rocksdb/db/compaction/compaction_outputs.cc

793 lines
32 KiB
C++
Raw Normal View History

// Copyright (c) Meta Platforms, Inc. and affiliates.
//
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "db/compaction/compaction_outputs.h"
#include "db/builder.h"
namespace ROCKSDB_NAMESPACE {
void CompactionOutputs::NewBuilder(const TableBuilderOptions& tboptions) {
builder_.reset(NewTableBuilder(tboptions, file_writer_.get()));
}
Status CompactionOutputs::Finish(const Status& intput_status,
const SeqnoToTimeMapping& seqno_time_mapping) {
FileMetaData* meta = GetMetaData();
assert(meta != nullptr);
Status s = intput_status;
if (s.ok()) {
std::string seqno_time_mapping_str;
seqno_time_mapping.Encode(seqno_time_mapping_str, meta->fd.smallest_seqno,
meta->fd.largest_seqno, meta->file_creation_time);
builder_->SetSeqnoTimeTableProperties(seqno_time_mapping_str,
meta->oldest_ancester_time);
s = builder_->Finish();
} else {
builder_->Abandon();
}
Status io_s = builder_->io_status();
if (s.ok()) {
s = io_s;
} else {
io_s.PermitUncheckedError();
}
const uint64_t current_bytes = builder_->FileSize();
if (s.ok()) {
meta->fd.file_size = current_bytes;
Record and use the tail size to prefetch table tail (#11406) Summary: **Context:** We prefetch the tail part of a SST file (i.e, the blocks after data blocks till the end of the file) during each SST file open in hope to prefetch all the stuff at once ahead of time for later read e.g, footer, meta index, filter/index etc. The existing approach to estimate the tail size to prefetch is through `TailPrefetchStats` heuristics introduced in https://github.com/facebook/rocksdb/pull/4156, which has caused small reads in unlucky case (e.g, small read into the tail buffer during table open in thread 1 under the same BlockBasedTableFactory object can make thread 2's tail prefetching use a small size that it shouldn't) and is hard to debug. Therefore we decide to record the exact tail size and use it directly to prefetch tail of the SST instead of relying heuristics. **Summary:** - Obtain and record in manifest the tail size in `BlockBasedTableBuilder::Finish()` - For backward compatibility, we fall back to TailPrefetchStats and last to simple heuristics that the tail size is a linear portion of the file size - see PR conversation for more. - Make`tail_start_offset` part of the table properties and deduct tail size to record in manifest for external files (e.g, file ingestion, import CF) and db repair (with no access to manifest). Pull Request resolved: https://github.com/facebook/rocksdb/pull/11406 Test Plan: 1. New UT 2. db bench Note: db bench on /tmp/ where direct read is supported is too slow to finish and the default pinning setting in db bench is not helpful to profile # sst read of Get. Therefore I hacked the following to obtain the following comparison. ``` diff --git a/table/block_based/block_based_table_reader.cc b/table/block_based/block_based_table_reader.cc index bd5669f0f..791484c1f 100644 --- a/table/block_based/block_based_table_reader.cc +++ b/table/block_based/block_based_table_reader.cc @@ -838,7 +838,7 @@ Status BlockBasedTable::PrefetchTail( &tail_prefetch_size); // Try file system prefetch - if (!file->use_direct_io() && !force_direct_prefetch) { + if (false && !file->use_direct_io() && !force_direct_prefetch) { if (!file->Prefetch(prefetch_off, prefetch_len, ro.rate_limiter_priority) .IsNotSupported()) { prefetch_buffer->reset(new FilePrefetchBuffer( diff --git a/tools/db_bench_tool.cc b/tools/db_bench_tool.cc index ea40f5fa0..39a0ac385 100644 --- a/tools/db_bench_tool.cc +++ b/tools/db_bench_tool.cc @@ -4191,6 +4191,8 @@ class Benchmark { std::shared_ptr<TableFactory>(NewCuckooTableFactory(table_options)); } else { BlockBasedTableOptions block_based_options; + block_based_options.metadata_cache_options.partition_pinning = + PinningTier::kAll; block_based_options.checksum = static_cast<ChecksumType>(FLAGS_checksum_type); if (FLAGS_use_hash_search) { ``` Create DB ``` ./db_bench --bloom_bits=3 --use_existing_db=1 --seed=1682546046158958 --partition_index_and_filters=1 --statistics=1 -db=/dev/shm/testdb/ -benchmarks=readrandom -key_size=3200 -value_size=512 -num=1000000 -write_buffer_size=6550000 -disable_auto_compactions=false -target_file_size_base=6550000 -compression_type=none ``` ReadRandom ``` ./db_bench --bloom_bits=3 --use_existing_db=1 --seed=1682546046158958 --partition_index_and_filters=1 --statistics=1 -db=/dev/shm/testdb/ -benchmarks=readrandom -key_size=3200 -value_size=512 -num=1000000 -write_buffer_size=6550000 -disable_auto_compactions=false -target_file_size_base=6550000 -compression_type=none ``` (a) Existing (Use TailPrefetchStats for tail size + use seperate prefetch buffer in PartitionedFilter/IndexReader::CacheDependencies()) ``` rocksdb.table.open.prefetch.tail.hit COUNT : 3395 rocksdb.sst.read.micros P50 : 5.655570 P95 : 9.931396 P99 : 14.845454 P100 : 585.000000 COUNT : 999905 SUM : 6590614 ``` (b) This PR (Record tail size + use the same tail buffer in PartitionedFilter/IndexReader::CacheDependencies()) ``` rocksdb.table.open.prefetch.tail.hit COUNT : 14257 rocksdb.sst.read.micros P50 : 5.173347 P95 : 9.015017 P99 : 12.912610 P100 : 228.000000 COUNT : 998547 SUM : 5976540 ``` As we can see, we increase the prefetch tail hit count and decrease SST read count with this PR 3. Test backward compatibility by stepping through reading with post-PR code on a db generated pre-PR. Reviewed By: pdillinger Differential Revision: D45413346 Pulled By: hx235 fbshipit-source-id: 7d5e36a60a72477218f79905168d688452a4c064
2023-05-08 20:14:28 +00:00
meta->tail_size = builder_->GetTailSize();
meta->marked_for_compaction = builder_->NeedCompact();
meta->user_defined_timestamps_persisted = static_cast<bool>(
builder_->GetTableProperties().user_defined_timestamps_persisted);
}
current_output().finished = true;
stats_.bytes_written += current_bytes;
stats_.num_output_files = outputs_.size();
return s;
}
IOStatus CompactionOutputs::WriterSyncClose(const Status& input_status,
SystemClock* clock,
Statistics* statistics,
bool use_fsync) {
IOStatus io_s;
if (input_status.ok()) {
StopWatch sw(clock, statistics, COMPACTION_OUTFILE_SYNC_MICROS);
io_s = file_writer_->Sync(use_fsync);
}
if (input_status.ok() && io_s.ok()) {
io_s = file_writer_->Close();
}
if (input_status.ok() && io_s.ok()) {
FileMetaData* meta = GetMetaData();
meta->file_checksum = file_writer_->GetFileChecksum();
meta->file_checksum_func_name = file_writer_->GetFileChecksumFuncName();
}
file_writer_.reset();
return io_s;
}
bool CompactionOutputs::UpdateFilesToCutForTTLStates(
const Slice& internal_key) {
if (!files_to_cut_for_ttl_.empty()) {
const InternalKeyComparator* icmp =
&compaction_->column_family_data()->internal_comparator();
if (cur_files_to_cut_for_ttl_ != -1) {
// Previous key is inside the range of a file
if (icmp->Compare(internal_key,
files_to_cut_for_ttl_[cur_files_to_cut_for_ttl_]
->largest.Encode()) > 0) {
next_files_to_cut_for_ttl_ = cur_files_to_cut_for_ttl_ + 1;
cur_files_to_cut_for_ttl_ = -1;
return true;
}
} else {
// Look for the key position
while (next_files_to_cut_for_ttl_ <
static_cast<int>(files_to_cut_for_ttl_.size())) {
if (icmp->Compare(internal_key,
files_to_cut_for_ttl_[next_files_to_cut_for_ttl_]
->smallest.Encode()) >= 0) {
if (icmp->Compare(internal_key,
files_to_cut_for_ttl_[next_files_to_cut_for_ttl_]
->largest.Encode()) <= 0) {
// With in the current file
cur_files_to_cut_for_ttl_ = next_files_to_cut_for_ttl_;
return true;
}
// Beyond the current file
next_files_to_cut_for_ttl_++;
} else {
// Still fall into the gap
break;
}
}
}
}
return false;
}
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
size_t CompactionOutputs::UpdateGrandparentBoundaryInfo(
const Slice& internal_key) {
size_t curr_key_boundary_switched_num = 0;
const std::vector<FileMetaData*>& grandparents = compaction_->grandparents();
if (grandparents.empty()) {
return curr_key_boundary_switched_num;
}
const Comparator* ucmp = compaction_->column_family_data()->user_comparator();
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
// Move the grandparent_index_ to the file containing the current user_key.
