2016-02-10 00:12:00 +01:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
2015-07-21 12:05:57 +02:00
|
|
|
// This source code is licensed under the BSD-style license found in the
|
|
|
|
// LICENSE file in the root directory of this source tree. An additional grant
|
|
|
|
// of patent rights can be found in the PATENTS file in the same directory.
|
|
|
|
//
|
|
|
|
// 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.
|
|
|
|
|
2015-10-12 22:05:42 +02:00
|
|
|
#include "db/db_test_util.h"
|
2015-07-21 12:05:57 +02:00
|
|
|
#include "port/stack_trace.h"
|
|
|
|
#include "rocksdb/experimental.h"
|
2015-12-29 22:22:13 +01:00
|
|
|
#include "rocksdb/utilities/convenience.h"
|
2015-07-21 12:05:57 +02:00
|
|
|
#include "util/sync_point.h"
|
|
|
|
namespace rocksdb {
|
|
|
|
|
2015-08-20 23:08:24 +02:00
|
|
|
// SYNC_POINT is not supported in released Windows mode.
|
2015-10-23 00:15:37 +02:00
|
|
|
#if !defined(ROCKSDB_LITE)
|
2015-08-20 23:08:24 +02:00
|
|
|
|
2015-07-21 12:05:57 +02:00
|
|
|
class DBCompactionTest : public DBTestBase {
|
|
|
|
public:
|
|
|
|
DBCompactionTest() : DBTestBase("/db_compaction_test") {}
|
|
|
|
};
|
|
|
|
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
class DBCompactionTestWithParam
|
|
|
|
: public DBTestBase,
|
|
|
|
public testing::WithParamInterface<std::tuple<uint32_t, bool>> {
|
2015-08-05 07:19:07 +02:00
|
|
|
public:
|
|
|
|
DBCompactionTestWithParam() : DBTestBase("/db_compaction_test") {
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
max_subcompactions_ = std::get<0>(GetParam());
|
|
|
|
exclusive_manual_compaction_ = std::get<1>(GetParam());
|
2015-08-05 07:19:07 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
// Required if inheriting from testing::WithParamInterface<>
|
|
|
|
static void SetUpTestCase() {}
|
|
|
|
static void TearDownTestCase() {}
|
|
|
|
|
2015-08-21 23:25:34 +02:00
|
|
|
uint32_t max_subcompactions_;
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
bool exclusive_manual_compaction_;
|
2015-08-05 07:19:07 +02:00
|
|
|
};
|
|
|
|
|
2015-07-21 12:05:57 +02:00
|
|
|
namespace {
|
|
|
|
|
2015-12-10 02:28:46 +01:00
|
|
|
class FlushedFileCollector : public EventListener {
|
|
|
|
public:
|
|
|
|
FlushedFileCollector() {}
|
|
|
|
~FlushedFileCollector() {}
|
|
|
|
|
|
|
|
virtual void OnFlushCompleted(DB* db, const FlushJobInfo& info) override {
|
|
|
|
std::lock_guard<std::mutex> lock(mutex_);
|
|
|
|
flushed_files_.push_back(info.file_path);
|
|
|
|
}
|
|
|
|
|
|
|
|
std::vector<std::string> GetFlushedFiles() {
|
|
|
|
std::lock_guard<std::mutex> lock(mutex_);
|
|
|
|
std::vector<std::string> result;
|
|
|
|
for (auto fname : flushed_files_) {
|
|
|
|
result.push_back(fname);
|
|
|
|
}
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
|
|
|
void ClearFlushedFiles() { flushed_files_.clear(); }
|
|
|
|
|
|
|
|
private:
|
|
|
|
std::vector<std::string> flushed_files_;
|
|
|
|
std::mutex mutex_;
|
|
|
|
};
|
|
|
|
|
2015-07-21 12:05:57 +02:00
|
|
|
static const int kCDTValueSize = 1000;
|
|
|
|
static const int kCDTKeysPerBuffer = 4;
|
|
|
|
static const int kCDTNumLevels = 8;
|
2016-04-08 02:40:42 +02:00
|
|
|
Options DeletionTriggerOptions(Options options) {
|
2015-07-21 12:05:57 +02:00
|
|
|
options.compression = kNoCompression;
|
|
|
|
options.write_buffer_size = kCDTKeysPerBuffer * (kCDTValueSize + 24);
|
|
|
|
options.min_write_buffer_number_to_merge = 1;
|
|
|
|
options.max_write_buffer_number_to_maintain = 0;
|
|
|
|
options.num_levels = kCDTNumLevels;
|
|
|
|
options.level0_file_num_compaction_trigger = 1;
|
|
|
|
options.target_file_size_base = options.write_buffer_size * 2;
|
|
|
|
options.target_file_size_multiplier = 2;
|
|
|
|
options.max_bytes_for_level_base =
|
|
|
|
options.target_file_size_base * options.target_file_size_multiplier;
|
|
|
|
options.max_bytes_for_level_multiplier = 2;
|
|
|
|
options.disable_auto_compactions = false;
|
|
|
|
return options;
|
|
|
|
}
|
|
|
|
|
|
|
|
bool HaveOverlappingKeyRanges(
|
|
|
|
const Comparator* c,
|
|
|
|
const SstFileMetaData& a, const SstFileMetaData& b) {
|
|
|
|
if (c->Compare(a.smallestkey, b.smallestkey) >= 0) {
|
|
|
|
if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
|
|
|
|
// b.smallestkey <= a.smallestkey <= b.largestkey
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
} else if (c->Compare(a.largestkey, b.smallestkey) >= 0) {
|
|
|
|
// a.smallestkey < b.smallestkey <= a.largestkey
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
if (c->Compare(a.largestkey, b.largestkey) <= 0) {
|
|
|
|
if (c->Compare(a.largestkey, b.smallestkey) >= 0) {
|
|
|
|
// b.smallestkey <= a.largestkey <= b.largestkey
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
} else if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
|
|
|
|
// a.smallestkey <= b.largestkey < a.largestkey
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Identifies all files between level "min_level" and "max_level"
|
|
|
|
// which has overlapping key range with "input_file_meta".
|
|
|
|
void GetOverlappingFileNumbersForLevelCompaction(
|
|
|
|
const ColumnFamilyMetaData& cf_meta,
|
|
|
|
const Comparator* comparator,
|
|
|
|
int min_level, int max_level,
|
|
|
|
const SstFileMetaData* input_file_meta,
|
|
|
|
std::set<std::string>* overlapping_file_names) {
|
|
|
|
std::set<const SstFileMetaData*> overlapping_files;
|
|
|
|
overlapping_files.insert(input_file_meta);
|
|
|
|
for (int m = min_level; m <= max_level; ++m) {
|
|
|
|
for (auto& file : cf_meta.levels[m].files) {
|
|
|
|
for (auto* included_file : overlapping_files) {
|
|
|
|
if (HaveOverlappingKeyRanges(
|
|
|
|
comparator, *included_file, file)) {
|
|
|
|
overlapping_files.insert(&file);
|
|
|
|
overlapping_file_names->insert(file.name);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void VerifyCompactionResult(
|
|
|
|
const ColumnFamilyMetaData& cf_meta,
|
|
|
|
const std::set<std::string>& overlapping_file_numbers) {
|
|
|
|
#ifndef NDEBUG
|
|
|
|
for (auto& level : cf_meta.levels) {
|
|
|
|
for (auto& file : level.files) {
|
|
|
|
assert(overlapping_file_numbers.find(file.name) ==
|
|
|
|
overlapping_file_numbers.end());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
}
|
|
|
|
|
|
|
|
const SstFileMetaData* PickFileRandomly(
|
|
|
|
const ColumnFamilyMetaData& cf_meta,
|
|
|
|
Random* rand,
|
|
|
|
int* level = nullptr) {
|
|
|
|
auto file_id = rand->Uniform(static_cast<int>(
|
|
|
|
cf_meta.file_count)) + 1;
|
|
|
|
for (auto& level_meta : cf_meta.levels) {
|
|
|
|
if (file_id <= level_meta.files.size()) {
|
|
|
|
if (level != nullptr) {
|
|
|
|
*level = level_meta.level;
|
|
|
|
}
|
|
|
|
auto result = rand->Uniform(file_id);
|
|
|
|
return &(level_meta.files[result]);
|
|
|
|
}
|
2015-11-15 19:49:14 +01:00
|
|
|
file_id -= static_cast<uint32_t>(level_meta.files.size());
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
assert(false);
|
|
|
|
return nullptr;
|
|
|
|
}
|
|
|
|
} // anonymous namespace
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
// All the TEST_P tests run once with sub_compactions disabled (i.e.
|
2015-08-21 23:25:34 +02:00
|
|
|
// options.max_subcompactions = 1) and once with it enabled
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, CompactionDeletionTrigger) {
|
2015-08-04 22:48:16 +02:00
|
|
|
for (int tid = 0; tid < 3; ++tid) {
|
2015-07-21 12:05:57 +02:00
|
|
|
uint64_t db_size[2];
|
2016-04-08 02:40:42 +02:00
|
|
|
Options options = DeletionTriggerOptions(CurrentOptions());
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
|
|
|
|
if (tid == 1) {
|
2015-08-04 22:48:16 +02:00
|
|
|
// the following only disable stats update in DB::Open()
|
|
|
|
// and should not affect the result of this test.
|
|
|
|
options.skip_stats_update_on_db_open = true;
|
|
|
|
} else if (tid == 2) {
|
|
|
|
// third pass with universal compaction
|
2015-07-21 12:05:57 +02:00
|
|
|
options.compaction_style = kCompactionStyleUniversal;
|
|
|
|
options.num_levels = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
Random rnd(301);
|
|
|
|
|
2015-09-01 23:43:23 +02:00
|
|
|
const int kTestSize = kCDTKeysPerBuffer * 1024;
|
2015-07-21 12:05:57 +02:00
|
|
|
std::vector<std::string> values;
|
|
|
|
for (int k = 0; k < kTestSize; ++k) {
|
|
|
|
values.push_back(RandomString(&rnd, kCDTValueSize));
|
|
|
|
ASSERT_OK(Put(Key(k), values[k]));
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
db_size[0] = Size(Key(0), Key(kTestSize - 1));
|
|
|
|
|
|
|
|
for (int k = 0; k < kTestSize; ++k) {
|
|
|
|
ASSERT_OK(Delete(Key(k)));
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
db_size[1] = Size(Key(0), Key(kTestSize - 1));
|
|
|
|
|
|
|
|
// must have much smaller db size.
|
|
|
|
ASSERT_GT(db_size[0] / 3, db_size[1]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-08-04 22:48:16 +02:00
|
|
|
TEST_F(DBCompactionTest, SkipStatsUpdateTest) {
|
2016-06-14 00:00:02 +02:00
|
|
|
// This test verify UpdateAccumulatedStats is not on
|
|
|
|
// if options.skip_stats_update_on_db_open = true
|
2015-08-04 22:48:16 +02:00
|
|
|
// The test will need to be updated if the internal behavior changes.
|
|
|
|
|
2016-04-08 02:40:42 +02:00
|
|
|
Options options = DeletionTriggerOptions(CurrentOptions());
|
2015-08-04 22:48:16 +02:00
|
|
|
options.env = env_;
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
Random rnd(301);
|
|
|
|
|
|
|
|
const int kTestSize = kCDTKeysPerBuffer * 512;
|
|
|
|
std::vector<std::string> values;
|
|
|
|
for (int k = 0; k < kTestSize; ++k) {
|
|
|
|
values.push_back(RandomString(&rnd, kCDTValueSize));
|
|
|
|
ASSERT_OK(Put(Key(k), values[k]));
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
// Reopen the DB with stats-update disabled
|
|
|
|
options.skip_stats_update_on_db_open = true;
|
|
|
|
env_->random_file_open_counter_.store(0);
|
|
|
|
Reopen(options);
|
2015-08-05 00:47:05 +02:00
|
|
|
|
2015-11-12 16:43:28 +01:00
|
|
|
// As stats-update is disabled, we expect a very low number of
|
|
|
|
// random file open.
|
|
|
|
// Note that this number must be changed accordingly if we change
|
|
|
|
// the number of files needed to be opened in the DB::Open process.
|
|
|
|
const int kMaxFileOpenCount = 10;
|
|
|
|
ASSERT_LT(env_->random_file_open_counter_.load(), kMaxFileOpenCount);
|
2015-08-04 22:48:16 +02:00
|
|
|
|
|
|
|
// Repeat the reopen process, but this time we enable
|
|
|
|
// stats-update.
|
|
|
|
options.skip_stats_update_on_db_open = false;
|
|
|
|
env_->random_file_open_counter_.store(0);
|
|
|
|
Reopen(options);
|
2015-08-05 00:47:05 +02:00
|
|
|
|
2015-08-04 22:48:16 +02:00
|
|
|
// Since we do a normal stats update on db-open, there
|
|
|
|
// will be more random open files.
