = user_merge_operator_->PartialMergeMulti(
orig_ikey.user_key,
std::deque<Slice>(operands_.begin(), operands_.end()),
&merge_result, logger_);
RecordTick(stats_, MERGE_OPERATION_TOTAL_TIME,
Remove wasteful instrumentation in FullMerge (stacked on D59577) Summary: [ This diff is stacked on top of D59577 ] We keep calling timer.ElapsedNanos() on every call to MergeOperator::FullMerge even when statistics are disabled, this is wasteful. I run the readseq benchmark on a DB containing 100K merge operands for 100K keys (1 operand per key) with 1GB block cache I see slight performance improvment Original results ``` $ ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=100000 --num=100000 --db="/dev/shm/100K_merge_compacted/" --cache_size=1073741824 --use_existing_db --disable_auto_compactions ------------------------------------------------ DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.498 micros/op 2006597 ops/sec; 222.0 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.295 micros/op 3393627 ops/sec; 375.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.285 micros/op 3511155 ops/sec; 388.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.286 micros/op 3500470 ops/sec; 387.2 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.283 micros/op 3530751 ops/sec; 390.6 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.289 micros/op 3464811 ops/sec; 383.3 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.277 micros/op 3612814 ops/sec; 399.7 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.283 micros/op 3539640 ops/sec; 391.6 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.285 micros/op 3503766 ops/sec; 387.6 MB/s ``` After patch ``` $ ./db_bench --benchmarks="readseq,readseq,readseq,readseq,readseq,readseq,readseq,readseq,readseq" --merge_operator="max" --merge_keys=100000 --num=100000 --db="/dev/shm/100K_merge_compacted/" --cache_size=1073741824 --use_existing_db --disable_auto_compactions ------------------------------------------------ DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.476 micros/op 2100119 ops/sec; 232.3 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.278 micros/op 3600887 ops/sec; 398.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.275 micros/op 3636698 ops/sec; 402.3 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.271 micros/op 3691661 ops/sec; 408.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.273 micros/op 3661534 ops/sec; 405.1 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.276 micros/op 3627106 ops/sec; 401.3 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.272 micros/op 3682635 ops/sec; 407.4 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.266 micros/op 3758331 ops/sec; 415.8 MB/s DB path: [/dev/shm/100K_merge_compacted/] readseq : 0.266 micros/op 3761907 ops/sec; 416.2 MB/s ``` Test Plan: make check -j64 Reviewers: yhchiang, sdong Reviewed By: sdong Subscribers: andrewkr, dhruba Differential Revision: https://reviews.facebook.net/D59583
2016-06-14 01:22:14 +02:00
stats_ ? timer.ElapsedNanosSafe() : 0);
}
if (merge_success) {
// Merging of operands (associative merge) was successful.
// Replace operands with the merge result
operands_.clear();
operands_.emplace_front(std::move(merge_result));
keys_.erase(keys_.begin(), keys_.end() - 1);
}
}
}
return s;
}
MergeOutputIterator::MergeOutputIterator(const MergeHelper* merge_helper)
: merge_helper_(merge_helper) {
it_keys_ = merge_helper_->keys().rend();
it_values_ = merge_helper_->values().rend();
}
void MergeOutputIterator::SeekToFirst() {
const auto& keys = merge_helper_->keys();
const auto& values = merge_helper_->values();
assert(keys.size() == values.size());
it_keys_ = keys.rbegin();
it_values_ = values.rbegin();
}
void MergeOutputIterator::Next() {
++it_keys_;
++it_values_;
}
bool MergeHelper::FilterMerge(const Slice& user_key, const Slice& value_slice) {
if (compaction_filter_ == nullptr) {
return false;
}
if (stats_ != nullptr) {
filter_timer_.Start();
}
bool to_delete =
compaction_filter_->FilterMergeOperand(level_, user_key, value_slice);
total_filter_time_ += filter_timer_.ElapsedNanosSafe();
return to_delete;
}
} // namespace rocksdb