// If there are multiple files containing the same user_key, make sure the
// index points to the last file containing the key.
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
while (grandparent_index_ < grandparents.size()) {
if (being_grandparent_gap_) {
if (sstableKeyCompare(ucmp, internal_key,
grandparents[grandparent_index_]->smallest) < 0) {
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
break;
}
if (seen_key_) {
curr_key_boundary_switched_num++;
grandparent_overlapped_bytes_ +=
grandparents[grandparent_index_]->fd.GetFileSize();
grandparent_boundary_switched_num_++;
}
being_grandparent_gap_ = false;
} else {
int cmp_result = sstableKeyCompare(
ucmp, internal_key, grandparents[grandparent_index_]->largest);
// If it's same key, make sure grandparent_index_ is pointing to the last
// one.
if (cmp_result < 0 ||
(cmp_result == 0 &&
(grandparent_index_ == grandparents.size() - 1 ||
sstableKeyCompare(ucmp, internal_key,
grandparents[grandparent_index_ + 1]->smallest) <
0))) {
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
break;
}
if (seen_key_) {
curr_key_boundary_switched_num++;
grandparent_boundary_switched_num_++;
}
being_grandparent_gap_ = true;
grandparent_index_++;
}
}
// If the first key is in the middle of a grandparent file, adding it to the
// overlap
if (!seen_key_ && !being_grandparent_gap_) {
assert(grandparent_overlapped_bytes_ == 0);
grandparent_overlapped_bytes_ =
GetCurrentKeyGrandparentOverlappedBytes(internal_key);
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
}
seen_key_ = true;
return curr_key_boundary_switched_num;
}
uint64_t CompactionOutputs::GetCurrentKeyGrandparentOverlappedBytes(
const Slice& internal_key) const {
// no overlap with any grandparent file
if (being_grandparent_gap_) {
return 0;
}
uint64_t overlapped_bytes = 0;
const std::vector<FileMetaData*>& grandparents = compaction_->grandparents();
const Comparator* ucmp = compaction_->column_family_data()->user_comparator();
InternalKey ikey;
ikey.DecodeFrom(internal_key);
#ifndef NDEBUG
// make sure the grandparent_index_ is pointing to the last files containing
// the current key.
int cmp_result =
sstableKeyCompare(ucmp, ikey, grandparents[grandparent_index_]->largest);
assert(
cmp_result < 0 ||
(cmp_result == 0 &&
(grandparent_index_ == grandparents.size() - 1 ||
sstableKeyCompare(
ucmp, ikey, grandparents[grandparent_index_ + 1]->smallest) < 0)));
assert(sstableKeyCompare(ucmp, ikey,
grandparents[grandparent_index_]->smallest) >= 0);
#endif
overlapped_bytes += grandparents[grandparent_index_]->fd.GetFileSize();
// go backwards to find all overlapped files, one key can overlap multiple
// files. In the following example, if the current output key is `c`, and one
// compaction file was cut before `c`, current `c` can overlap with 3 files:
// [a b] [c...
// [b, b] [c, c] [c, c] [c, d]
for (int64_t i = static_cast<int64_t>(grandparent_index_) - 1;
i >= 0 && sstableKeyCompare(ucmp, ikey, grandparents[i]->largest) == 0;
i--) {
overlapped_bytes += grandparents[i]->fd.GetFileSize();
}
return overlapped_bytes;
}
bool CompactionOutputs::ShouldStopBefore(const CompactionIterator& c_iter) {
assert(c_iter.Valid());
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
const Slice& internal_key = c_iter.key();
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
#ifndef NDEBUG
bool should_stop = false;
std::pair<bool*, const Slice> p{&should_stop, internal_key};
TEST_SYNC_POINT_CALLBACK(
"CompactionOutputs::ShouldStopBefore::manual_decision", (void*)&p);
if (should_stop) {
return true;
}
#endif // NDEBUG
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
const uint64_t previous_overlapped_bytes = grandparent_overlapped_bytes_;
const InternalKeyComparator* icmp =
&compaction_->column_family_data()->internal_comparator();
size_t num_grandparent_boundaries_crossed = 0;
bool should_stop_for_ttl = false;
// Always update grandparent information like overlapped file number, size
// etc., and TTL states.
// If compaction_->output_level() == 0, there is no need to update grandparent
// info, and that `grandparent` should be empty.
if (compaction_->output_level() > 0) {
num_grandparent_boundaries_crossed =
UpdateGrandparentBoundaryInfo(internal_key);
should_stop_for_ttl = UpdateFilesToCutForTTLStates(internal_key);
}
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
if (!HasBuilder()) {
return false;
}
if (should_stop_for_ttl) {
return true;
}
// If there's user defined partitioner, check that first
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
if (partitioner_ && partitioner_->ShouldPartition(PartitionerRequest(
last_key_for_partitioner_, c_iter.user_key(),
current_output_file_size_)) == kRequired) {
return true;
}
// files output to Level 0 won't be split
if (compaction_->output_level() == 0) {
return false;
}
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
// reach the max file size
if (current_output_file_size_ >= compaction_->max_output_file_size()) {
return true;
}
// Check if it needs to split for RoundRobin
// Invalid local_output_split_key indicates that we do not need to split
if (local_output_split_key_ != nullptr && !is_split_) {
// Split occurs when the next key is larger than/equal to the cursor
if (icmp->Compare(internal_key, local_output_split_key_->Encode()) >= 0) {
is_split_ = true;
return true;
}
}
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
// only check if the current key is going to cross the grandparents file
// boundary (either the file beginning or ending).
if (num_grandparent_boundaries_crossed > 0) {
// Cut the file before the current key if the size of the current output
// file + its overlapped grandparent files is bigger than
// max_compaction_bytes. Which is to prevent future bigger than
// max_compaction_bytes compaction from the current output level.
if (grandparent_overlapped_bytes_ + current_output_file_size_ >
compaction_->max_compaction_bytes()) {
return true;
}
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
// Cut the file if including the key is going to add a skippable file on
// the grandparent level AND its size is reasonably big (1/8 of target file
// size). For example, if it's compacting the files L0 + L1:
// L0: [1, 21]
// L1: [3, 23]
// L2: [2, 4] [11, 15] [22, 24]
// Without this break, it will output as:
// L1: [1,3, 21,23]
// With this break, it will output as (assuming [11, 15] at L2 is bigger
// than 1/8 of target size):
// L1: [1,3] [21,23]
// Then for the future compactions, [11,15] won't be included.
// For random datasets (either evenly distributed or skewed), it rarely
// triggers this condition, but if the user is adding 2 different datasets
// without any overlap, it may likely happen.
// More details, check PR #1963
const size_t num_skippable_boundaries_crossed =
being_grandparent_gap_ ? 2 : 3;
if (compaction_->immutable_options()->compaction_style ==
kCompactionStyleLevel &&
compaction_->immutable_options()->level_compaction_dynamic_file_size &&
num_grandparent_boundaries_crossed >=
num_skippable_boundaries_crossed &&
grandparent_overlapped_bytes_ - previous_overlapped_bytes >
compaction_->target_output_file_size() / 8) {
return true;
}
// Pre-cut the output file if it's reaching a certain size AND it's at the
// boundary of a grandparent file. It can reduce the future compaction size,
// the cost is having smaller files.