|
2015-11-12 16:43:28 +01:00
|
|
|
ASSERT_GT(env_->random_file_open_counter_.load(), kMaxFileOpenCount);
|
2015-08-04 22:48:16 +02:00
|
|
|
}
|
|
|
|
|
2015-08-21 07:33:25 +02:00
|
|
|
TEST_F(DBCompactionTest, TestTableReaderForCompaction) {
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-08-21 07:33:25 +02:00
|
|
|
options.env = env_;
|
|
|
|
options.new_table_reader_for_compaction_inputs = true;
|
|
|
|
options.max_open_files = 100;
|
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
Random rnd(301);
|
|
|
|
|
|
|
|
int num_table_cache_lookup = 0;
|
|
|
|
int num_new_table_reader = 0;
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"TableCache::FindTable:0", [&](void* arg) {
|
|
|
|
assert(arg != nullptr);
|
|
|
|
bool no_io = *(reinterpret_cast<bool*>(arg));
|
|
|
|
if (!no_io) {
|
|
|
|
// filter out cases for table properties queries.
|
|
|
|
num_table_cache_lookup++;
|
|
|
|
}
|
|
|
|
});
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"TableCache::GetTableReader:0",
|
|
|
|
[&](void* arg) { num_new_table_reader++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
|
|
|
for (int k = 0; k < options.level0_file_num_compaction_trigger; ++k) {
|
|
|
|
ASSERT_OK(Put(Key(k), Key(k)));
|
|
|
|
ASSERT_OK(Put(Key(10 - k), "bar"));
|
|
|
|
if (k < options.level0_file_num_compaction_trigger - 1) {
|
|
|
|
num_table_cache_lookup = 0;
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
// preloading iterator issues one table cache lookup and create
|
|
|
|
// a new table reader.
|
|
|
|
ASSERT_EQ(num_table_cache_lookup, 1);
|
|
|
|
ASSERT_EQ(num_new_table_reader, 1);
|
|
|
|
|
|
|
|
num_table_cache_lookup = 0;
|
|
|
|
num_new_table_reader = 0;
|
|
|
|
ASSERT_EQ(Key(k), Get(Key(k)));
|
|
|
|
// lookup iterator from table cache and no need to create a new one.
|
2016-11-22 06:08:06 +01:00
|
|
|
ASSERT_EQ(num_table_cache_lookup, 1);
|
2015-08-21 07:33:25 +02:00
|
|
|
ASSERT_EQ(num_new_table_reader, 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
num_table_cache_lookup = 0;
|
|
|
|
num_new_table_reader = 0;
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
// Preloading iterator issues one table cache lookup and creates
|
|
|
|
// a new table reader. One file is created for flush and one for compaction.
|
2016-11-22 06:08:06 +01:00
|
|
|
// Compaction inputs make no table cache look-up for data/range deletion
|
|
|
|
// iterators
|
|
|
|
ASSERT_EQ(num_table_cache_lookup, 2);
|
2015-08-21 07:33:25 +02:00
|
|
|
// Create new iterator for:
|
|
|
|
// (1) 1 for verifying flush results
|
|
|
|
// (2) 3 for compaction input files
|
|
|
|
// (3) 1 for verifying compaction results.
|
|
|
|
ASSERT_EQ(num_new_table_reader, 5);
|
|
|
|
|
|
|
|
num_table_cache_lookup = 0;
|
|
|
|
num_new_table_reader = 0;
|
|
|
|
ASSERT_EQ(Key(1), Get(Key(1)));
|
2016-11-22 06:08:06 +01:00
|
|
|
ASSERT_EQ(num_table_cache_lookup, 1);
|
2015-08-21 07:33:25 +02:00
|
|
|
ASSERT_EQ(num_new_table_reader, 0);
|
|
|
|
|
|
|
|
num_table_cache_lookup = 0;
|
|
|
|
num_new_table_reader = 0;
|
|
|
|
CompactRangeOptions cro;
|
|
|
|
cro.change_level = true;
|
|
|
|
cro.target_level = 2;
|
|
|
|
cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
|
|
|
|
db_->CompactRange(cro, nullptr, nullptr);
|
2016-11-22 06:08:06 +01:00
|
|
|
// Only verifying compaction outputs issues one table cache lookup
|
|
|
|
// for both data block and range deletion block).
|
|
|
|
ASSERT_EQ(num_table_cache_lookup, 1);
|
2015-08-21 07:33:25 +02:00
|
|
|
// One for compaction input, one for verifying compaction results.
|
|
|
|
ASSERT_EQ(num_new_table_reader, 2);
|
|
|
|
|
|
|
|
num_table_cache_lookup = 0;
|
|
|
|
num_new_table_reader = 0;
|
|
|
|
ASSERT_EQ(Key(1), Get(Key(1)));
|
2016-11-22 06:08:06 +01:00
|
|
|
ASSERT_EQ(num_table_cache_lookup, 1);
|
2015-08-21 07:33:25 +02:00
|
|
|
ASSERT_EQ(num_new_table_reader, 0);
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->ClearAllCallBacks();
|
|
|
|
}
|
2015-08-05 07:19:07 +02:00
|
|
|
|
|
|
|
TEST_P(DBCompactionTestWithParam, CompactionDeletionTriggerReopen) {
|
2015-07-21 12:05:57 +02:00
|
|
|
for (int tid = 0; tid < 2; ++tid) {
|
|
|
|
uint64_t db_size[3];
|
2016-04-08 02:40:42 +02:00
|
|
|
Options options = DeletionTriggerOptions(CurrentOptions());
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
|
|
|
|
if (tid == 1) {
|
|
|
|
// second pass with universal compaction
|
|
|
|
options.compaction_style = kCompactionStyleUniversal;
|
|
|
|
options.num_levels = 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
Random rnd(301);
|
|
|
|
|
|
|
|
// round 1 --- insert key/value pairs.
|
|
|
|
const int kTestSize = kCDTKeysPerBuffer * 512;
|
|
|
|
std::vector<std::string> values;
|
|
|
|
for (int k = 0; k < kTestSize; ++k) {
|
|
|
|
values.push_back(RandomString(&rnd, kCDTValueSize));
|
|
|
|
ASSERT_OK(Put(Key(k), values[k]));
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
db_size[0] = Size(Key(0), Key(kTestSize - 1));
|
|
|
|
Close();
|
|
|
|
|
|
|
|
// round 2 --- disable auto-compactions and issue deletions.
|
|
|
|
options.create_if_missing = false;
|
|
|
|
options.disable_auto_compactions = true;
|
|
|
|
Reopen(options);
|
|
|
|
|
|
|
|
for (int k = 0; k < kTestSize; ++k) {
|
|
|
|
ASSERT_OK(Delete(Key(k)));
|
|
|
|
}
|
|
|
|
db_size[1] = Size(Key(0), Key(kTestSize - 1));
|
|
|
|
Close();
|
|
|
|
// as auto_compaction is off, we shouldn't see too much reduce
|
|
|
|
// in db size.
|
|
|
|
ASSERT_LT(db_size[0] / 3, db_size[1]);
|
|
|
|
|
|
|
|
// round 3 --- reopen db with auto_compaction on and see if
|
|
|
|
// deletion compensation still work.
|
|
|
|
options.disable_auto_compactions = false;
|
|
|
|
Reopen(options);
|
|
|
|
// insert relatively small amount of data to trigger auto compaction.
|
|
|
|
for (int k = 0; k < kTestSize / 10; ++k) {
|
|
|
|
ASSERT_OK(Put(Key(k), values[k]));
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
db_size[2] = Size(Key(0), Key(kTestSize - 1));
|
|
|
|
// this time we're expecting significant drop in size.
|
|
|
|
ASSERT_GT(db_size[0] / 3, db_size[2]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-08-04 22:48:16 +02:00
|
|
|
TEST_F(DBCompactionTest, DisableStatsUpdateReopen) {
|
|
|
|
uint64_t db_size[3];
|
|
|
|
for (int test = 0; test < 2; ++test) {
|
2016-04-08 02:40:42 +02:00
|
|
|
Options options = DeletionTriggerOptions(CurrentOptions());
|
2015-08-04 22:48:16 +02:00
|
|
|
options.skip_stats_update_on_db_open = (test == 0);
|
|
|
|
|
|
|
|
env_->random_read_counter_.Reset();
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
Random rnd(301);
|
|
|
|
|
|
|
|
// round 1 --- insert key/value pairs.
|
|
|
|
const int kTestSize = kCDTKeysPerBuffer * 512;
|
|
|
|
std::vector<std::string> values;
|
|
|
|
for (int k = 0; k < kTestSize; ++k) {
|
|
|
|
values.push_back(RandomString(&rnd, kCDTValueSize));
|
|
|
|
ASSERT_OK(Put(Key(k), values[k]));
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
db_size[0] = Size(Key(0), Key(kTestSize - 1));
|
|
|
|
Close();
|
|
|
|
|
|
|
|
// round 2 --- disable auto-compactions and issue deletions.
|
|
|
|
options.create_if_missing = false;
|
|
|
|
options.disable_auto_compactions = true;
|
|
|
|
|
|
|
|
env_->random_read_counter_.Reset();
|
|
|
|
Reopen(options);
|
|
|
|
|
|
|
|
for (int k = 0; k < kTestSize; ++k) {
|
|
|
|
ASSERT_OK(Delete(Key(k)));
|
|
|
|
}
|
|
|
|
db_size[1] = Size(Key(0), Key(kTestSize - 1));
|
|
|
|
Close();
|
|
|
|
// as auto_compaction is off, we shouldn't see too much reduce
|
|
|
|
// in db size.
|
|
|
|
ASSERT_LT(db_size[0] / 3, db_size[1]);
|
|
|
|
|
|
|
|
// round 3 --- reopen db with auto_compaction on and see if
|
|
|
|
// deletion compensation still work.
|
|
|
|
options.disable_auto_compactions = false;
|
|
|
|
Reopen(options);
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
db_size[2] = Size(Key(0), Key(kTestSize - 1));
|
|
|
|
|
|
|
|
if (options.skip_stats_update_on_db_open) {
|
|
|
|
// If update stats on DB::Open is disable, we don't expect
|
|
|
|
// deletion entries taking effect.
|
|
|
|
ASSERT_LT(db_size[0] / 3, db_size[2]);
|
|
|
|
} else {
|
|
|
|
// Otherwise, we should see a significant drop in db size.
|
|
|
|
ASSERT_GT(db_size[0] / 3, db_size[2]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
|
|
|
|
TEST_P(DBCompactionTestWithParam, CompactionTrigger) {
|
2015-12-02 23:06:33 +01:00
|
|
|
const int kNumKeysPerFile = 100;
|
2015-12-01 07:16:49 +01:00
|
|
|
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-12-02 23:06:33 +01:00
|
|
|
options.write_buffer_size = 110 << 10; // 110KB
|
2015-08-26 23:19:31 +02:00
|
|
|
options.arena_block_size = 4 << 10;
|
2015-07-21 12:05:57 +02:00
|
|
|
options.num_levels = 3;
|
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-12-01 07:16:49 +01:00
|
|
|
options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
|
2015-07-21 12:05:57 +02:00
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
|
|
|
|
for (int num = 0; num < options.level0_file_num_compaction_trigger - 1;
|
|
|
|
num++) {
|
|
|
|
std::vector<std::string> values;
|
2015-08-26 23:19:31 +02:00
|
|
|
// Write 100KB (100 values, each 1K)
|
2015-12-01 07:16:49 +01:00
|
|
|
for (int i = 0; i < kNumKeysPerFile; i++) {
|
2015-08-26 23:19:31 +02:00
|
|
|
values.push_back(RandomString(&rnd, 990));
|
2015-07-21 12:05:57 +02:00
|
|
|
ASSERT_OK(Put(1, Key(i), values[i]));
|
|
|
|
}
|
2015-12-02 23:06:33 +01:00
|
|
|
// put extra key to trigger flush
|
2015-12-01 07:16:49 +01:00
|
|
|
ASSERT_OK(Put(1, "", ""));
|
2015-07-21 12:05:57 +02:00
|
|
|
dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0, 1), num + 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
// generate one more file in level-0, and should trigger level-0 compaction
|
|
|
|
std::vector<std::string> values;
|
2015-12-02 23:06:33 +01:00
|
|
|
for (int i = 0; i < kNumKeysPerFile; i++) {
|
2015-08-26 23:19:31 +02:00
|
|
|
values.push_back(RandomString(&rnd, 990));
|
2015-07-21 12:05:57 +02:00
|
|
|
ASSERT_OK(Put(1, Key(i), values[i]));
|
|
|
|
}
|
2015-12-02 23:06:33 +01:00
|
|
|
// put extra key to trigger flush
|
|
|
|
ASSERT_OK(Put(1, "", ""));
|
2015-07-21 12:05:57 +02:00
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(1, 1), 1);
|
|
|
|
}
|
|
|
|
|
Add options.base_background_compactions as a number of compaction threads for low compaction debt
Summary:
If options.base_background_compactions is given, we try to schedule number of compactions not existing this number, only when L0 files increase to certain number, or pending compaction bytes more than certain threshold, we schedule compactions based on options.max_background_compactions.
The watermarks are calculated based on slowdown thresholds.
Test Plan:
Add new test cases in column_family_test.
Adding more unit tests.
Reviewers: IslamAbdelRahman, yhchiang, kradhakrishnan, rven, anthony
Reviewed By: anthony
Subscribers: leveldb, dhruba, yoshinorim
Differential Revision: https://reviews.facebook.net/D53409
2016-01-28 20:56:16 +01:00
|
|
|
TEST_F(DBCompactionTest, BGCompactionsAllowed) {
|
|
|
|
// Create several column families. Make compaction triggers in all of them
|
|
|
|
// and see number of compactions scheduled to be less than allowed.
|
|
|
|
const int kNumKeysPerFile = 100;
|
|
|
|
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
Add options.base_background_compactions as a number of compaction threads for low compaction debt
Summary:
If options.base_background_compactions is given, we try to schedule number of compactions not existing this number, only when L0 files increase to certain number, or pending compaction bytes more than certain threshold, we schedule compactions based on options.max_background_compactions.
The watermarks are calculated based on slowdown thresholds.
Test Plan:
Add new test cases in column_family_test.
Adding more unit tests.
Reviewers: IslamAbdelRahman, yhchiang, kradhakrishnan, rven, anthony
Reviewed By: anthony
Subscribers: leveldb, dhruba, yoshinorim
Differential Revision: https://reviews.facebook.net/D53409
2016-01-28 20:56:16 +01:00
|
|
|
options.write_buffer_size = 110 << 10; // 110KB
|
|
|
|
options.arena_block_size = 4 << 10;
|
|
|
|
options.num_levels = 3;
|
|
|
|
// Should speed up compaction when there are 4 files.
|
|
|
|
options.level0_file_num_compaction_trigger = 2;
|
|
|
|
options.level0_slowdown_writes_trigger = 20;
|
|
|
|
options.soft_pending_compaction_bytes_limit = 1 << 30; // Infinitely large
|
|
|
|
options.base_background_compactions = 1;
|
|
|
|
options.max_background_compactions = 3;
|
|
|
|
options.memtable_factory.reset(new SpecialSkipListFactory(kNumKeysPerFile));
|
|
|
|
|
|
|
|
// Block all threads in thread pool.
|
|
|
|
const size_t kTotalTasks = 4;
|
|
|
|
env_->SetBackgroundThreads(4, Env::LOW);
|
|
|
|
test::SleepingBackgroundTask sleeping_tasks[kTotalTasks];
|
|
|
|
for (size_t i = 0; i < kTotalTasks; i++) {
|
|
|
|
env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
|
|
|
|
&sleeping_tasks[i], Env::Priority::LOW);
|
|
|
|
sleeping_tasks[i].WaitUntilSleeping();
|
|
|
|
}
|
|
|
|
|
|
|
|
CreateAndReopenWithCF({"one", "two", "three"}, options);
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
for (int cf = 0; cf < 4; cf++) {
|
|
|
|
for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
|
|
|
|
for (int i = 0; i < kNumKeysPerFile; i++) {
|
|
|
|
ASSERT_OK(Put(cf, Key(i), ""));
|
|
|
|
}
|
|
|
|
// put extra key to trigger flush
|
|
|
|
ASSERT_OK(Put(cf, "", ""));
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable(handles_[cf]);
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0, cf), num + 1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Now all column families qualify compaction but only one should be
|
|
|
|
// scheduled, because no column family hits speed up condition.
|
|
|
|
ASSERT_EQ(1, env_->GetThreadPoolQueueLen(Env::Priority::LOW));
|
|
|
|
|
|
|
|
// Create two more files for one column family, which triggers speed up
|
|
|
|
// condition, three compactions will be scheduled.
|
|
|
|
for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
|
|
|
|
for (int i = 0; i < kNumKeysPerFile; i++) {
|
|
|
|
ASSERT_OK(Put(2, Key(i), ""));
|
|
|
|
}
|
|
|
|
// put extra key to trigger flush
|
|
|
|
ASSERT_OK(Put(2, "", ""));
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable(handles_[2]);
|
|
|
|
ASSERT_EQ(options.level0_file_num_compaction_trigger + num + 1,
|
|
|
|
NumTableFilesAtLevel(0, 2));
|
|
|
|
}
|
|
|
|
ASSERT_EQ(3, env_->GetThreadPoolQueueLen(Env::Priority::LOW));
|
|
|
|
|
|
|
|
// Unblock all threads to unblock all compactions.
|
|
|
|
for (size_t i = 0; i < kTotalTasks; i++) {
|
|
|
|
sleeping_tasks[i].WakeUp();
|
|
|
|
sleeping_tasks[i].WaitUntilDone();
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
// Verify number of compactions allowed will come back to 1.
|
|
|
|
|
|
|
|
for (size_t i = 0; i < kTotalTasks; i++) {
|
|
|
|
sleeping_tasks[i].Reset();
|
|
|
|
env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask,
|
|
|
|
&sleeping_tasks[i], Env::Priority::LOW);
|
|
|
|
sleeping_tasks[i].WaitUntilSleeping();
|
|
|
|
}
|
|
|
|
for (int cf = 0; cf < 4; cf++) {
|
|
|
|
for (int num = 0; num < options.level0_file_num_compaction_trigger; num++) {
|
|
|
|
for (int i = 0; i < kNumKeysPerFile; i++) {
|
|
|
|
ASSERT_OK(Put(cf, Key(i), ""));
|
|
|
|
}
|
|
|
|
// put extra key to trigger flush
|
|
|
|
ASSERT_OK(Put(cf, "", ""));
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable(handles_[cf]);
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0, cf), num + 1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Now all column families qualify compaction but only one should be
|
|
|
|
// scheduled, because no column family hits speed up condition.
|
|
|
|
ASSERT_EQ(1, env_->GetThreadPoolQueueLen(Env::Priority::LOW));
|
|
|
|
|
|
|
|
for (size_t i = 0; i < kTotalTasks; i++) {
|
|
|
|
sleeping_tasks[i].WakeUp();
|
|
|
|
sleeping_tasks[i].WaitUntilDone();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, CompactionsGenerateMultipleFiles) {
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-07-21 12:05:57 +02:00
|
|
|
options.write_buffer_size = 100000000; // Large write buffer
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
|
|
|
|
// Write 8MB (80 values, each 100K)
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
|
|
|
|
std::vector<std::string> values;
|
|
|
|
for (int i = 0; i < 80; i++) {
|
|
|
|
values.push_back(RandomString(&rnd, 100000));
|
|
|
|
ASSERT_OK(Put(1, Key(i), values[i]));
|
|
|
|
}
|
|
|
|
|
|
|
|
// Reopening moves updates to level-0
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, options);
|
|
|
|
dbfull()->TEST_CompactRange(0, nullptr, nullptr, handles_[1],
|
|
|
|
true /* disallow trivial move */);
|
|
|
|
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
|
|
|
|
ASSERT_GT(NumTableFilesAtLevel(1, 1), 1);
|
|
|
|
for (int i = 0; i < 80; i++) {
|
|
|
|
ASSERT_EQ(Get(1, Key(i)), values[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST_F(DBCompactionTest, MinorCompactionsHappen) {
|
|
|
|
do {
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-07-21 12:05:57 +02:00
|
|
|
options.write_buffer_size = 10000;
|
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
|
|
|
|
const int N = 500;
|
|
|
|
|
|
|
|
int starting_num_tables = TotalTableFiles(1);
|
|
|
|
for (int i = 0; i < N; i++) {
|
|
|
|
ASSERT_OK(Put(1, Key(i), Key(i) + std::string(1000, 'v')));
|
|
|
|
}
|
|
|
|
int ending_num_tables = TotalTableFiles(1);
|
|
|
|
ASSERT_GT(ending_num_tables, starting_num_tables);
|
|
|
|
|
|
|
|
for (int i = 0; i < N; i++) {
|
|
|
|
ASSERT_EQ(Key(i) + std::string(1000, 'v'), Get(1, Key(i)));
|
|
|
|
}
|
|
|
|
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, options);
|
|
|
|
|
|
|
|
for (int i = 0; i < N; i++) {
|
|
|
|
ASSERT_EQ(Key(i) + std::string(1000, 'v'), Get(1, Key(i)));
|
|
|
|
}
|
|
|
|
} while (ChangeCompactOptions());
|
|
|
|
}
|
|
|
|
|
2016-07-26 03:08:49 +02:00
|
|
|
TEST_F(DBCompactionTest, UserKeyCrossFile1) {
|
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.compaction_style = kCompactionStyleLevel;
|
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
// create first file and flush to l0
|
|
|
|
Put("4", "A");
|
|
|
|
Put("3", "A");
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
|
|
|
|
Put("2", "A");
|
|
|
|
Delete("3");
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
ASSERT_EQ("NOT_FOUND", Get("3"));
|
|
|
|
|
|
|
|
// move both files down to l1
|
|
|
|
dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
|
|
|
|
ASSERT_EQ("NOT_FOUND", Get("3"));
|
|
|
|
|
|
|
|
for (int i = 0; i < 3; i++) {
|
|
|
|
Put("2", "B");
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
ASSERT_EQ("NOT_FOUND", Get("3"));
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST_F(DBCompactionTest, UserKeyCrossFile2) {
|
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.compaction_style = kCompactionStyleLevel;
|
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
// create first file and flush to l0
|
|
|
|
Put("4", "A");
|
|
|
|
Put("3", "A");
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
|
|
|
|
Put("2", "A");
|
|
|
|
SingleDelete("3");
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
ASSERT_EQ("NOT_FOUND", Get("3"));
|
|
|
|
|
|
|
|
// move both files down to l1
|
|
|
|
dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);
|
|
|
|
ASSERT_EQ("NOT_FOUND", Get("3"));
|
|
|
|
|
|
|
|
for (int i = 0; i < 3; i++) {
|
|
|
|
Put("2", "B");
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
ASSERT_EQ("NOT_FOUND", Get("3"));
|
|
|
|
}
|
|
|
|
|
2015-12-10 02:28:46 +01:00
|
|
|
TEST_F(DBCompactionTest, ZeroSeqIdCompaction) {
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-12-10 02:28:46 +01:00
|
|
|
options.compaction_style = kCompactionStyleLevel;
|
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
|
|
|
|
|
|
|
FlushedFileCollector* collector = new FlushedFileCollector();
|
|
|
|
options.listeners.emplace_back(collector);
|
|
|
|
|
|
|
|
// compaction options
|
|
|
|
CompactionOptions compact_opt;
|
|
|
|
compact_opt.compression = kNoCompression;
|
|
|
|
compact_opt.output_file_size_limit = 4096;
|
2015-12-21 16:55:15 +01:00
|
|
|
const size_t key_len =
|
|
|
|
static_cast<size_t>(compact_opt.output_file_size_limit) / 5;
|
2015-12-10 02:28:46 +01:00
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
std::vector<const Snapshot*> snaps;
|
|
|
|
|
|
|
|
// create first file and flush to l0
|
|
|
|
for (auto& key : {"1", "2", "3", "3", "3", "3"}) {
|
|
|
|
Put(key, std::string(key_len, 'A'));
|
|
|
|
snaps.push_back(dbfull()->GetSnapshot());
|
|
|
|
}
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
|
|
|
|
// create second file and flush to l0
|
|
|
|
for (auto& key : {"3", "4", "5", "6", "7", "8"}) {
|
|
|
|
Put(key, std::string(key_len, 'A'));
|
|
|
|
snaps.push_back(dbfull()->GetSnapshot());
|
|
|
|
}
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
|
|
|
|
// move both files down to l1
|
|
|
|
dbfull()->CompactFiles(compact_opt, collector->GetFlushedFiles(), 1);
|
|
|
|
|
|
|
|
// release snap so that first instance of key(3) can have seqId=0
|
|
|
|
for (auto snap : snaps) {
|
|
|
|
dbfull()->ReleaseSnapshot(snap);
|
|
|
|
}
|
|
|
|
|
|
|
|
// create 3 files in l0 so to trigger compaction
|
|
|
|
for (int i = 0; i < options.level0_file_num_compaction_trigger; i++) {
|
|
|
|
Put("2", std::string(1, 'A'));
|
|
|
|
Flush();
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
}
|
|
|
|
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
ASSERT_OK(Put("", ""));
|
|
|
|
}
|
|
|
|
|
2015-07-21 12:05:57 +02:00
|
|
|
// Check that writes done during a memtable compaction are recovered
|
|
|
|
// if the database is shutdown during the memtable compaction.