// The pre-cut size threshold is based on how many grandparent boundaries
// it has seen before. Basically, if it has seen no boundary at all, then it
// will pre-cut at 50% target file size. Every boundary it has seen
// increases the threshold by 5%, max at 90%, which it will always cut.
// The idea is based on if it has seen more boundaries before, it will more
// likely to see another boundary (file cutting opportunity) before the
// target file size. The test shows it can generate larger files than a
// static threshold like 75% and has a similar write amplification
// improvement.
if (compaction_->immutable_options()->compaction_style ==
kCompactionStyleLevel &&
compaction_->immutable_options()->level_compaction_dynamic_file_size &&
current_output_file_size_ >=
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
((compaction_->target_output_file_size() + 99) / 100) *
(50 + std::min(grandparent_boundary_switched_num_ * 5,
size_t{40}))) {
return true;
}
}
return false;
}
Status CompactionOutputs::AddToOutput(
const CompactionIterator& c_iter,
const CompactionFileOpenFunc& open_file_func,
const CompactionFileCloseFunc& close_file_func) {
Status s;
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
bool is_range_del = c_iter.IsDeleteRangeSentinelKey();
if (is_range_del && compaction_->bottommost_level()) {
// We don't consider range tombstone for bottommost level since:
// 1. there is no grandparent and hence no overlap to consider
// 2. range tombstone may be dropped at bottommost level.
return s;
}
const Slice& key = c_iter.key();
if (ShouldStopBefore(c_iter) && HasBuilder()) {
s = close_file_func(*this, c_iter.InputStatus(), key);
if (!s.ok()) {
return s;
}
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
// reset grandparent information
grandparent_boundary_switched_num_ = 0;
grandparent_overlapped_bytes_ =
GetCurrentKeyGrandparentOverlappedBytes(key);
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
if (UNLIKELY(is_range_del)) {
// lower bound for this new output file, this is needed as the lower bound
// does not come from the smallest point key in this case.
range_tombstone_lower_bound_.DecodeFrom(key);
} else {
range_tombstone_lower_bound_.Clear();
}
}
// Open output file if necessary
if (!HasBuilder()) {
s = open_file_func(*this);
if (!s.ok()) {
return s;
}
}
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
// c_iter may emit range deletion keys, so update `last_key_for_partitioner_`
// here before returning below when `is_range_del` is true
if (partitioner_) {
last_key_for_partitioner_.assign(c_iter.user_key().data_,
c_iter.user_key().size_);
}
if (UNLIKELY(is_range_del)) {
return s;
}
assert(builder_ != nullptr);
const Slice& value = c_iter.value();
Align compaction output file boundaries to the next level ones (#10655) Summary: Try to align the compaction output file boundaries to the next level ones (grandparent level), to reduce the level compaction write-amplification. In level compaction, there are "wasted" data at the beginning and end of the output level files. Align the file boundary can avoid such "wasted" compaction. With this PR, it tries to align the non-bottommost level file boundaries to its next level ones. It may cut file when the file size is large enough (at least 50% of target_file_size) and not too large (2x target_file_size). db_bench shows about 12.56% compaction reduction: ``` TEST_TMPDIR=/data/dbbench2 ./db_bench --benchmarks=fillrandom,readrandom -max_background_jobs=12 -num=400000000 -target_file_size_base=33554432 # baseline: Flush(GB): cumulative 25.882, interval 7.216 Cumulative compaction: 285.90 GB write, 162.36 MB/s write, 269.68 GB read, 153.15 MB/s read, 2926.7 seconds # with this change: Flush(GB): cumulative 25.882, interval 7.753 Cumulative compaction: 249.97 GB write, 141.96 MB/s write, 233.74 GB read, 132.74 MB/s read, 2534.9 seconds ``` The compaction simulator shows a similar result (14% with 100G random data). As a side effect, with this PR, the SST file size can exceed the target_file_size, but is capped at 2x target_file_size. And there will be smaller files. Here are file size statistics when loading 100GB with the target file size 32MB: ``` baseline this_PR count 1.656000e+03 1.705000e+03 mean 3.116062e+07 3.028076e+07 std 7.145242e+06 8.046139e+06 ``` The feature is enabled by default, to revert to the old behavior disable it with `AdvancedColumnFamilyOptions.level_compaction_dynamic_file_size = false` Also includes https://github.com/facebook/rocksdb/issues/1963 to cut file before skippable grandparent file. Which is for use case like user adding 2 or more non-overlapping data range at the same time, it can reduce the overlapping of 2 datasets in the lower levels. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10655 Reviewed By: cbi42 Differential Revision: D39552321 Pulled By: jay-zhuang fbshipit-source-id: 640d15f159ab0cd973f2426cfc3af266fc8bdde2
2022-09-30 02:43:55 +00:00
s = current_output().validator.Add(key, value);
if (!s.ok()) {
return s;
}
builder_->Add(key, value);
stats_.num_output_records++;
current_output_file_size_ = builder_->EstimatedFileSize();
if (blob_garbage_meter_) {
s = blob_garbage_meter_->ProcessOutFlow(key, value);
}
if (!s.ok()) {
return s;
}
const ParsedInternalKey& ikey = c_iter.ikey();
s = current_output().meta.UpdateBoundaries(key, value, ikey.sequence,
ikey.type);
return s;
}
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
namespace {
void SetMaxSeqAndTs(InternalKey& internal_key, const Slice& user_key,
const size_t ts_sz) {
if (ts_sz) {
static constexpr char kTsMax[] = "\xff\xff\xff\xff\xff\xff\xff\xff\xff";
if (ts_sz <= strlen(kTsMax)) {
internal_key = InternalKey(user_key, kMaxSequenceNumber,
kTypeRangeDeletion, Slice(kTsMax, ts_sz));
} else {
internal_key =
InternalKey(user_key, kMaxSequenceNumber, kTypeRangeDeletion,
std::string(ts_sz, '\xff'));
}
} else {
internal_key.Set(user_key, kMaxSequenceNumber, kTypeRangeDeletion);
}
}
} // namespace
Status CompactionOutputs::AddRangeDels(
User-defined timestamp support for `DeleteRange()` (#10661) Summary: Add user-defined timestamp support for range deletion. The new API is `DeleteRange(opt, cf, begin_key, end_key, ts)`. Most of the change is to update the comparator to compare without timestamp. Other than that, major changes are - internal range tombstone data structures (`FragmentedRangeTombstoneList`, `RangeTombstone`, etc.) to store timestamps. - Garbage collection of range tombstones and range tombstone covered keys during compaction. - Get()/MultiGet() to return the timestamp of a range tombstone when needed. - Get/Iterator with range tombstones bounded by readoptions.timestamp. - timestamp crash test now issues DeleteRange by default. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10661 Test Plan: - Added unit test: `make check` - Stress test: `python3 tools/db_crashtest.py --enable_ts whitebox --readpercent=57 --prefixpercent=4 --writepercent=25 -delpercent=5 --iterpercent=5 --delrangepercent=4` - Ran `db_bench` to measure regression when timestamp is not enabled. The tests are for write (with some range deletion) and iterate with DB fitting in memory: `./db_bench--benchmarks=fillrandom,seekrandom --writes_per_range_tombstone=200 --max_write_buffer_number=100 --min_write_buffer_number_to_merge=100 --writes=500000 --reads=500000 --seek_nexts=10 --disable_auto_compactions -disable_wal=true --max_num_range_tombstones=1000`. Did not see consistent regression in no timestamp case. | micros/op | fillrandom | seekrandom | | --- | --- | --- | |main| 2.58 |10.96| |PR 10661| 2.68 |10.63| Reviewed By: riversand963 Differential Revision: D39441192 Pulled By: cbi42 fbshipit-source-id: f05aca3c41605caf110daf0ff405919f300ddec2
2022-09-30 23:13:03 +00:00
const Slice* comp_start_user_key, const Slice* comp_end_user_key,
CompactionIterationStats& range_del_out_stats, bool bottommost_level,
const InternalKeyComparator& icmp, SequenceNumber earliest_snapshot,
User-defined timestamp support for `DeleteRange()` (#10661) Summary: Add user-defined timestamp support for range deletion. The new API is `DeleteRange(opt, cf, begin_key, end_key, ts)`. Most of the change is to update the comparator to compare without timestamp. Other than that, major changes are - internal range tombstone data structures (`FragmentedRangeTombstoneList`, `RangeTombstone`, etc.) to store timestamps. - Garbage collection of range tombstones and range tombstone covered keys during compaction. - Get()/MultiGet() to return the timestamp of a range tombstone when needed. - Get/Iterator with range tombstones bounded by readoptions.timestamp. - timestamp crash test now issues DeleteRange by default. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10661 Test Plan: - Added unit test: `make check` - Stress test: `python3 tools/db_crashtest.py --enable_ts whitebox --readpercent=57 --prefixpercent=4 --writepercent=25 -delpercent=5 --iterpercent=5 --delrangepercent=4` - Ran `db_bench` to measure regression when timestamp is not enabled. The tests are for write (with some range deletion) and iterate with DB fitting in memory: `./db_bench--benchmarks=fillrandom,seekrandom --writes_per_range_tombstone=200 --max_write_buffer_number=100 --min_write_buffer_number_to_merge=100 --writes=500000 --reads=500000 --seek_nexts=10 --disable_auto_compactions -disable_wal=true --max_num_range_tombstones=1000`. Did not see consistent regression in no timestamp case. | micros/op | fillrandom | seekrandom | | --- | --- | --- | |main| 2.58 |10.96| |PR 10661| 2.68 |10.63| Reviewed By: riversand963 Differential Revision: D39441192 Pulled By: cbi42 fbshipit-source-id: f05aca3c41605caf110daf0ff405919f300ddec2
2022-09-30 23:13:03 +00:00
const Slice& next_table_min_key, const std::string& full_history_ts_low) {
// The following example does not happen since
// CompactionOutput::ShouldStopBefore() always return false for the first
// point key. But we should consider removing this dependency. Suppose for the
// first compaction output file,
// - next_table_min_key.user_key == comp_start_user_key
// - no point key is in the output file
// - there is a range tombstone @seqno to be added that covers
// comp_start_user_key
// Then meta.smallest will be set to comp_start_user_key@seqno
// and meta.largest will be set to comp_start_user_key@kMaxSequenceNumber
// which violates the assumption that meta.smallest should be <= meta.largest.