|
|
|
|
TEST_F(DBCompactionTest, RecoverDuringMemtableCompaction) {
|
|
|
|
do {
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-07-21 12:05:57 +02:00
|
|
|
options.env = env_;
|
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
|
|
|
|
// Trigger a long memtable compaction and reopen the database during it
|
|
|
|
ASSERT_OK(Put(1, "foo", "v1")); // Goes to 1st log file
|
|
|
|
ASSERT_OK(Put(1, "big1", std::string(10000000, 'x'))); // Fills memtable
|
|
|
|
ASSERT_OK(Put(1, "big2", std::string(1000, 'y'))); // Triggers compaction
|
|
|
|
ASSERT_OK(Put(1, "bar", "v2")); // Goes to new log file
|
|
|
|
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, options);
|
|
|
|
ASSERT_EQ("v1", Get(1, "foo"));
|
|
|
|
ASSERT_EQ("v2", Get(1, "bar"));
|
|
|
|
ASSERT_EQ(std::string(10000000, 'x'), Get(1, "big1"));
|
|
|
|
ASSERT_EQ(std::string(1000, 'y'), Get(1, "big2"));
|
|
|
|
} while (ChangeOptions());
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, TrivialMoveOneFile) {
|
2015-07-21 12:05:57 +02:00
|
|
|
int32_t trivial_move = 0;
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:TrivialMove",
|
|
|
|
[&](void* arg) { trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-07-21 12:05:57 +02:00
|
|
|
options.write_buffer_size = 100000000;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
int32_t num_keys = 80;
|
|
|
|
int32_t value_size = 100 * 1024; // 100 KB
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
std::vector<std::string> values;
|
|
|
|
for (int i = 0; i < num_keys; i++) {
|
|
|
|
values.push_back(RandomString(&rnd, value_size));
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
|
|
|
|
// Reopening moves updates to L0
|
|
|
|
Reopen(options);
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0, 0), 1); // 1 file in L0
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(1, 0), 0); // 0 files in L1
|
|
|
|
|
|
|
|
std::vector<LiveFileMetaData> metadata;
|
|
|
|
db_->GetLiveFilesMetaData(&metadata);
|
|
|
|
ASSERT_EQ(metadata.size(), 1U);
|
|
|
|
LiveFileMetaData level0_file = metadata[0]; // L0 file meta
|
|
|
|
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
CompactRangeOptions cro;
|
|
|
|
cro.exclusive_manual_compaction = exclusive_manual_compaction_;
|
|
|
|
|
2015-07-21 12:05:57 +02:00
|
|
|
// Compaction will initiate a trivial move from L0 to L1
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
dbfull()->CompactRange(cro, nullptr, nullptr);
|
2015-07-21 12:05:57 +02:00
|
|
|
|
|
|
|
// File moved From L0 to L1
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0, 0), 0); // 0 files in L0
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(1, 0), 1); // 1 file in L1
|
|
|
|
|
|
|
|
metadata.clear();
|
|
|
|
db_->GetLiveFilesMetaData(&metadata);
|
|
|
|
ASSERT_EQ(metadata.size(), 1U);
|
|
|
|
ASSERT_EQ(metadata[0].name /* level1_file.name */, level0_file.name);
|
|
|
|
ASSERT_EQ(metadata[0].size /* level1_file.size */, level0_file.size);
|
|
|
|
|
|
|
|
for (int i = 0; i < num_keys; i++) {
|
|
|
|
ASSERT_EQ(Get(Key(i)), values[i]);
|
|
|
|
}
|
|
|
|
|
|
|
|
ASSERT_EQ(trivial_move, 1);
|
|
|
|
rocksdb::SyncPoint::GetInstance()->DisableProcessing();
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, TrivialMoveNonOverlappingFiles) {
|
2015-07-21 12:05:57 +02:00
|
|
|
int32_t trivial_move = 0;
|
|
|
|
int32_t non_trivial_move = 0;
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:TrivialMove",
|
|
|
|
[&](void* arg) { trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:NonTrivial",
|
|
|
|
[&](void* arg) { non_trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.disable_auto_compactions = true;
|
|
|
|
options.write_buffer_size = 10 * 1024 * 1024;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
// non overlapping ranges
|
|
|
|
std::vector<std::pair<int32_t, int32_t>> ranges = {
|
|
|
|
{100, 199},
|
|
|
|
{300, 399},
|
|
|
|
{0, 99},
|
|
|
|
{200, 299},
|
|
|
|
{600, 699},
|
|
|
|
{400, 499},
|
|
|
|
{500, 550},
|
|
|
|
{551, 599},
|
|
|
|
};
|
|
|
|
int32_t value_size = 10 * 1024; // 10 KB
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
std::map<int32_t, std::string> values;
|
2015-12-16 00:26:20 +01:00
|
|
|
for (size_t i = 0; i < ranges.size(); i++) {
|
2015-07-21 12:05:57 +02:00
|
|
|
for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
|
|
|
|
values[j] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(j), values[j]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
}
|
|
|
|
|
|
|
|
int32_t level0_files = NumTableFilesAtLevel(0, 0);
|
|
|
|
ASSERT_EQ(level0_files, ranges.size()); // Multiple files in L0
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(1, 0), 0); // No files in L1
|
|
|
|
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
CompactRangeOptions cro;
|
|
|
|
cro.exclusive_manual_compaction = exclusive_manual_compaction_;
|
|
|
|
|
2015-07-21 12:05:57 +02:00
|
|
|
// Since data is non-overlapping we expect compaction to initiate
|
|
|
|
// a trivial move
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
db_->CompactRange(cro, nullptr, nullptr);
|
2015-07-21 12:05:57 +02:00
|
|
|
// We expect that all the files were trivially moved from L0 to L1
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0, 0), 0);
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(1, 0) /* level1_files */, level0_files);
|
|
|
|
|
2015-12-16 00:26:20 +01:00
|
|
|
for (size_t i = 0; i < ranges.size(); i++) {
|
2015-07-21 12:05:57 +02:00
|
|
|
for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
|
|
|
|
ASSERT_EQ(Get(Key(j)), values[j]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
ASSERT_EQ(trivial_move, 1);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
trivial_move = 0;
|
|
|
|
non_trivial_move = 0;
|
|
|
|
values.clear();
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
// Same ranges as above but overlapping
|
|
|
|
ranges = {
|
|
|
|
{100, 199},
|
|
|
|
{300, 399},
|
|
|
|
{0, 99},
|
|
|
|
{200, 299},
|
|
|
|
{600, 699},
|
|
|
|
{400, 499},
|
|
|
|
{500, 560}, // this range overlap with the next one
|
|
|
|
{551, 599},
|
|
|
|
};
|
2015-12-16 00:26:20 +01:00
|
|
|
for (size_t i = 0; i < ranges.size(); i++) {
|
2015-07-21 12:05:57 +02:00
|
|
|
for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
|
|
|
|
values[j] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(j), values[j]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
}
|
|
|
|
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
db_->CompactRange(cro, nullptr, nullptr);
|
2015-07-21 12:05:57 +02:00
|
|
|
|
2015-12-16 00:26:20 +01:00
|
|
|
for (size_t i = 0; i < ranges.size(); i++) {
|
2015-07-21 12:05:57 +02:00
|
|
|
for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
|
|
|
|
ASSERT_EQ(Get(Key(j)), values[j]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ASSERT_EQ(trivial_move, 0);
|
|
|
|
ASSERT_EQ(non_trivial_move, 1);
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->DisableProcessing();
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, TrivialMoveTargetLevel) {
|
2015-07-21 12:05:57 +02:00
|
|
|
int32_t trivial_move = 0;
|
|
|
|
int32_t non_trivial_move = 0;
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:TrivialMove",
|
|
|
|
[&](void* arg) { trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:NonTrivial",
|
|
|
|
[&](void* arg) { non_trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.disable_auto_compactions = true;
|
|
|
|
options.write_buffer_size = 10 * 1024 * 1024;
|
|
|
|
options.num_levels = 7;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
int32_t value_size = 10 * 1024; // 10 KB
|
|
|
|
|
|
|
|
// Add 2 non-overlapping files
|
|
|
|
Random rnd(301);
|
|
|
|
std::map<int32_t, std::string> values;
|
|
|
|
|
|
|
|
// file 1 [0 => 300]
|
|
|
|
for (int32_t i = 0; i <= 300; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// file 2 [600 => 700]
|
|
|
|
for (int32_t i = 600; i <= 700; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// 2 files in L0
|
|
|
|
ASSERT_EQ("2", FilesPerLevel(0));
|
|
|
|
CompactRangeOptions compact_options;
|
|
|
|
compact_options.change_level = true;
|
|
|
|
compact_options.target_level = 6;
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
|
2015-07-21 12:05:57 +02:00
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
|
|
|
|
// 2 files in L6
|
|
|
|
ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel(0));
|
|
|
|
|
|
|
|
ASSERT_EQ(trivial_move, 1);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
for (int32_t i = 0; i <= 300; i++) {
|
|
|
|
ASSERT_EQ(Get(Key(i)), values[i]);
|
|
|
|
}
|
|
|
|
for (int32_t i = 600; i <= 700; i++) {
|
|
|
|
ASSERT_EQ(Get(Key(i)), values[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
TEST_P(DBCompactionTestWithParam, ManualCompactionPartial) {
|
|
|
|
int32_t trivial_move = 0;
|
|
|
|
int32_t non_trivial_move = 0;
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:TrivialMove",
|
|
|
|
[&](void* arg) { trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:NonTrivial",
|
|
|
|
[&](void* arg) { non_trivial_move++; });
|
|
|
|
bool first = true;
|
2016-07-07 20:29:14 +02:00
|
|
|
// Purpose of dependencies:
|
|
|
|
// 4 -> 1: ensure the order of two non-trivial compactions
|
|
|
|
// 5 -> 2 and 5 -> 3: ensure we do a check before two non-trivial compactions
|
|
|
|
// are installed
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
rocksdb::SyncPoint::GetInstance()->LoadDependency(
|
|
|
|
{{"DBCompaction::ManualPartial:4", "DBCompaction::ManualPartial:1"},
|
2016-07-07 20:29:14 +02:00
|
|
|
{"DBCompaction::ManualPartial:5", "DBCompaction::ManualPartial:2"},
|
|
|
|
{"DBCompaction::ManualPartial:5", "DBCompaction::ManualPartial:3"}});
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
|
|
|
|
if (first) {
|
|
|
|
first = false;
|
2016-07-07 20:29:14 +02:00
|
|
|
TEST_SYNC_POINT("DBCompaction::ManualPartial:4");
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
TEST_SYNC_POINT("DBCompaction::ManualPartial:3");
|
2016-07-07 20:29:14 +02:00
|
|
|
} else { // second non-trivial compaction
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
TEST_SYNC_POINT("DBCompaction::ManualPartial:2");
|
|
|
|
}
|
|
|
|
});
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.write_buffer_size = 10 * 1024 * 1024;
|
|
|
|
options.num_levels = 7;
|
|
|
|
options.max_subcompactions = max_subcompactions_;
|
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
|
|
|
options.max_background_compactions = 3;
|
|
|
|
options.target_file_size_base = 1 << 23; // 8 MB
|
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
int32_t value_size = 10 * 1024; // 10 KB
|
|
|
|
|
|
|
|
// Add 2 non-overlapping files
|
|
|
|
Random rnd(301);
|
|
|
|
std::map<int32_t, std::string> values;
|
|
|
|
|
|
|
|
// file 1 [0 => 100]
|
|
|
|
for (int32_t i = 0; i < 100; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// file 2 [100 => 300]
|
|
|
|
for (int32_t i = 100; i < 300; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// 2 files in L0
|
|
|
|
ASSERT_EQ("2", FilesPerLevel(0));
|
|
|
|
CompactRangeOptions compact_options;
|
|
|
|
compact_options.change_level = true;
|
|
|
|
compact_options.target_level = 6;
|
|
|
|
compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
|
|
|
|
// Trivial move the two non-overlapping files to level 6
|
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
|
|
|
|
// 2 files in L6
|
|
|
|
ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel(0));
|
|
|
|
|
|
|
|
ASSERT_EQ(trivial_move, 1);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
// file 3 [ 0 => 200]
|
|
|
|
for (int32_t i = 0; i < 200; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// 1 files in L0
|
|
|
|
ASSERT_EQ("1,0,0,0,0,0,2", FilesPerLevel(0));
|
|
|
|
ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr, false));
|
|
|
|
ASSERT_OK(dbfull()->TEST_CompactRange(1, nullptr, nullptr, nullptr, false));
|
|
|
|
ASSERT_OK(dbfull()->TEST_CompactRange(2, nullptr, nullptr, nullptr, false));
|
|
|
|
ASSERT_OK(dbfull()->TEST_CompactRange(3, nullptr, nullptr, nullptr, false));
|
|
|
|
ASSERT_OK(dbfull()->TEST_CompactRange(4, nullptr, nullptr, nullptr, false));
|
|
|
|
// 2 files in L6, 1 file in L5
|
|
|
|
ASSERT_EQ("0,0,0,0,0,1,2", FilesPerLevel(0));
|
|
|
|
|
|
|
|
ASSERT_EQ(trivial_move, 6);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
std::thread threads([&] {
|
|
|
|
compact_options.change_level = false;
|
|
|
|
compact_options.exclusive_manual_compaction = false;
|
|
|
|
std::string begin_string = Key(0);
|
|
|
|
std::string end_string = Key(199);
|
|
|
|
Slice begin(begin_string);
|
|
|
|
Slice end(end_string);
|
2016-07-07 20:29:14 +02:00
|
|
|
// First non-trivial compaction is triggered
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, &begin, &end));
|
|
|
|
});
|
|
|
|
|
|
|
|
TEST_SYNC_POINT("DBCompaction::ManualPartial:1");
|
|
|
|
// file 4 [300 => 400)
|
|
|
|
for (int32_t i = 300; i <= 400; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// file 5 [400 => 500)
|
|
|
|
for (int32_t i = 400; i <= 500; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// file 6 [500 => 600)
|
|
|
|
for (int32_t i = 500; i <= 600; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
2016-07-07 20:29:14 +02:00
|
|
|
// Second non-trivial compaction is triggered
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
2016-07-07 20:29:14 +02:00
|
|
|
// Before two non-trivial compactions are installed, there are 3 files in L0
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
ASSERT_EQ("3,0,0,0,0,1,2", FilesPerLevel(0));
|
2016-04-13 19:38:45 +02:00
|
|
|
TEST_SYNC_POINT("DBCompaction::ManualPartial:5");
|
|
|
|
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
2016-07-07 20:29:14 +02:00
|
|
|
// After two non-trivial compactions are installed, there is 1 file in L6, and
|
|
|
|
// 1 file in L1
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
ASSERT_EQ("0,1,0,0,0,0,1", FilesPerLevel(0));
|
|
|
|
threads.join();
|
|
|
|
|
|
|
|
for (int32_t i = 0; i < 600; i++) {
|
|
|
|
ASSERT_EQ(Get(Key(i)), values[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST_F(DBCompactionTest, ManualPartialFill) {
|
|
|
|
int32_t trivial_move = 0;
|
|
|
|
int32_t non_trivial_move = 0;
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:TrivialMove",
|
|
|
|
[&](void* arg) { trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:NonTrivial",
|
|
|
|
[&](void* arg) { non_trivial_move++; });
|
|
|
|
bool first = true;
|
|
|
|
bool second = true;
|
|
|
|
rocksdb::SyncPoint::GetInstance()->LoadDependency(
|
|
|
|
{{"DBCompaction::PartialFill:4", "DBCompaction::PartialFill:1"},
|
|
|
|
{"DBCompaction::PartialFill:2", "DBCompaction::PartialFill:3"}});
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:NonTrivial:AfterRun", [&](void* arg) {
|
|
|
|
if (first) {
|
|
|
|
TEST_SYNC_POINT("DBCompaction::PartialFill:4");
|
|
|
|
first = false;
|
|
|
|
TEST_SYNC_POINT("DBCompaction::PartialFill:3");
|
|
|
|
} else if (second) {
|
|
|
|
}
|
|
|
|
});
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.write_buffer_size = 10 * 1024 * 1024;
|
|
|
|
options.max_bytes_for_level_multiplier = 2;
|
|
|
|
options.num_levels = 4;
|
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
|
|
|
options.max_background_compactions = 3;
|
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
int32_t value_size = 10 * 1024; // 10 KB
|
|
|
|
|
|
|
|
// Add 2 non-overlapping files
|
|
|
|
Random rnd(301);
|
|
|
|
std::map<int32_t, std::string> values;
|
|
|
|
|
|
|
|
// file 1 [0 => 100]
|
|
|
|
for (int32_t i = 0; i < 100; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// file 2 [100 => 300]
|
|
|
|
for (int32_t i = 100; i < 300; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// 2 files in L0
|
|
|
|
ASSERT_EQ("2", FilesPerLevel(0));
|
|
|
|
CompactRangeOptions compact_options;
|
|
|
|
compact_options.change_level = true;
|
|
|
|
compact_options.target_level = 2;
|
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
|
|
|
|
// 2 files in L2
|
|
|
|
ASSERT_EQ("0,0,2", FilesPerLevel(0));
|
|
|
|
|
|
|
|
ASSERT_EQ(trivial_move, 1);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
// file 3 [ 0 => 200]
|
|
|
|
for (int32_t i = 0; i < 200; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// 2 files in L2, 1 in L0
|
|
|
|
ASSERT_EQ("1,0,2", FilesPerLevel(0));
|
|
|
|
ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr, nullptr, false));
|
|
|
|
// 2 files in L2, 1 in L1
|
|
|
|
ASSERT_EQ("0,1,2", FilesPerLevel(0));
|
|
|
|
|
|
|
|
ASSERT_EQ(trivial_move, 2);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
std::thread threads([&] {
|
|
|
|
compact_options.change_level = false;
|
|
|
|
compact_options.exclusive_manual_compaction = false;
|
|
|
|
std::string begin_string = Key(0);
|
|
|
|
std::string end_string = Key(199);
|
|
|
|
Slice begin(begin_string);
|
|
|
|
Slice end(end_string);
|
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, &begin, &end));
|
|
|
|
});
|
|
|
|
|
|
|
|
TEST_SYNC_POINT("DBCompaction::PartialFill:1");
|
|
|
|
// Many files 4 [300 => 4300)
|
|
|
|
for (int32_t i = 0; i <= 5; i++) {
|
|
|
|
for (int32_t j = 300; j < 4300; j++) {
|
|
|
|
if (j == 2300) {
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
}
|
|
|
|
values[j] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(j), values[j]));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Verify level sizes
|
|
|
|
uint64_t target_size = 4 * options.max_bytes_for_level_base;
|
|
|
|
for (int32_t i = 1; i < options.num_levels; i++) {
|
|
|
|
ASSERT_LE(SizeAtLevel(i), target_size);
|
2016-11-02 05:05:32 +01:00
|
|
|
target_size = static_cast<uint64_t>(target_size *
|
|
|
|
options.max_bytes_for_level_multiplier);
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
}
|
|
|
|
|
|
|
|
TEST_SYNC_POINT("DBCompaction::PartialFill:2");
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
threads.join();
|
|
|
|
|
|
|
|
for (int32_t i = 0; i < 4300; i++) {
|
|
|
|
ASSERT_EQ(Get(Key(i)), values[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-12-29 22:22:13 +01:00
|
|
|
TEST_F(DBCompactionTest, DeleteFileRange) {
|
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.write_buffer_size = 10 * 1024 * 1024;
|
|
|
|
options.max_bytes_for_level_multiplier = 2;
|
|
|
|
options.num_levels = 4;
|
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
|
|
|
options.max_background_compactions = 3;
|
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
int32_t value_size = 10 * 1024; // 10 KB
|
|
|
|
|
|
|
|
// Add 2 non-overlapping files
|
|
|
|
Random rnd(301);
|
|
|
|
std::map<int32_t, std::string> values;
|
|
|
|
|
|
|
|
// file 1 [0 => 100]
|
|
|
|
for (int32_t i = 0; i < 100; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// file 2 [100 => 300]
|
|
|
|
for (int32_t i = 100; i < 300; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// 2 files in L0
|
|
|
|
ASSERT_EQ("2", FilesPerLevel(0));
|
|
|
|
CompactRangeOptions compact_options;
|
|
|
|
compact_options.change_level = true;
|
|
|
|
compact_options.target_level = 2;
|
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
|
|
|
|
// 2 files in L2
|
|
|
|
ASSERT_EQ("0,0,2", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// file 3 [ 0 => 200]
|
|
|
|
for (int32_t i = 0; i < 200; i++) {
|
|
|
|
values[i] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
// Many files 4 [300 => 4300)
|
|
|
|
for (int32_t i = 0; i <= 5; i++) {
|
|
|
|
for (int32_t j = 300; j < 4300; j++) {
|
|
|
|
if (j == 2300) {
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
}
|
|
|
|
values[j] = RandomString(&rnd, value_size);
|
|
|
|
ASSERT_OK(Put(Key(j), values[j]));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
// Verify level sizes
|
|
|
|
uint64_t target_size = 4 * options.max_bytes_for_level_base;
|
|
|
|
for (int32_t i = 1; i < options.num_levels; i++) {
|
|
|
|
ASSERT_LE(SizeAtLevel(i), target_size);
|
2016-11-02 05:05:32 +01:00
|
|
|
target_size = static_cast<uint64_t>(target_size *
|
|
|
|
options.max_bytes_for_level_multiplier);
|
2015-12-29 22:22:13 +01:00
|
|
|
}
|
|
|
|
|
2015-12-30 01:44:12 +01:00
|
|
|
size_t old_num_files = CountFiles();
|
2015-12-29 22:22:13 +01:00
|
|
|
std::string begin_string = Key(1000);
|
|
|
|
std::string end_string = Key(2000);
|
|
|
|
Slice begin(begin_string);
|
|
|
|
Slice end(end_string);
|
|
|
|
ASSERT_OK(DeleteFilesInRange(db_, db_->DefaultColumnFamily(), &begin, &end));
|
|
|
|
|
|
|
|
int32_t deleted_count = 0;
|
|
|
|
for (int32_t i = 0; i < 4300; i++) {
|
|
|
|
if (i < 1000 || i > 2000) {
|
|
|
|
ASSERT_EQ(Get(Key(i)), values[i]);
|
|
|
|
} else {
|
|
|
|
ReadOptions roptions;
|
|
|
|
std::string result;
|
|
|
|
Status s = db_->Get(roptions, Key(i), &result);
|
|
|
|
ASSERT_TRUE(s.IsNotFound() || s.ok());
|
|
|
|
if (s.IsNotFound()) {
|
|
|
|
deleted_count++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ASSERT_GT(deleted_count, 0);
|
2016-01-04 19:55:31 +01:00
|
|
|
begin_string = Key(5000);
|
|
|
|
end_string = Key(6000);
|
|
|
|
Slice begin1(begin_string);
|
|
|
|
Slice end1(end_string);
|
|
|
|
// Try deleting files in range which contain no keys
|
|
|
|
ASSERT_OK(
|
|
|
|
DeleteFilesInRange(db_, db_->DefaultColumnFamily(), &begin1, &end1));
|
2015-12-29 22:22:13 +01:00
|
|
|
|
2016-01-07 23:48:45 +01:00
|
|
|
// Push data from level 0 to level 1 to force all data to be deleted
|
|
|
|
// Note that we don't delete level 0 files
|
|
|
|
compact_options.change_level = true;
|
|
|
|
compact_options.target_level = 1;
|
|
|
|
ASSERT_OK(dbfull()->TEST_CompactRange(0, nullptr, nullptr));
|
|
|
|
|
2015-12-29 22:22:13 +01:00
|
|
|
ASSERT_OK(
|
|
|
|
DeleteFilesInRange(db_, db_->DefaultColumnFamily(), nullptr, nullptr));
|
|
|
|
|
|
|
|
int32_t deleted_count2 = 0;
|
|
|
|
for (int32_t i = 0; i < 4300; i++) {
|
|
|
|
ReadOptions roptions;
|
|
|
|
std::string result;
|
|
|
|
Status s = db_->Get(roptions, Key(i), &result);
|
|
|
|
ASSERT_TRUE(s.IsNotFound());
|
|
|
|
deleted_count2++;
|
|
|
|
}
|
|
|
|
ASSERT_GT(deleted_count2, deleted_count);
|
2015-12-30 01:44:12 +01:00
|
|
|
size_t new_num_files = CountFiles();
|
2015-12-29 22:22:13 +01:00
|
|
|
ASSERT_GT(old_num_files, new_num_files);
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, TrivialMoveToLastLevelWithFiles) {
|
2015-07-21 12:05:57 +02:00
|
|
|
int32_t trivial_move = 0;
|
|
|
|
int32_t non_trivial_move = 0;
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:TrivialMove",
|
|
|
|
[&](void* arg) { trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:NonTrivial",
|
|
|
|
[&](void* arg) { non_trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-07-21 12:05:57 +02:00
|
|
|
options.write_buffer_size = 100000000;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
int32_t value_size = 10 * 1024; // 10 KB
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
std::vector<std::string> values;
|
|
|
|
// File with keys [ 0 => 99 ]
|
|
|
|
for (int i = 0; i < 100; i++) {
|
|
|
|
values.push_back(RandomString(&rnd, value_size));
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
ASSERT_EQ("1", FilesPerLevel(0));
|
|
|
|
// Compaction will do L0=>L1 (trivial move) then move L1 files to L3
|
|
|
|
CompactRangeOptions compact_options;
|
|
|
|
compact_options.change_level = true;
|
|
|
|
compact_options.target_level = 3;
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
|
2015-07-21 12:05:57 +02:00
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
|
|
|
|
ASSERT_EQ("0,0,0,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(trivial_move, 1);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
// File with keys [ 100 => 199 ]
|
|
|
|
for (int i = 100; i < 200; i++) {
|
|
|
|
values.push_back(RandomString(&rnd, value_size));
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
ASSERT_EQ("1,0,0,1", FilesPerLevel(0));
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
CompactRangeOptions cro;
|
|
|
|
cro.exclusive_manual_compaction = exclusive_manual_compaction_;
|
2015-07-21 12:05:57 +02:00
|
|
|
// Compaction will do L0=>L1 L1=>L2 L2=>L3 (3 trivial moves)
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
|
2015-07-21 12:05:57 +02:00
|
|
|
ASSERT_EQ("0,0,0,2", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(trivial_move, 4);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
for (int i = 0; i < 200; i++) {
|
|
|
|
ASSERT_EQ(Get(Key(i)), values[i]);
|
|
|
|
}
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->DisableProcessing();
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, LevelCompactionThirdPath) {
|
2015-07-21 12:05:57 +02:00
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.db_paths.emplace_back(dbname_, 500 * 1024);
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_2", 4 * 1024 * 1024);
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_3", 1024 * 1024 * 1024);
|
2015-12-07 20:57:00 +01:00
|
|
|
options.memtable_factory.reset(
|
|
|
|
new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
|
2015-07-21 12:05:57 +02:00
|
|
|
options.compaction_style = kCompactionStyleLevel;
|
2015-08-26 23:19:31 +02:00
|
|
|
options.write_buffer_size = 110 << 10; // 110KB
|
|
|
|
options.arena_block_size = 4 << 10;
|
2015-07-21 12:05:57 +02:00
|
|
|
options.level0_file_num_compaction_trigger = 2;
|
|
|
|
options.num_levels = 4;
|
|
|
|
options.max_bytes_for_level_base = 400 * 1024;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
// options = CurrentOptions(options);
|
|
|
|
|
|
|
|
std::vector<std::string> filenames;
|
|
|
|
env_->GetChildren(options.db_paths[1].path, &filenames);
|
|
|
|
// Delete archival files.
|
|
|
|
for (size_t i = 0; i < filenames.size(); ++i) {
|
|
|
|
env_->DeleteFile(options.db_paths[1].path + "/" + filenames[i]);
|
|
|
|
}
|
|
|
|
env_->DeleteDir(options.db_paths[1].path);
|
|
|
|
Reopen(options);
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
int key_idx = 0;
|
|
|
|
|
|
|
|
// First three 110KB files are not going to second path.