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
assert(HasRangeDel());
FileMetaData& meta = current_output().meta;
const Comparator* ucmp = icmp.user_comparator();
InternalKey lower_bound_buf, upper_bound_buf;
Slice lower_bound_guard, upper_bound_guard;
std::string smallest_user_key;
const Slice *lower_bound, *upper_bound;
// We first determine the internal key lower_bound and upper_bound for
// this output file. All and only range tombstones that overlap with
// [lower_bound, upper_bound] should be added to this file. File
// boundaries (meta.smallest/largest) should be updated accordingly when
// extended by range tombstones.
size_t output_size = outputs_.size();
if (output_size == 1) {
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
// This is the first file in the subcompaction.
//
// When outputting a range tombstone that spans a subcompaction boundary,
// the files on either side of that boundary need to include that
// boundary's user key. Otherwise, the spanning range tombstone would lose
// coverage.
//
// To achieve this while preventing files from overlapping in internal key
// (an LSM invariant violation), we allow the earlier file to include the
// boundary user key up to `kMaxSequenceNumber,kTypeRangeDeletion`. The
// later file can begin at the boundary user key at the newest key version
// it contains. At this point that version number is unknown since we have
// not processed the range tombstones yet, so permit any version. Same story
// applies to timestamp, and a non-nullptr `comp_start_user_key` should have
// `kMaxTs` here, which similarly permits any timestamp.
if (comp_start_user_key) {
lower_bound_buf.Set(*comp_start_user_key, kMaxSequenceNumber,
kTypeRangeDeletion);
lower_bound_guard = lower_bound_buf.Encode();
lower_bound = &lower_bound_guard;
} else {
lower_bound = nullptr;
}
} else {
// For subsequent output tables, only include range tombstones from min
// key onwards since the previous file was extended to contain range
// tombstones falling before min key.
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
if (range_tombstone_lower_bound_.size() > 0) {
assert(meta.smallest.size() == 0 ||
icmp.Compare(range_tombstone_lower_bound_, meta.smallest) < 0);
lower_bound_guard = range_tombstone_lower_bound_.Encode();
} else {
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
assert(meta.smallest.size() > 0);
lower_bound_guard = meta.smallest.Encode();
}
lower_bound = &lower_bound_guard;
}
const size_t ts_sz = ucmp->timestamp_size();
if (next_table_min_key.empty()) {
// Last file of the subcompaction.
if (comp_end_user_key) {
upper_bound_buf.Set(*comp_end_user_key, kMaxSequenceNumber,
kTypeRangeDeletion);
upper_bound_guard = upper_bound_buf.Encode();
upper_bound = &upper_bound_guard;
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
} else {
upper_bound = nullptr;
}
} else {
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
// There is another file coming whose coverage will begin at
// `next_table_min_key`. The current file needs to extend range tombstone
// coverage through its own keys (through `meta.largest`) and through user
// keys preceding `next_table_min_key`'s user key.
ParsedInternalKey next_table_min_key_parsed;
ParseInternalKey(next_table_min_key, &next_table_min_key_parsed,
false /* log_err_key */)
.PermitUncheckedError();
assert(next_table_min_key_parsed.sequence < kMaxSequenceNumber);
assert(meta.largest.size() == 0 ||
icmp.Compare(meta.largest.Encode(), next_table_min_key) < 0);
assert(!lower_bound || icmp.Compare(*lower_bound, next_table_min_key) <= 0);
if (meta.largest.size() > 0 &&
ucmp->EqualWithoutTimestamp(meta.largest.user_key(),
next_table_min_key_parsed.user_key)) {
// Caution: this assumes meta.largest.Encode() lives longer than
// upper_bound, which is only true if meta.largest is never updated.
// This just happens to be the case here since meta.largest serves
// as the upper_bound.
upper_bound_guard = meta.largest.Encode();
} else {
SetMaxSeqAndTs(upper_bound_buf, next_table_min_key_parsed.user_key,
ts_sz);
upper_bound_guard = upper_bound_buf.Encode();
}
upper_bound = &upper_bound_guard;
}
if (lower_bound && upper_bound &&
icmp.Compare(*lower_bound, *upper_bound) > 0) {
assert(meta.smallest.size() == 0 &&
ucmp->EqualWithoutTimestamp(ExtractUserKey(*lower_bound),
ExtractUserKey(*upper_bound)));
// This can only happen when lower_bound have the same user key as
// next_table_min_key and that there is no point key in the current
// compaction output file.
return Status::OK();
}
// The end key of the subcompaction must be bigger or equal to the upper
// bound. If the end of subcompaction is null or the upper bound is null,
// it means that this file is the last file in the compaction. So there
// will be no overlapping between this file and others.