|
|
|
|
// After that, (100K, 200K)
|
|
|
|
for (int num = 0; num < 3; num++) {
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Another 110KB triggers a compaction to 400K file to fill up first path
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ(3, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
|
|
|
|
// (1, 4)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// (1, 4, 1)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// (1, 4, 2)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,2", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(2, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// (1, 4, 3)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,3", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(3, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// (1, 4, 4)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,4", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// (1, 4, 5)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,5", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(5, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// (1, 4, 6)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,6", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(6, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// (1, 4, 7)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,7", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(7, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// (1, 4, 8)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,8", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(8, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
for (int i = 0; i < key_idx; i++) {
|
|
|
|
auto v = Get(Key(i));
|
|
|
|
ASSERT_NE(v, "NOT_FOUND");
|
2015-08-26 23:19:31 +02:00
|
|
|
ASSERT_TRUE(v.size() == 1 || v.size() == 990);
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
Reopen(options);
|
|
|
|
|
|
|
|
for (int i = 0; i < key_idx; i++) {
|
|
|
|
auto v = Get(Key(i));
|
|
|
|
ASSERT_NE(v, "NOT_FOUND");
|
2015-08-26 23:19:31 +02:00
|
|
|
ASSERT_TRUE(v.size() == 1 || v.size() == 990);
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
Destroy(options);
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, LevelCompactionPathUse) {
|
2015-07-21 12:05:57 +02:00
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.db_paths.emplace_back(dbname_, 500 * 1024);
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_2", 4 * 1024 * 1024);
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_3", 1024 * 1024 * 1024);
|
2015-12-07 20:57:00 +01:00
|
|
|
options.memtable_factory.reset(
|
|
|
|
new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
|
2015-07-21 12:05:57 +02:00
|
|
|
options.compaction_style = kCompactionStyleLevel;
|
2015-09-01 08:11:12 +02:00
|
|
|
options.write_buffer_size = 110 << 10; // 110KB
|
2015-08-26 23:19:31 +02:00
|
|
|
options.arena_block_size = 4 << 10;
|
2015-07-21 12:05:57 +02:00
|
|
|
options.level0_file_num_compaction_trigger = 2;
|
|
|
|
options.num_levels = 4;
|
|
|
|
options.max_bytes_for_level_base = 400 * 1024;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
// options = CurrentOptions(options);
|
|
|
|
|
|
|
|
std::vector<std::string> filenames;
|
|
|
|
env_->GetChildren(options.db_paths[1].path, &filenames);
|
|
|
|
// Delete archival files.
|
|
|
|
for (size_t i = 0; i < filenames.size(); ++i) {
|
|
|
|
env_->DeleteFile(options.db_paths[1].path + "/" + filenames[i]);
|
|
|
|
}
|
|
|
|
env_->DeleteDir(options.db_paths[1].path);
|
|
|
|
Reopen(options);
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
int key_idx = 0;
|
|
|
|
|
|
|
|
// Always gets compacted into 1 Level1 file,
|
|
|
|
// 0/1 Level 0 file
|
|
|
|
for (int num = 0; num < 3; num++) {
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
}
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("0,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("0,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("0,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("0,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
key_idx = 0;
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
for (int i = 0; i < key_idx; i++) {
|
|
|
|
auto v = Get(Key(i));
|
|
|
|
ASSERT_NE(v, "NOT_FOUND");
|
2015-08-26 23:19:31 +02:00
|
|
|
ASSERT_TRUE(v.size() == 1 || v.size() == 990);
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
Reopen(options);
|
|
|
|
|
|
|
|
for (int i = 0; i < key_idx; i++) {
|
|
|
|
auto v = Get(Key(i));
|
|
|
|
ASSERT_NE(v, "NOT_FOUND");
|
2015-08-26 23:19:31 +02:00
|
|
|
ASSERT_TRUE(v.size() == 1 || v.size() == 990);
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
Destroy(options);
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, ConvertCompactionStyle) {
|
2015-07-21 12:05:57 +02:00
|
|
|
Random rnd(301);
|
|
|
|
int max_key_level_insert = 200;
|
|
|
|
int max_key_universal_insert = 600;
|
|
|
|
|
|
|
|
// Stage 1: generate a db with level compaction
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-08-26 23:19:31 +02:00
|
|
|
options.write_buffer_size = 110 << 10; // 110KB
|
|
|
|
options.arena_block_size = 4 << 10;
|
2015-07-21 12:05:57 +02:00
|
|
|
options.num_levels = 4;
|
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
|
|
|
options.max_bytes_for_level_base = 500 << 10; // 500KB
|
|
|
|
options.max_bytes_for_level_multiplier = 1;
|
|
|
|
options.target_file_size_base = 200 << 10; // 200KB
|
|
|
|
options.target_file_size_multiplier = 1;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
|
|
|
|
for (int i = 0; i <= max_key_level_insert; i++) {
|
|
|
|
// each value is 10K
|
|
|
|
ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush(1));
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
ASSERT_GT(TotalTableFiles(1, 4), 1);
|
|
|
|
int non_level0_num_files = 0;
|
|
|
|
for (int i = 1; i < options.num_levels; i++) {
|
|
|
|
non_level0_num_files += NumTableFilesAtLevel(i, 1);
|
|
|
|
}
|
|
|
|
ASSERT_GT(non_level0_num_files, 0);
|
|
|
|
|
|
|
|
// Stage 2: reopen with universal compaction - should fail
|
|
|
|
options = CurrentOptions();
|
|
|
|
options.compaction_style = kCompactionStyleUniversal;
|
|
|
|
options.num_levels = 1;
|
|
|
|
options = CurrentOptions(options);
|
|
|
|
Status s = TryReopenWithColumnFamilies({"default", "pikachu"}, options);
|
|
|
|
ASSERT_TRUE(s.IsInvalidArgument());
|
|
|
|
|
|
|
|
// Stage 3: compact into a single file and move the file to level 0
|
|
|
|
options = CurrentOptions();
|
|
|
|
options.disable_auto_compactions = true;
|
|
|
|
options.target_file_size_base = INT_MAX;
|
|
|
|
options.target_file_size_multiplier = 1;
|
|
|
|
options.max_bytes_for_level_base = INT_MAX;
|
|
|
|
options.max_bytes_for_level_multiplier = 1;
|
|
|
|
options.num_levels = 4;
|
|
|
|
options = CurrentOptions(options);
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, options);
|
|
|
|
|
|
|
|
CompactRangeOptions compact_options;
|
|
|
|
compact_options.change_level = true;
|
|
|
|
compact_options.target_level = 0;
|
|
|
|
compact_options.bottommost_level_compaction =
|
|
|
|
BottommostLevelCompaction::kForce;
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
|
2015-07-21 12:05:57 +02:00
|
|
|
dbfull()->CompactRange(compact_options, handles_[1], nullptr, nullptr);
|
|
|
|
|
|
|
|
// Only 1 file in L0
|
|
|
|
ASSERT_EQ("1", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Stage 4: re-open in universal compaction style and do some db operations
|
|
|
|
options = CurrentOptions();
|
|
|
|
options.compaction_style = kCompactionStyleUniversal;
|
|
|
|
options.num_levels = 4;
|
2015-08-26 23:19:31 +02:00
|
|
|
options.write_buffer_size = 110 << 10; // 110KB
|
|
|
|
options.arena_block_size = 4 << 10;
|
2015-07-21 12:05:57 +02:00
|
|
|
options.level0_file_num_compaction_trigger = 3;
|
|
|
|
options = CurrentOptions(options);
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, options);
|
|
|
|
|
|
|
|
options.num_levels = 1;
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, options);
|
|
|
|
|
|
|
|
for (int i = max_key_level_insert / 2; i <= max_key_universal_insert; i++) {
|
|
|
|
ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
|
|
|
|
}
|
|
|
|
dbfull()->Flush(FlushOptions());
|
|
|
|
ASSERT_OK(Flush(1));
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
for (int i = 1; i < options.num_levels; i++) {
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(i, 1), 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
// verify keys inserted in both level compaction style and universal
|
|
|
|
// compaction style
|
|
|
|
std::string keys_in_db;
|
|
|
|
Iterator* iter = dbfull()->NewIterator(ReadOptions(), handles_[1]);
|
|
|
|
for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
|
|
|
|
keys_in_db.append(iter->key().ToString());
|
|
|
|
keys_in_db.push_back(',');
|
|
|
|
}
|
|
|
|
delete iter;
|
|
|
|
|
|
|
|
std::string expected_keys;
|
|
|
|
for (int i = 0; i <= max_key_universal_insert; i++) {
|
|
|
|
expected_keys.append(Key(i));
|
|
|
|
expected_keys.push_back(',');
|
|
|
|
}
|
|
|
|
|
|
|
|
ASSERT_EQ(keys_in_db, expected_keys);
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST_F(DBCompactionTest, L0_CompactionBug_Issue44_a) {
|
|
|
|
do {
|
|
|
|
CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
|
|
|
|
ASSERT_OK(Put(1, "b", "v"));
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
ASSERT_OK(Delete(1, "b"));
|
|
|
|
ASSERT_OK(Delete(1, "a"));
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
ASSERT_OK(Delete(1, "a"));
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
ASSERT_OK(Put(1, "a", "v"));
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
ASSERT_EQ("(a->v)", Contents(1));
|
|
|
|
env_->SleepForMicroseconds(1000000); // Wait for compaction to finish
|
|
|
|
ASSERT_EQ("(a->v)", Contents(1));
|
|
|
|
} while (ChangeCompactOptions());
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST_F(DBCompactionTest, L0_CompactionBug_Issue44_b) {
|
|
|
|
do {
|
|
|
|
CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
|
|
|
|
Put(1, "", "");
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
Delete(1, "e");
|
|
|
|
Put(1, "", "");
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
Put(1, "c", "cv");
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
Put(1, "", "");
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
Put(1, "", "");
|
|
|
|
env_->SleepForMicroseconds(1000000); // Wait for compaction to finish
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
Put(1, "d", "dv");
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
Put(1, "", "");
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
Delete(1, "d");
|
|
|
|
Delete(1, "b");
|
|
|
|
ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
|
|
|
|
ASSERT_EQ("(->)(c->cv)", Contents(1));
|
|
|
|
env_->SleepForMicroseconds(1000000); // Wait for compaction to finish
|
|
|
|
ASSERT_EQ("(->)(c->cv)", Contents(1));
|
|
|
|
} while (ChangeCompactOptions());
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, ManualCompaction) {
|
2015-07-21 12:05:57 +02:00
|
|
|
Options options = CurrentOptions();
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-10-20 08:40:15 +02:00
|
|
|
options.statistics = rocksdb::CreateDBStatistics();
|
2015-07-21 12:05:57 +02:00
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
|
|
|
|
// iter - 0 with 7 levels
|
|
|
|
// iter - 1 with 3 levels
|
|
|
|
for (int iter = 0; iter < 2; ++iter) {
|
|
|
|
MakeTables(3, "p", "q", 1);
|
|
|
|
ASSERT_EQ("1,1,1", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Compaction range falls before files
|
|
|
|
Compact(1, "", "c");
|
|
|
|
ASSERT_EQ("1,1,1", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Compaction range falls after files
|
|
|
|
Compact(1, "r", "z");
|
|
|
|
ASSERT_EQ("1,1,1", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Compaction range overlaps files
|
|
|
|
Compact(1, "p1", "p9");
|
|
|
|
ASSERT_EQ("0,0,1", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Populate a different range
|
|
|
|
MakeTables(3, "c", "e", 1);
|
|
|
|
ASSERT_EQ("1,1,2", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Compact just the new range
|
|
|
|
Compact(1, "b", "f");
|
|
|
|
ASSERT_EQ("0,0,2", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Compact all
|
|
|
|
MakeTables(1, "a", "z", 1);
|
|
|
|
ASSERT_EQ("1,0,2", FilesPerLevel(1));
|
2015-10-20 08:40:15 +02:00
|
|
|
|
|
|
|
uint64_t prev_block_cache_add =
|
|
|
|
options.statistics->getTickerCount(BLOCK_CACHE_ADD);
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
CompactRangeOptions cro;
|
|
|
|
cro.exclusive_manual_compaction = exclusive_manual_compaction_;
|
|
|
|
db_->CompactRange(cro, handles_[1], nullptr, nullptr);
|
2015-10-20 08:40:15 +02:00
|
|
|
// Verify manual compaction doesn't fill block cache
|
|
|
|
ASSERT_EQ(prev_block_cache_add,
|
|
|
|
options.statistics->getTickerCount(BLOCK_CACHE_ADD));
|
|
|
|
|
2015-07-21 12:05:57 +02:00
|
|
|
ASSERT_EQ("0,0,1", FilesPerLevel(1));
|
|
|
|
|
|
|
|
if (iter == 0) {
|
|
|
|
options = CurrentOptions();
|
|
|
|
options.max_background_flushes = 0;
|
|
|
|
options.num_levels = 3;
|
|
|
|
options.create_if_missing = true;
|
2015-10-20 08:40:15 +02:00
|
|
|
options.statistics = rocksdb::CreateDBStatistics();
|
2015-07-21 12:05:57 +02:00
|
|
|
DestroyAndReopen(options);
|
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, ManualLevelCompactionOutputPathId) {
|
2015-07-21 12:05:57 +02:00
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_2", 2 * 10485760);
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_3", 100 * 10485760);
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_4", 120 * 10485760);
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
|
|
|
|
// iter - 0 with 7 levels
|
|
|
|
// iter - 1 with 3 levels
|
|
|
|
for (int iter = 0; iter < 2; ++iter) {
|
|
|
|
for (int i = 0; i < 3; ++i) {
|
|
|
|
ASSERT_OK(Put(1, "p", "begin"));
|
|
|
|
ASSERT_OK(Put(1, "q", "end"));
|
|
|
|
ASSERT_OK(Flush(1));
|
|
|
|
}
|
|
|
|
ASSERT_EQ("3", FilesPerLevel(1));
|
|
|
|
ASSERT_EQ(3, GetSstFileCount(options.db_paths[0].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// Compaction range falls before files
|
|
|
|
Compact(1, "", "c");
|
|
|
|
ASSERT_EQ("3", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Compaction range falls after files
|
|
|
|
Compact(1, "r", "z");
|
|
|
|
ASSERT_EQ("3", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Compaction range overlaps files
|
|
|
|
Compact(1, "p1", "p9", 1);
|
|
|
|
ASSERT_EQ("0,1", FilesPerLevel(1));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[0].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// Populate a different range
|
|
|
|
for (int i = 0; i < 3; ++i) {
|
|
|
|
ASSERT_OK(Put(1, "c", "begin"));
|
|
|
|
ASSERT_OK(Put(1, "e", "end"));
|
|
|
|
ASSERT_OK(Flush(1));
|
|
|
|
}
|
|
|
|
ASSERT_EQ("3,1", FilesPerLevel(1));
|
|
|
|
|
|
|
|
// Compact just the new range
|
|
|
|
Compact(1, "b", "f", 1);
|
|
|
|
ASSERT_EQ("0,2", FilesPerLevel(1));
|
|
|
|
ASSERT_EQ(2, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[0].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// Compact all
|
|
|
|
ASSERT_OK(Put(1, "a", "begin"));
|
|
|
|
ASSERT_OK(Put(1, "z", "end"));
|
|
|
|
ASSERT_OK(Flush(1));
|
|
|
|
ASSERT_EQ("1,2", FilesPerLevel(1));
|
|
|
|
ASSERT_EQ(2, GetSstFileCount(options.db_paths[1].path));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[0].path));
|
|
|
|
CompactRangeOptions compact_options;
|
|
|
|
compact_options.target_path_id = 1;
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
compact_options.exclusive_manual_compaction = exclusive_manual_compaction_;
|
2015-07-21 12:05:57 +02:00
|
|
|
db_->CompactRange(compact_options, handles_[1], nullptr, nullptr);
|
2015-08-05 07:19:07 +02:00
|
|
|
|
2015-08-18 23:56:31 +02:00
|
|
|
ASSERT_EQ("0,1", FilesPerLevel(1));
|
|
|
|
ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
|
2015-07-21 12:05:57 +02:00
|
|
|
ASSERT_EQ(0, GetSstFileCount(options.db_paths[0].path));
|
|
|
|
ASSERT_EQ(0, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
if (iter == 0) {
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
options = CurrentOptions();
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_2", 2 * 10485760);
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_3", 100 * 10485760);
|
|
|
|
options.db_paths.emplace_back(dbname_ + "_4", 120 * 10485760);
|
|
|
|
options.max_background_flushes = 1;
|
|
|
|
options.num_levels = 3;
|
|
|
|
options.create_if_missing = true;
|
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST_F(DBCompactionTest, FilesDeletedAfterCompaction) {
|
|
|
|
do {
|
|
|
|
CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
|
|
|
|
ASSERT_OK(Put(1, "foo", "v2"));
|
|
|
|
Compact(1, "a", "z");
|
|
|
|
const size_t num_files = CountLiveFiles();
|
|
|
|
for (int i = 0; i < 10; i++) {
|
|
|
|
ASSERT_OK(Put(1, "foo", "v2"));
|
|
|
|
Compact(1, "a", "z");
|
|
|
|
}
|
|
|
|
ASSERT_EQ(CountLiveFiles(), num_files);
|
|
|
|
} while (ChangeCompactOptions());
|
|
|
|
}
|
|
|
|
|
Fix CompactFiles by adding all necessary files
Summary:
The compact files API had a bug where some overlapping files
are not added. These are files which overlap with files which were
added to the compaction input files, but not to the original set of
input files. This happens only when there are more than two levels
involved in the compaction. An example will illustrate this better.
Level 2 has 1 input file 1.sst which spans [20,30].
Level 3 has added file 2.sst which spans [10,25]
Level 4 has file 3.sst which spans [35,40] and
input file 4.sst which spans [46,50].
The existing code would not add 3.sst to the set of input_files because
it only becomes an overlapping file in level 4 and it wasn't one in
level 3.
When installing the results of the compaction, 3.sst would overlap with
output file from the compact files and result in the assertion in
version_set.cc:1130
// Must not overlap
assert(level <= 0 || level_files->empty() ||
internal_comparator_->Compare(
(*level_files)[level_files->size() - 1]->largest, f->smallest) <
0);
This change now adds overlapping files from the current level to the set
of input files also so that we don't hit the assertion above.
Test Plan:
d=/tmp/j; rm -rf $d; seq 1000 | parallel --gnu --eta
'd=/tmp/j/d-{}; mkdir -p $d; TEST_TMPDIR=$d ./db_compaction_test
--gtest_filter=*CompactilesOnLevel* --gtest_also_run_disabled_tests >&
'$d'/log-{}'
Reviewers: igor, yhchiang, sdong
Reviewed By: yhchiang
Subscribers: dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D43437
2015-08-04 00:53:22 +02:00
|
|
|
// Check level comapction with compact files
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, DISABLED_CompactFilesOnLevelCompaction) {
|
2015-07-21 12:05:57 +02:00
|
|
|
const int kTestKeySize = 16;
|
|
|
|
const int kTestValueSize = 984;
|
|
|
|
const int kEntrySize = kTestKeySize + kTestValueSize;
|
|
|
|
const int kEntriesPerBuffer = 100;
|
|
|
|
Options options;
|
|
|
|
options.create_if_missing = true;
|
|
|
|
options.write_buffer_size = kEntrySize * kEntriesPerBuffer;
|
|
|
|
options.compaction_style = kCompactionStyleLevel;
|
|
|
|
options.target_file_size_base = options.write_buffer_size;
|
|
|
|
options.max_bytes_for_level_base = options.target_file_size_base * 2;
|
|
|
|
options.level0_stop_writes_trigger = 2;
|
|
|
|
options.max_bytes_for_level_multiplier = 2;
|
|
|
|
options.compression = kNoCompression;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
options = CurrentOptions(options);
|
|
|
|
CreateAndReopenWithCF({"pikachu"}, options);
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
for (int key = 64 * kEntriesPerBuffer; key >= 0; --key) {
|
|
|
|
ASSERT_OK(Put(1, ToString(key), RandomString(&rnd, kTestValueSize)));
|
|
|
|
}
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
ColumnFamilyMetaData cf_meta;
|
|
|
|
dbfull()->GetColumnFamilyMetaData(handles_[1], &cf_meta);
|
|
|
|
int output_level = static_cast<int>(cf_meta.levels.size()) - 1;
|
|
|
|
for (int file_picked = 5; file_picked > 0; --file_picked) {
|
|
|
|
std::set<std::string> overlapping_file_names;
|
|
|
|
std::vector<std::string> compaction_input_file_names;
|
|
|
|
for (int f = 0; f < file_picked; ++f) {
|
2016-01-19 05:45:21 +01:00
|
|
|
int level = 0;
|
2015-07-21 12:05:57 +02:00
|
|
|
auto file_meta = PickFileRandomly(cf_meta, &rnd, &level);
|
|
|
|
compaction_input_file_names.push_back(file_meta->name);
|
|
|
|
GetOverlappingFileNumbersForLevelCompaction(
|
|
|
|
cf_meta, options.comparator, level, output_level,
|
|
|
|
file_meta, &overlapping_file_names);
|
|
|
|
}
|
|
|
|
|
|
|
|
ASSERT_OK(dbfull()->CompactFiles(
|
|
|
|
CompactionOptions(), handles_[1],
|
|
|
|
compaction_input_file_names,
|
|
|
|
output_level));
|
|
|
|
|
|
|
|
// Make sure all overlapping files do not exist after compaction
|
|
|
|
dbfull()->GetColumnFamilyMetaData(handles_[1], &cf_meta);
|
|
|
|
VerifyCompactionResult(cf_meta, overlapping_file_names);
|
|
|
|
}
|
|
|
|
|
|
|
|
// make sure all key-values are still there.
|
|
|
|
for (int key = 64 * kEntriesPerBuffer; key >= 0; --key) {
|
|
|
|
ASSERT_NE(Get(1, ToString(key)), "NOT_FOUND");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, PartialCompactionFailure) {
|
2015-07-21 12:05:57 +02:00
|
|
|
Options options;
|
|
|
|
const int kKeySize = 16;
|
|
|
|
const int kKvSize = 1000;
|
|
|
|
const int kKeysPerBuffer = 100;
|
|
|
|
const int kNumL1Files = 5;
|
|
|
|
options.create_if_missing = true;
|
|
|
|
options.write_buffer_size = kKeysPerBuffer * kKvSize;
|
|
|
|
options.max_write_buffer_number = 2;
|
|
|
|
options.target_file_size_base =
|
|
|
|
options.write_buffer_size *
|
|
|
|
(options.max_write_buffer_number - 1);
|
|
|
|
options.level0_file_num_compaction_trigger = kNumL1Files;
|
|
|
|
options.max_bytes_for_level_base =
|
|
|
|
options.level0_file_num_compaction_trigger *
|
|
|
|
options.target_file_size_base;
|
|
|
|
options.max_bytes_for_level_multiplier = 2;
|
|
|
|
options.compression = kNoCompression;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
|
|
|
|
env_->SetBackgroundThreads(1, Env::HIGH);
|
|
|
|
env_->SetBackgroundThreads(1, Env::LOW);
|
|
|
|
// stop the compaction thread until we simulate the file creation failure.
|
2015-09-25 19:29:44 +02:00
|
|
|
test::SleepingBackgroundTask sleeping_task_low;
|
|
|
|
env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask, &sleeping_task_low,
|
2015-07-21 12:05:57 +02:00
|
|
|
Env::Priority::LOW);
|
|
|
|
|
|
|
|
options.env = env_;
|
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
const int kNumInsertedKeys =
|
|
|
|
options.level0_file_num_compaction_trigger *
|
|
|
|
(options.max_write_buffer_number - 1) *
|
|
|
|
kKeysPerBuffer;
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
std::vector<std::string> keys;
|
|
|
|
std::vector<std::string> values;
|
|
|
|
for (int k = 0; k < kNumInsertedKeys; ++k) {
|
|
|
|
keys.emplace_back(RandomString(&rnd, kKeySize));
|
|
|
|
values.emplace_back(RandomString(&rnd, kKvSize - kKeySize));
|
|
|
|
ASSERT_OK(Put(Slice(keys[k]), Slice(values[k])));
|
2015-07-22 22:45:08 +02:00
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
dbfull()->TEST_FlushMemTable(true);
|
|
|
|
// Make sure the number of L0 files can trigger compaction.
|
|
|
|
ASSERT_GE(NumTableFilesAtLevel(0),
|
|
|
|
options.level0_file_num_compaction_trigger);
|
|
|
|
|
|
|
|
auto previous_num_level0_files = NumTableFilesAtLevel(0);
|
|
|
|
|
|
|
|
// Fail the first file creation.
|
|
|
|
env_->non_writable_count_ = 1;
|
|
|
|
sleeping_task_low.WakeUp();
|
|
|
|
sleeping_task_low.WaitUntilDone();
|
|
|
|
|
|
|
|
// Expect compaction to fail here as one file will fail its
|
|
|
|
// creation.
|
|
|
|
ASSERT_TRUE(!dbfull()->TEST_WaitForCompact().ok());
|
|
|
|
|
|
|
|
// Verify L0 -> L1 compaction does fail.