User-defined timestamp support for `DeleteRange()` (#10661) Summary: Add user-defined timestamp support for range deletion. The new API is `DeleteRange(opt, cf, begin_key, end_key, ts)`. Most of the change is to update the comparator to compare without timestamp. Other than that, major changes are - internal range tombstone data structures (`FragmentedRangeTombstoneList`, `RangeTombstone`, etc.) to store timestamps. - Garbage collection of range tombstones and range tombstone covered keys during compaction. - Get()/MultiGet() to return the timestamp of a range tombstone when needed. - Get/Iterator with range tombstones bounded by readoptions.timestamp. - timestamp crash test now issues DeleteRange by default. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10661 Test Plan: - Added unit test: `make check` - Stress test: `python3 tools/db_crashtest.py --enable_ts whitebox --readpercent=57 --prefixpercent=4 --writepercent=25 -delpercent=5 --iterpercent=5 --delrangepercent=4` - Ran `db_bench` to measure regression when timestamp is not enabled. The tests are for write (with some range deletion) and iterate with DB fitting in memory: `./db_bench--benchmarks=fillrandom,seekrandom --writes_per_range_tombstone=200 --max_write_buffer_number=100 --min_write_buffer_number_to_merge=100 --writes=500000 --reads=500000 --seek_nexts=10 --disable_auto_compactions -disable_wal=true --max_num_range_tombstones=1000`. Did not see consistent regression in no timestamp case. | micros/op | fillrandom | seekrandom | | --- | --- | --- | |main| 2.58 |10.96| |PR 10661| 2.68 |10.63| Reviewed By: riversand963 Differential Revision: D39441192 Pulled By: cbi42 fbshipit-source-id: f05aca3c41605caf110daf0ff405919f300ddec2
2022-09-30 23:13:03 +00:00
assert(comp_end_user_key == nullptr || upper_bound == nullptr ||
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
ucmp->CompareWithoutTimestamp(ExtractUserKey(*upper_bound),
*comp_end_user_key) <= 0);
auto it = range_del_agg_->NewIterator(lower_bound, upper_bound);
Slice last_tombstone_start_user_key{};
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
bool reached_lower_bound = false;
Group rocksdb.sst.read.micros stat by IOActivity flush and compaction (#11288) Summary: **Context:** The existing stat rocksdb.sst.read.micros does not reflect each of compaction and flush cases but aggregate them, which is not so helpful for us to understand IO read behavior of each of them. **Summary** - Update `StopWatch` and `RandomAccessFileReader` to record `rocksdb.sst.read.micros` and `rocksdb.file.{flush/compaction}.read.micros` - Fixed the default histogram in `RandomAccessFileReader` - New field `ReadOptions/IOOptions::io_activity`; Pass `ReadOptions` through paths under db open, flush and compaction to where we can prepare `IOOptions` and pass it to `RandomAccessFileReader` - Use `thread_status_util` for assertion in `DbStressFSWrapper` for continuous testing on we are passing correct `io_activity` under db open, flush and compaction Pull Request resolved: https://github.com/facebook/rocksdb/pull/11288 Test Plan: - **Stress test** - **Db bench 1: rocksdb.sst.read.micros COUNT ≈ sum of rocksdb.file.read.flush.micros's and rocksdb.file.read.compaction.micros's.** (without blob) - May not be exactly the same due to `HistogramStat::Add` only guarantees atomic not accuracy across threads. ``` ./db_bench -db=/dev/shm/testdb/ -statistics=true -benchmarks="fillseq" -key_size=32 -value_size=512 -num=50000 -write_buffer_size=655 -target_file_size_base=655 -disable_auto_compactions=false -compression_type=none -bloom_bits=3 (-use_plain_table=1 -prefix_size=10) ``` ``` // BlockBasedTable rocksdb.sst.read.micros P50 : 2.009374 P95 : 4.968548 P99 : 8.110362 P100 : 43.000000 COUNT : 40456 SUM : 114805 rocksdb.file.read.flush.micros P50 : 1.871841 P95 : 3.872407 P99 : 5.540541 P100 : 43.000000 COUNT : 2250 SUM : 6116 rocksdb.file.read.compaction.micros P50 : 2.023109 P95 : 5.029149 P99 : 8.196910 P100 : 26.000000 COUNT : 38206 SUM : 108689 // PlainTable Does not apply ``` - **Db bench 2: performance** **Read** SETUP: db with 900 files ``` ./db_bench -db=/dev/shm/testdb/ -benchmarks="fillseq" -key_size=32 -value_size=512 -num=50000 -write_buffer_size=655 -disable_auto_compactions=true -target_file_size_base=655 -compression_type=none ```run till convergence ``` ./db_bench -seed=1678564177044286 -use_existing_db=true -db=/dev/shm/testdb -benchmarks=readrandom[-X60] -statistics=true -num=1000000 -disable_auto_compactions=true -compression_type=none -bloom_bits=3 ``` Pre-change `readrandom [AVG 60 runs] : 21568 (± 248) ops/sec` Post-change (no regression, -0.3%) `readrandom [AVG 60 runs] : 21486 (± 236) ops/sec` **Compaction/Flush**run till convergence ``` ./db_bench -db=/dev/shm/testdb2/ -seed=1678564177044286 -benchmarks="fillseq[-X60]" -key_size=32 -value_size=512 -num=50000 -write_buffer_size=655 -disable_auto_compactions=false -target_file_size_base=655 -compression_type=none rocksdb.sst.read.micros COUNT : 33820 rocksdb.sst.read.flush.micros COUNT : 1800 rocksdb.sst.read.compaction.micros COUNT : 32020 ``` Pre-change `fillseq [AVG 46 runs] : 1391 (± 214) ops/sec; 0.7 (± 0.1) MB/sec` Post-change (no regression, ~-0.4%) `fillseq [AVG 46 runs] : 1385 (± 216) ops/sec; 0.7 (± 0.1) MB/sec` Reviewed By: ajkr Differential Revision: D44007011 Pulled By: hx235 fbshipit-source-id: a54c89e4846dfc9a135389edf3f3eedfea257132
2023-04-21 16:07:18 +00:00
const ReadOptions read_options(Env::IOActivity::kCompaction);
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
for (it->SeekToFirst(); it->Valid(); it->Next()) {
auto tombstone = it->Tombstone();
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
auto kv = tombstone.Serialize();
InternalKey tombstone_end = tombstone.SerializeEndKey();
// TODO: the underlying iterator should support clamping the bounds.
// tombstone_end.Encode is of form user_key@kMaxSeqno
// if it is equal to lower_bound, there is no need to include
// such range tombstone.
if (!reached_lower_bound && lower_bound &&
icmp.Compare(tombstone_end.Encode(), *lower_bound) <= 0) {
continue;
}
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
assert(!lower_bound ||
icmp.Compare(*lower_bound, tombstone_end.Encode()) <= 0);
reached_lower_bound = true;
User-defined timestamp support for `DeleteRange()` (#10661) Summary: Add user-defined timestamp support for range deletion. The new API is `DeleteRange(opt, cf, begin_key, end_key, ts)`. Most of the change is to update the comparator to compare without timestamp. Other than that, major changes are - internal range tombstone data structures (`FragmentedRangeTombstoneList`, `RangeTombstone`, etc.) to store timestamps. - Garbage collection of range tombstones and range tombstone covered keys during compaction. - Get()/MultiGet() to return the timestamp of a range tombstone when needed. - Get/Iterator with range tombstones bounded by readoptions.timestamp. - timestamp crash test now issues DeleteRange by default. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10661 Test Plan: - Added unit test: `make check` - Stress test: `python3 tools/db_crashtest.py --enable_ts whitebox --readpercent=57 --prefixpercent=4 --writepercent=25 -delpercent=5 --iterpercent=5 --delrangepercent=4` - Ran `db_bench` to measure regression when timestamp is not enabled. The tests are for write (with some range deletion) and iterate with DB fitting in memory: `./db_bench--benchmarks=fillrandom,seekrandom --writes_per_range_tombstone=200 --max_write_buffer_number=100 --min_write_buffer_number_to_merge=100 --writes=500000 --reads=500000 --seek_nexts=10 --disable_auto_compactions -disable_wal=true --max_num_range_tombstones=1000`. Did not see consistent regression in no timestamp case. | micros/op | fillrandom | seekrandom | | --- | --- | --- | |main| 2.58 |10.96| |PR 10661| 2.68 |10.63| Reviewed By: riversand963 Differential Revision: D39441192 Pulled By: cbi42 fbshipit-source-id: f05aca3c41605caf110daf0ff405919f300ddec2
2022-09-30 23:13:03 +00:00
// Garbage collection for range tombstones.
// If user-defined timestamp is enabled, range tombstones are dropped if
// they are at bottommost_level, below full_history_ts_low and not visible
// in any snapshot. trim_ts_ is passed to the constructor for
// range_del_agg_, and range_del_agg_ internally drops tombstones above
// trim_ts_.
bool consider_drop =
tombstone.seq_ <= earliest_snapshot &&
User-defined timestamp support for `DeleteRange()` (#10661) Summary: Add user-defined timestamp support for range deletion. The new API is `DeleteRange(opt, cf, begin_key, end_key, ts)`. Most of the change is to update the comparator to compare without timestamp. Other than that, major changes are - internal range tombstone data structures (`FragmentedRangeTombstoneList`, `RangeTombstone`, etc.) to store timestamps. - Garbage collection of range tombstones and range tombstone covered keys during compaction. - Get()/MultiGet() to return the timestamp of a range tombstone when needed. - Get/Iterator with range tombstones bounded by readoptions.timestamp. - timestamp crash test now issues DeleteRange by default. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10661 Test Plan: - Added unit test: `make check` - Stress test: `python3 tools/db_crashtest.py --enable_ts whitebox --readpercent=57 --prefixpercent=4 --writepercent=25 -delpercent=5 --iterpercent=5 --delrangepercent=4` - Ran `db_bench` to measure regression when timestamp is not enabled. The tests are for write (with some range deletion) and iterate with DB fitting in memory: `./db_bench--benchmarks=fillrandom,seekrandom --writes_per_range_tombstone=200 --max_write_buffer_number=100 --min_write_buffer_number_to_merge=100 --writes=500000 --reads=500000 --seek_nexts=10 --disable_auto_compactions -disable_wal=true --max_num_range_tombstones=1000`. Did not see consistent regression in no timestamp case. | micros/op | fillrandom | seekrandom | | --- | --- | --- | |main| 2.58 |10.96| |PR 10661| 2.68 |10.63| Reviewed By: riversand963 Differential Revision: D39441192 Pulled By: cbi42 fbshipit-source-id: f05aca3c41605caf110daf0ff405919f300ddec2
2022-09-30 23:13:03 +00:00
(ts_sz == 0 ||
(!full_history_ts_low.empty() &&
ucmp->CompareTimestamp(tombstone.ts_, full_history_ts_low) < 0));
if (consider_drop && bottommost_level) {
// TODO(andrewkr): tombstones that span multiple output files are
// counted for each compaction output file, so lots of double
// counting.
range_del_out_stats.num_range_del_drop_obsolete++;
range_del_out_stats.num_record_drop_obsolete++;
continue;
}
assert(lower_bound == nullptr ||
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
ucmp->CompareWithoutTimestamp(ExtractUserKey(*lower_bound),
kv.second) < 0);
InternalKey tombstone_start = kv.first;
if (lower_bound &&
ucmp->CompareWithoutTimestamp(tombstone_start.user_key(),
ExtractUserKey(*lower_bound)) < 0) {
// This just updates the non-timestamp portion of `tombstone_start`'s user
// key. Ideally there would be a simpler API usage
ParsedInternalKey tombstone_start_parsed;
ParseInternalKey(tombstone_start.Encode(), &tombstone_start_parsed,
false /* log_err_key */)
.PermitUncheckedError();
// timestamp should be from where sequence number is from, which is from
// tombstone in this case
std::string ts =
tombstone_start_parsed.GetTimestamp(ucmp->timestamp_size())
.ToString();
tombstone_start_parsed.user_key = ExtractUserKey(*lower_bound);
tombstone_start.SetFrom(tombstone_start_parsed, ts);
}
if (upper_bound != nullptr &&
icmp.Compare(*upper_bound, tombstone_start.Encode()) < 0) {
break;
}
if (lower_bound &&
icmp.Compare(tombstone_start.Encode(), *lower_bound) < 0) {
tombstone_start.DecodeFrom(*lower_bound);
}
if (upper_bound && icmp.Compare(*upper_bound, tombstone_end.Encode()) < 0) {
tombstone_end.DecodeFrom(*upper_bound);
}
if (consider_drop && compaction_->KeyRangeNotExistsBeyondOutputLevel(
tombstone_start.user_key(),
tombstone_end.user_key(), &level_ptrs_)) {
range_del_out_stats.num_range_del_drop_obsolete++;
range_del_out_stats.num_record_drop_obsolete++;
continue;
}
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
// Here we show that *only* range tombstones that overlap with
// [lower_bound, upper_bound] are added to the current file, and
// sanity checking invariants that should hold:
// - [tombstone_start, tombstone_end] overlaps with [lower_bound,
// upper_bound]
// - meta.smallest <= meta.largest
// Corresponding assertions are made, the proof is broken is any of them
// fails.
// TODO: show that *all* range tombstones that overlap with
// [lower_bound, upper_bound] are added.
// TODO: some invariant about boundaries are correctly updated.
//
// Note that `tombstone_start` is updated in the if condition above, we use
// tombstone_start to refer to its initial value, i.e.,
// it->Tombstone().first, and use tombstone_start* to refer to its value
// after the update.
//
// To show [lower_bound, upper_bound] overlaps with [tombstone_start,
// tombstone_end]:
// lower_bound <= upper_bound from the if condition right after all
// bounds are initialized. We assume each tombstone fragment has
// start_key.user_key < end_key.user_key, so
// tombstone_start < tombstone_end by
// FragmentedTombstoneIterator::Tombstone(). So these two ranges are both
// non-emtpy. The flag `reached_lower_bound` and the if logic before it
// ensures lower_bound <= tombstone_end. tombstone_start is only updated
// if it has a smaller user_key than lower_bound user_key, so
// tombstone_start <= tombstone_start*. The above if condition implies
// tombstone_start* <= upper_bound. So we have
// tombstone_start <= upper_bound and lower_bound <= tombstone_end
// and the two ranges overlap.
//
// To show meta.smallest <= meta.largest:
// From the implementation of UpdateBoundariesForRange(), it suffices to
// prove that when it is first called in this function, its parameters
// satisfy `start <= end`, where start = max(tombstone_start*, lower_bound)
// and end = min(tombstone_end, upper_bound). From the above proof we have
// lower_bound <= tombstone_end and lower_bound <= upper_bound. We only need
// to show that tombstone_start* <= min(tombstone_end, upper_bound).
// Note that tombstone_start*.user_key = max(tombstone_start.user_key,
// lower_bound.user_key). Assuming tombstone_end always has
// kMaxSequenceNumber and lower_bound.seqno < kMaxSequenceNumber.
// Since lower_bound <= tombstone_end and lower_bound.seqno <
// tombstone_end.seqno (in absolute number order, not internal key order),
// lower_bound.user_key < tombstone_end.user_key.
// Since lower_bound.user_key < tombstone_end.user_key and
// tombstone_start.user_key < tombstone_end.user_key, tombstone_start* <
// tombstone_end. Since tombstone_start* <= upper_bound from the above proof
// and tombstone_start* < tombstone_end, tombstone_start* <=
// min(tombstone_end, upper_bound), so the two ranges overlap.
// Range tombstone is not supported by output validator yet.
builder_->Add(kv.first.Encode(), kv.second);
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
assert(icmp.Compare(tombstone_start, tombstone_end) <= 0);
meta.UpdateBoundariesForRange(tombstone_start, tombstone_end,
tombstone.seq_, icmp);
Include estimated bytes deleted by range tombstones in compensated file size (#10734) Summary: compensate file sizes in compaction picking so files with range tombstones are preferred, such that they get compacted down earlier as they tend to delete a lot of data. This PR adds a `compensated_range_deletion_size` field in FileMeta that is computed during Flush/Compaction and persisted in MANIFEST. This value is added to `compensated_file_size` which will be used for compaction picking. Currently, for a file in level L, `compensated_range_deletion_size` is set to the estimated bytes deleted by range tombstone of this file in all levels > L. This helps to reduce space amp when data in older levels are covered by range tombstones in level L. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10734 Test Plan: - Added unit tests. - benchmark to check if the above definition `compensated_range_deletion_size` is reducing space amp as intended, without affecting write amp too much. The experiment set up favorable for this optimization: large range tombstone issued infrequently. Command used: ``` ./db_bench -benchmarks=fillrandom,waitforcompaction,stats,levelstats -use_existing_db=false -avoid_flush_during_recovery=true -write_buffer_size=33554432 -level_compaction_dynamic_level_bytes=true -max_background_jobs=8 -max_bytes_for_level_base=134217728 -target_file_size_base=33554432 -writes_per_range_tombstone=500000 -range_tombstone_width=5000000 -num=50000000 -benchmark_write_rate_limit=8388608 -threads=16 -duration=1800 --max_num_range_tombstones=1000000000 ``` In this experiment, each thread wrote 16 range tombstones over the duration of 30 minutes, each range tombstone has width 5M that is the 10% of the key space width. Results shows this PR generates a smaller DB size. Compaction stats from this PR: ``` Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) CompMergeCPU(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Rblob(GB) Wblob(GB) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 2/0 31.54 MB 0.5 0.0 0.0 0.0 8.4 8.4 0.0 1.0 0.0 63.4 135.56 110.94 544 0.249 0 0 0.0 0.0 L4 3/0 96.55 MB 0.8 18.5 6.7 11.8 18.4 6.6 0.0 2.7 65.3 64.9 290.08 284.03 108 2.686 284M 1957K 0.0 0.0 L5 15/0 404.41 MB 1.0 19.1 7.7 11.4 18.8 7.4 0.3 2.5 66.6 65.7 292.93 285.34 220 1.332 293M 3808K 0.0 0.0 L6 143/0 4.12 GB 0.0 45.0 7.5 37.5 41.6 4.1 0.0 5.5 71.2 65.9 647.00 632.66 251 2.578 739M 47M 0.0 0.0 Sum 163/0 4.64 GB 0.0 82.6 21.9 60.7 87.2 26.5 0.3 10.4 61.9 65.4 1365.58 1312.97 1123 1.216 1318M 52M 0.0 0.0 ``` Compaction stats from main: ``` Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) CompMergeCPU(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Rblob(GB) Wblob(GB) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 0/0 0.00 KB 0.0 0.0 0.0 0.0 8.4 8.4 0.0 1.0 0.0 60.5 142.12 115.89 569 0.250 0 0 0.0 0.0 L4 3/0 85.68 MB 1.0 17.7 6.8 10.9 17.6 6.7 0.0 2.6 62.7 62.3 289.05 281.79 112 2.581 272M 2309K 0.0 0.0 L5 11/0 293.73 MB 1.0 18.8 7.5 11.2 18.5 7.2 0.5 2.5 64.9 63.9 296.07 288.50 220 1.346 288M 4365K 0.0 0.0 L6 130/0 3.94 GB 0.0 51.5 7.6 43.9 47.9 3.9 0.0 6.3 67.2 62.4 784.95 765.92 258 3.042 848M 51M 0.0 0.0 Sum 144/0 4.31 GB 0.0 88.0 21.9 66.0 92.3 26.3 0.5 11.0 59.6 62.5 1512.19 1452.09 1159 1.305 1409M 58M 0.0 0.0``` Reviewed By: ajkr Differential Revision: D39834713 Pulled By: cbi42 fbshipit-source-id: fe9341040b8704a8fbb10cad5cf5c43e962c7e6b
2022-12-29 21:28:24 +00:00
if (!bottommost_level) {
bool start_user_key_changed =
last_tombstone_start_user_key.empty() ||
ucmp->CompareWithoutTimestamp(last_tombstone_start_user_key,
it->start_key()) < 0;
last_tombstone_start_user_key = it->start_key();
if (start_user_key_changed) {
Refactor AddRangeDels() + consider range tombstone during compaction file cutting (#11113) Summary: A second attempt after https://github.com/facebook/rocksdb/issues/10802, with bug fixes and refactoring. This PR updates compaction logic to take range tombstones into account when determining whether to cut the current compaction output file (https://github.com/facebook/rocksdb/issues/4811). Before this change, only point keys were considered, and range tombstones could cause large compactions. For example, if the current compaction outputs is a range tombstone [a, b) and 2 point keys y, z, they would be added to the same file, and may overlap with too many files in the next level and cause a large compaction in the future. This PR also includes ajkr's effort to simplify the logic to add range tombstones to compaction output files in `AddRangeDels()` ([https://github.com/facebook/rocksdb/issues/11078](https://github.com/facebook/rocksdb/pull/11078#issuecomment-1386078861)). The main change is for `CompactionIterator` to emit range tombstone start keys to be processed by `CompactionOutputs`. A new class `CompactionMergingIterator` is introduced to replace `MergingIterator` under `CompactionIterator` to enable emitting of range tombstone start keys. Further improvement after this PR include cutting compaction output at some grandparent boundary key (instead of the next output key) when cutting within a range tombstone to reduce overlap with grandparents. Pull Request resolved: https://github.com/facebook/rocksdb/pull/11113 Test Plan: * added unit test in db_range_del_test * crash test with a small key range: `python3 tools/db_crashtest.py blackbox --simple --max_key=100 --interval=600 --write_buffer_size=262144 --target_file_size_base=256 --max_bytes_for_level_base=262144 --block_size=128 --value_size_mult=33 --subcompactions=10 --use_multiget=1 --delpercent=3 --delrangepercent=2 --verify_iterator_with_expected_state_one_in=2 --num_iterations=10` Reviewed By: ajkr Differential Revision: D42655709 Pulled By: cbi42 fbshipit-source-id: 8367e36ef5640e8f21c14a3855d4a8d6e360a34c
2023-02-22 20:28:18 +00:00
// If tombstone_start >= tombstone_end, then either no key range is
// covered, or that they have the same user key. If they have the same
// user key, then the internal key range should only be within this
// level, and no keys from older levels is covered.
if (ucmp->CompareWithoutTimestamp(tombstone_start.user_key(),
tombstone_end.user_key()) < 0) {
SizeApproximationOptions approx_opts;
approx_opts.files_size_error_margin = 0.1;
auto approximate_covered_size =
compaction_->input_version()->version_set()->ApproximateSize(
Group rocksdb.sst.read.micros stat by IOActivity flush and compaction (#11288) Summary: **Context:** The existing stat rocksdb.sst.read.micros does not reflect each of compaction and flush cases but aggregate them, which is not so helpful for us to understand IO read behavior of each of them. **Summary** - Update `StopWatch` and `RandomAccessFileReader` to record `rocksdb.sst.read.micros` and `rocksdb.file.{flush/compaction}.read.micros` - Fixed the default histogram in `RandomAccessFileReader` - New field `ReadOptions/IOOptions::io_activity`; Pass `ReadOptions` through paths under db open, flush and compaction to where we can prepare `IOOptions` and pass it to `RandomAccessFileReader` - Use `thread_status_util` for assertion in `DbStressFSWrapper` for continuous testing on we are passing correct `io_activity` under db open, flush and compaction Pull Request resolved: https://github.com/facebook/rocksdb/pull/11288 Test Plan: - **Stress test** - **Db bench 1: rocksdb.sst.read.micros COUNT ≈ sum of rocksdb.file.read.flush.micros's and rocksdb.file.read.compaction.micros's.** (without blob) - May not be exactly the same due to `HistogramStat::Add` only guarantees atomic not accuracy across threads. ``` ./db_bench -db=/dev/shm/testdb/ -statistics=true -benchmarks="fillseq" -key_size=32 -value_size=512 -num=50000 -write_buffer_size=655 -target_file_size_base=655 -disable_auto_compactions=false -compression_type=none -bloom_bits=3 (-use_plain_table=1 -prefix_size=10) ``` ``` // BlockBasedTable rocksdb.sst.read.micros P50 : 2.009374 P95 : 4.968548 P99 : 8.110362 P100 : 43.000000 COUNT : 40456 SUM : 114805 rocksdb.file.read.flush.micros P50 : 1.871841 P95 : 3.872407 P99 : 5.540541 P100 : 43.000000 COUNT : 2250 SUM : 6116 rocksdb.file.read.compaction.micros P50 : 2.023109 P95 : 5.029149 P99 : 8.196910 P100 : 26.000000 COUNT : 38206 SUM : 108689 // PlainTable Does not apply ``` - **Db bench 2: performance** **Read** SETUP: db with 900 files ``` ./db_bench -db=/dev/shm/testdb/ -benchmarks="fillseq" -key_size=32 -value_size=512 -num=50000 -write_buffer_size=655 -disable_auto_compactions=true -target_file_size_base=655 -compression_type=none ```run till convergence ``` ./db_bench -seed=1678564177044286 -use_existing_db=true -db=/dev/shm/testdb -benchmarks=readrandom[-X60] -statistics=true -num=1000000 -disable_auto_compactions=true -compression_type=none -bloom_bits=3 ``` Pre-change `readrandom [AVG 60 runs] : 21568 (± 248) ops/sec` Post-change (no regression, -0.3%) `readrandom [AVG 60 runs] : 21486 (± 236) ops/sec` **Compaction/Flush**run till convergence ``` ./db_bench -db=/dev/shm/testdb2/ -seed=1678564177044286 -benchmarks="fillseq[-X60]" -key_size=32 -value_size=512 -num=50000 -write_buffer_size=655 -disable_auto_compactions=false -target_file_size_base=655 -compression_type=none rocksdb.sst.read.micros COUNT : 33820 rocksdb.sst.read.flush.micros COUNT : 1800 rocksdb.sst.read.compaction.micros COUNT : 32020 ``` Pre-change `fillseq [AVG 46 runs] : 1391 (± 214) ops/sec; 0.7 (± 0.1) MB/sec` Post-change (no regression, ~-0.4%) `fillseq [AVG 46 runs] : 1385 (± 216) ops/sec; 0.7 (± 0.1) MB/sec` Reviewed By: ajkr Differential Revision: D44007011 Pulled By: hx235 fbshipit-source-id: a54c89e4846dfc9a135389edf3f3eedfea257132
2023-04-21 16:07:18 +00:00
approx_opts, read_options, compaction_->input_version(),
tombstone_start.Encode(), tombstone_end.Encode(),
compaction_->output_level() + 1 /* start_level */,
-1 /* end_level */, kCompaction);
meta.compensated_range_deletion_size += approximate_covered_size;
}
}
Include estimated bytes deleted by range tombstones in compensated file size (#10734) Summary: compensate file sizes in compaction picking so files with range tombstones are preferred, such that they get compacted down earlier as they tend to delete a lot of data. This PR adds a `compensated_range_deletion_size` field in FileMeta that is computed during Flush/Compaction and persisted in MANIFEST. This value is added to `compensated_file_size` which will be used for compaction picking. Currently, for a file in level L, `compensated_range_deletion_size` is set to the estimated bytes deleted by range tombstone of this file in all levels > L. This helps to reduce space amp when data in older levels are covered by range tombstones in level L. Pull Request resolved: https://github.com/facebook/rocksdb/pull/10734 Test Plan: - Added unit tests. - benchmark to check if the above definition `compensated_range_deletion_size` is reducing space amp as intended, without affecting write amp too much. The experiment set up favorable for this optimization: large range tombstone issued infrequently. Command used: ``` ./db_bench -benchmarks=fillrandom,waitforcompaction,stats,levelstats -use_existing_db=false -avoid_flush_during_recovery=true -write_buffer_size=33554432 -level_compaction_dynamic_level_bytes=true -max_background_jobs=8 -max_bytes_for_level_base=134217728 -target_file_size_base=33554432 -writes_per_range_tombstone=500000 -range_tombstone_width=5000000 -num=50000000 -benchmark_write_rate_limit=8388608 -threads=16 -duration=1800 --max_num_range_tombstones=1000000000 ``` In this experiment, each thread wrote 16 range tombstones over the duration of 30 minutes, each range tombstone has width 5M that is the 10% of the key space width. Results shows this PR generates a smaller DB size. Compaction stats from this PR: ``` Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) CompMergeCPU(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Rblob(GB) Wblob(GB) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 2/0 31.54 MB 0.5 0.0 0.0 0.0 8.4 8.4 0.0 1.0 0.0 63.4 135.56 110.94 544 0.249 0 0 0.0 0.0 L4 3/0 96.55 MB 0.8 18.5 6.7 11.8 18.4 6.6 0.0 2.7 65.3 64.9 290.08 284.03 108 2.686 284M 1957K 0.0 0.0 L5 15/0 404.41 MB 1.0 19.1 7.7 11.4 18.8 7.4 0.3 2.5 66.6 65.7 292.93 285.34 220 1.332 293M 3808K 0.0 0.0 L6 143/0 4.12 GB 0.0 45.0 7.5 37.5 41.6 4.1 0.0 5.5 71.2 65.9 647.00 632.66 251 2.578 739M 47M 0.0 0.0 Sum 163/0 4.64 GB 0.0 82.6 21.9 60.7 87.2 26.5 0.3 10.4 61.9 65.4 1365.58 1312.97 1123 1.216 1318M 52M 0.0 0.0 ``` Compaction stats from main: ``` Level Files Size Score Read(GB) Rn(GB) Rnp1(GB) Write(GB) Wnew(GB) Moved(GB) W-Amp Rd(MB/s) Wr(MB/s) Comp(sec) CompMergeCPU(sec) Comp(cnt) Avg(sec) KeyIn KeyDrop Rblob(GB) Wblob(GB) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ L0 0/0 0.00 KB 0.0 0.0 0.0 0.0 8.4 8.4 0.0 1.0 0.0 60.5 142.12 115.89 569 0.250 0 0 0.0 0.0 L4 3/0 85.68 MB 1.0 17.7 6.8 10.9 17.6 6.7 0.0 2.6 62.7 62.3 289.05 281.79 112 2.581 272M 2309K 0.0 0.0 L5 11/0 293.73 MB 1.0 18.8 7.5 11.2 18.5 7.2 0.5 2.5 64.9 63.9 296.07 288.50 220 1.346 288M 4365K 0.0 0.0 L6 130/0 3.94 GB 0.0 51.5 7.6 43.9 47.9 3.9 0.0 6.3 67.2 62.4 784.95 765.92 258 3.042 848M 51M 0.0 0.0 Sum 144/0 4.31 GB 0.0 88.0 21.9 66.0 92.3 26.3 0.5 11.0 59.6 62.5 1512.19 1452.09 1159 1.305 1409M 58M 0.0 0.0``` Reviewed By: ajkr Differential Revision: D39834713 Pulled By: cbi42 fbshipit-source-id: fe9341040b8704a8fbb10cad5cf5c43e962c7e6b
2022-12-29 21:28:24 +00:00
}
}
return Status::OK();
}
void CompactionOutputs::FillFilesToCutForTtl() {
if (compaction_->immutable_options()->compaction_style !=
kCompactionStyleLevel ||
compaction_->immutable_options()->compaction_pri !=
kMinOverlappingRatio ||
compaction_->mutable_cf_options()->ttl == 0 ||
compaction_->num_input_levels() < 2 || compaction_->bottommost_level()) {
return;
}
// We define new file with the oldest ancestor time to be younger than 1/4
// TTL, and an old one to be older than 1/2 TTL time.
int64_t temp_current_time;
auto get_time_status =
compaction_->immutable_options()->clock->GetCurrentTime(
&temp_current_time);
if (!get_time_status.ok()) {
return;
}
auto current_time = static_cast<uint64_t>(temp_current_time);
if (current_time < compaction_->mutable_cf_options()->ttl) {
return;
}
uint64_t old_age_thres =
current_time - compaction_->mutable_cf_options()->ttl / 2;
const std::vector<FileMetaData*>& olevel =
*(compaction_->inputs(compaction_->num_input_levels() - 1));
for (FileMetaData* file : olevel) {
// Worth filtering out by start and end?
uint64_t oldest_ancester_time = file->TryGetOldestAncesterTime();
// We put old files if they are not too small to prevent a flood
// of small files.
if (oldest_ancester_time < old_age_thres &&
file->fd.GetFileSize() >
compaction_->mutable_cf_options()->target_file_size_base / 2) {
files_to_cut_for_ttl_.push_back(file);
}
}
}
CompactionOutputs::CompactionOutputs(const Compaction* compaction,
const bool is_penultimate_level)
: compaction_(compaction), is_penultimate_level_(is_penultimate_level) {
partitioner_ = compaction->output_level() == 0
? nullptr
: compaction->CreateSstPartitioner();
if (compaction->output_level() != 0) {
FillFilesToCutForTtl();
}
level_ptrs_ = std::vector<size_t>(compaction_->number_levels(), 0);
}
} // namespace ROCKSDB_NAMESPACE