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(1), 0);
|
|
|
|
|
|
|
|
// Verify all L0 files are still there.
|
|
|
|
ASSERT_EQ(NumTableFilesAtLevel(0), previous_num_level0_files);
|
|
|
|
|
|
|
|
// All key-values must exist after compaction fails.
|
|
|
|
for (int k = 0; k < kNumInsertedKeys; ++k) {
|
|
|
|
ASSERT_EQ(values[k], Get(keys[k]));
|
|
|
|
}
|
|
|
|
|
|
|
|
env_->non_writable_count_ = 0;
|
|
|
|
|
|
|
|
// Make sure RocksDB will not get into corrupted state.
|
|
|
|
Reopen(options);
|
|
|
|
|
|
|
|
// Verify again after reopen.
|
|
|
|
for (int k = 0; k < kNumInsertedKeys; ++k) {
|
|
|
|
ASSERT_EQ(values[k], Get(keys[k]));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, DeleteMovedFileAfterCompaction) {
|
2015-07-21 12:05:57 +02:00
|
|
|
// iter 1 -- delete_obsolete_files_period_micros == 0
|
|
|
|
for (int iter = 0; iter < 2; ++iter) {
|
|
|
|
// This test triggers move compaction and verifies that the file is not
|
|
|
|
// deleted when it's part of move compaction
|
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.env = env_;
|
|
|
|
if (iter == 1) {
|
|
|
|
options.delete_obsolete_files_period_micros = 0;
|
|
|
|
}
|
|
|
|
options.create_if_missing = true;
|
|
|
|
options.level0_file_num_compaction_trigger =
|
|
|
|
2; // trigger compaction when we have 2 files
|
|
|
|
OnFileDeletionListener* listener = new OnFileDeletionListener();
|
|
|
|
options.listeners.emplace_back(listener);
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
// Create two 1MB sst files
|
|
|
|
for (int i = 0; i < 2; ++i) {
|
|
|
|
// Create 1MB sst file
|
|
|
|
for (int j = 0; j < 100; ++j) {
|
|
|
|
ASSERT_OK(Put(Key(i * 50 + j), RandomString(&rnd, 10 * 1024)));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
}
|
|
|
|
// this should execute L0->L1
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
ASSERT_EQ("0,1", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// block compactions
|
2015-09-25 19:29:44 +02:00
|
|
|
test::SleepingBackgroundTask sleeping_task;
|
|
|
|
env_->Schedule(&test::SleepingBackgroundTask::DoSleepTask, &sleeping_task,
|
2015-07-21 12:05:57 +02:00
|
|
|
Env::Priority::LOW);
|
|
|
|
|
|
|
|
options.max_bytes_for_level_base = 1024 * 1024; // 1 MB
|
|
|
|
Reopen(options);
|
|
|
|
std::unique_ptr<Iterator> iterator(db_->NewIterator(ReadOptions()));
|
|
|
|
ASSERT_EQ("0,1", FilesPerLevel(0));
|
|
|
|
// let compactions go
|
|
|
|
sleeping_task.WakeUp();
|
|
|
|
sleeping_task.WaitUntilDone();
|
|
|
|
|
|
|
|
// this should execute L1->L2 (move)
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
ASSERT_EQ("0,0,1", FilesPerLevel(0));
|
|
|
|
|
|
|
|
std::vector<LiveFileMetaData> metadata;
|
|
|
|
db_->GetLiveFilesMetaData(&metadata);
|
|
|
|
ASSERT_EQ(metadata.size(), 1U);
|
|
|
|
auto moved_file_name = metadata[0].name;
|
|
|
|
|
|
|
|
// Create two more 1MB sst files
|
|
|
|
for (int i = 0; i < 2; ++i) {
|
|
|
|
// Create 1MB sst file
|
|
|
|
for (int j = 0; j < 100; ++j) {
|
|
|
|
ASSERT_OK(Put(Key(i * 50 + j + 100), RandomString(&rnd, 10 * 1024)));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
}
|
|
|
|
// this should execute both L0->L1 and L1->L2 (merge with previous file)
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
|
|
|
|
ASSERT_EQ("0,0,2", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// iterator is holding the file
|
|
|
|
ASSERT_OK(env_->FileExists(dbname_ + moved_file_name));
|
|
|
|
|
|
|
|
listener->SetExpectedFileName(dbname_ + moved_file_name);
|
|
|
|
iterator.reset();
|
|
|
|
|
|
|
|
// this file should have been compacted away
|
|
|
|
ASSERT_NOK(env_->FileExists(dbname_ + moved_file_name));
|
|
|
|
listener->VerifyMatchedCount(1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, CompressLevelCompaction) {
|
2015-07-21 12:05:57 +02:00
|
|
|
if (!Zlib_Supported()) {
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
Options options = CurrentOptions();
|
2015-12-07 20:57:00 +01:00
|
|
|
options.memtable_factory.reset(
|
|
|
|
new SpecialSkipListFactory(KNumKeysByGenerateNewFile - 1));
|
2015-07-21 12:05:57 +02:00
|
|
|
options.compaction_style = kCompactionStyleLevel;
|
2015-09-01 08:11:12 +02:00
|
|
|
options.write_buffer_size = 110 << 10; // 110KB
|
2015-08-26 23:19:31 +02:00
|
|
|
options.arena_block_size = 4 << 10;
|
2015-07-21 12:05:57 +02:00
|
|
|
options.level0_file_num_compaction_trigger = 2;
|
|
|
|
options.num_levels = 4;
|
|
|
|
options.max_bytes_for_level_base = 400 * 1024;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
// First two levels have no compression, so that a trivial move between
|
|
|
|
// them will be allowed. Level 2 has Zlib compression so that a trivial
|
|
|
|
// move to level 3 will not be allowed
|
|
|
|
options.compression_per_level = {kNoCompression, kNoCompression,
|
|
|
|
kZlibCompression};
|
|
|
|
int matches = 0, didnt_match = 0, trivial_move = 0, non_trivial = 0;
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"Compaction::InputCompressionMatchesOutput:Matches",
|
|
|
|
[&](void* arg) { matches++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"Compaction::InputCompressionMatchesOutput:DidntMatch",
|
|
|
|
[&](void* arg) { didnt_match++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:NonTrivial",
|
|
|
|
[&](void* arg) { non_trivial++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:TrivialMove",
|
|
|
|
[&](void* arg) { trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
|
|
|
Reopen(options);
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
int key_idx = 0;
|
|
|
|
|
|
|
|
// First three 110KB files are going to level 0
|
|
|
|
// After that, (100K, 200K)
|
|
|
|
for (int num = 0; num < 3; num++) {
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Another 110KB triggers a compaction to 400K file to fill up level 0
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ(4, GetSstFileCount(dbname_));
|
|
|
|
|
|
|
|
// (1, 4)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// (1, 4, 1)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,1", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// (1, 4, 2)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,2", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// (1, 4, 3)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,3", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// (1, 4, 4)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,4", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// (1, 4, 5)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,5", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// (1, 4, 6)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,6", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// (1, 4, 7)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,7", FilesPerLevel(0));
|
|
|
|
|
|
|
|
// (1, 4, 8)
|
|
|
|
GenerateNewFile(&rnd, &key_idx);
|
|
|
|
ASSERT_EQ("1,4,8", FilesPerLevel(0));
|
|
|
|
|
|
|
|
ASSERT_EQ(matches, 12);
|
|
|
|
// Currently, the test relies on the number of calls to
|
|
|
|
// InputCompressionMatchesOutput() per compaction.
|
|
|
|
const int kCallsToInputCompressionMatch = 2;
|
|
|
|
ASSERT_EQ(didnt_match, 8 * kCallsToInputCompressionMatch);
|
|
|
|
ASSERT_EQ(trivial_move, 12);
|
|
|
|
ASSERT_EQ(non_trivial, 8);
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->DisableProcessing();
|
|
|
|
|
|
|
|
for (int i = 0; i < key_idx; i++) {
|
|
|
|
auto v = Get(Key(i));
|
|
|
|
ASSERT_NE(v, "NOT_FOUND");
|
2015-08-26 23:19:31 +02:00
|
|
|
ASSERT_TRUE(v.size() == 1 || v.size() == 990);
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
Reopen(options);
|
|
|
|
|
|
|
|
for (int i = 0; i < key_idx; i++) {
|
|
|
|
auto v = Get(Key(i));
|
|
|
|
ASSERT_NE(v, "NOT_FOUND");
|
2015-08-26 23:19:31 +02:00
|
|
|
ASSERT_TRUE(v.size() == 1 || v.size() == 990);
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
Destroy(options);
|
|
|
|
}
|
|
|
|
|
Add options.base_background_compactions as a number of compaction threads for low compaction debt
Summary:
If options.base_background_compactions is given, we try to schedule number of compactions not existing this number, only when L0 files increase to certain number, or pending compaction bytes more than certain threshold, we schedule compactions based on options.max_background_compactions.
The watermarks are calculated based on slowdown thresholds.
Test Plan:
Add new test cases in column_family_test.
Adding more unit tests.
Reviewers: IslamAbdelRahman, yhchiang, kradhakrishnan, rven, anthony
Reviewed By: anthony
Subscribers: leveldb, dhruba, yoshinorim
Differential Revision: https://reviews.facebook.net/D53409
2016-01-28 20:56:16 +01:00
|
|
|
TEST_F(DBCompactionTest, SanitizeCompactionOptionsTest) {
|
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.max_background_compactions = 5;
|
|
|
|
options.soft_pending_compaction_bytes_limit = 0;
|
|
|
|
options.hard_pending_compaction_bytes_limit = 100;
|
|
|
|
options.create_if_missing = true;
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
ASSERT_EQ(5, db_->GetOptions().base_background_compactions);
|
|
|
|
ASSERT_EQ(100, db_->GetOptions().soft_pending_compaction_bytes_limit);
|
|
|
|
|
|
|
|
options.base_background_compactions = 4;
|
|
|
|
options.max_background_compactions = 3;
|
|
|
|
options.soft_pending_compaction_bytes_limit = 200;
|
|
|
|
options.hard_pending_compaction_bytes_limit = 150;
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
ASSERT_EQ(3, db_->GetOptions().base_background_compactions);
|
|
|
|
ASSERT_EQ(150, db_->GetOptions().soft_pending_compaction_bytes_limit);
|
|
|
|
}
|
|
|
|
|
2015-07-21 12:05:57 +02:00
|
|
|
// This tests for a bug that could cause two level0 compactions running
|
|
|
|
// concurrently
|
2015-09-10 23:37:00 +02:00
|
|
|
// TODO(aekmekji): Make sure that the reason this fails when run with
|
|
|
|
// max_subcompactions > 1 is not a correctness issue but just inherent to
|
|
|
|
// running parallel L0-L1 compactions
|
|
|
|
TEST_F(DBCompactionTest, SuggestCompactRangeNoTwoLevel0Compactions) {
|
2015-07-21 12:05:57 +02:00
|
|
|
Options options = CurrentOptions();
|
|
|
|
options.compaction_style = kCompactionStyleLevel;
|
|
|
|
options.write_buffer_size = 110 << 10;
|
2015-08-26 23:19:31 +02:00
|
|
|
options.arena_block_size = 4 << 10;
|
2015-07-21 12:05:57 +02:00
|
|
|
options.level0_file_num_compaction_trigger = 4;
|
|
|
|
options.num_levels = 4;
|
|
|
|
options.compression = kNoCompression;
|
|
|
|
options.max_bytes_for_level_base = 450 << 10;
|
|
|
|
options.target_file_size_base = 98 << 10;
|
|
|
|
options.max_write_buffer_number = 2;
|
|
|
|
options.max_background_compactions = 2;
|
|
|
|
|
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
// fill up the DB
|
|
|
|
Random rnd(301);
|
|
|
|
for (int num = 0; num < 10; num++) {
|
|
|
|
GenerateNewRandomFile(&rnd);
|
|
|
|
}
|
|
|
|
db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->LoadDependency(
|
|
|
|
{{"CompactionJob::Run():Start",
|
|
|
|
"DBCompactionTest::SuggestCompactRangeNoTwoLevel0Compactions:1"},
|
|
|
|
{"DBCompactionTest::SuggestCompactRangeNoTwoLevel0Compactions:2",
|
|
|
|
"CompactionJob::Run():End"}});
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
|
|
|
// trigger L0 compaction
|
|
|
|
for (int num = 0; num < options.level0_file_num_compaction_trigger + 1;
|
|
|
|
num++) {
|
|
|
|
GenerateNewRandomFile(&rnd, /* nowait */ true);
|
2015-08-04 00:52:52 +02:00
|
|
|
ASSERT_OK(Flush());
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
TEST_SYNC_POINT(
|
|
|
|
"DBCompactionTest::SuggestCompactRangeNoTwoLevel0Compactions:1");
|
|
|
|
|
|
|
|
GenerateNewRandomFile(&rnd, /* nowait */ true);
|
|
|
|
dbfull()->TEST_WaitForFlushMemTable();
|
|
|
|
ASSERT_OK(experimental::SuggestCompactRange(db_, nullptr, nullptr));
|
|
|
|
for (int num = 0; num < options.level0_file_num_compaction_trigger + 1;
|
|
|
|
num++) {
|
|
|
|
GenerateNewRandomFile(&rnd, /* nowait */ true);
|
2015-08-04 00:52:52 +02:00
|
|
|
ASSERT_OK(Flush());
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
TEST_SYNC_POINT(
|
|
|
|
"DBCompactionTest::SuggestCompactRangeNoTwoLevel0Compactions:2");
|
2015-08-04 00:52:52 +02:00
|
|
|
dbfull()->TEST_WaitForCompact();
|
2015-07-21 12:05:57 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
|
2015-07-21 12:05:57 +02:00
|
|
|
int32_t trivial_move = 0;
|
|
|
|
int32_t non_trivial_move = 0;
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:TrivialMove",
|
|
|
|
[&](void* arg) { trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->SetCallBack(
|
|
|
|
"DBImpl::BackgroundCompaction:NonTrivial",
|
|
|
|
[&](void* arg) { non_trivial_move++; });
|
|
|
|
rocksdb::SyncPoint::GetInstance()->EnableProcessing();
|
|
|
|
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-07-21 12:05:57 +02:00
|
|
|
options.write_buffer_size = 100000000;
|
2015-08-21 23:25:34 +02:00
|
|
|
options.max_subcompactions = max_subcompactions_;
|
2015-07-21 12:05:57 +02:00
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
int32_t value_size = 10 * 1024; // 10 KB
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
std::vector<std::string> values;
|
|
|
|
// File with keys [ 0 => 99 ]
|
|
|
|
for (int i = 0; i < 100; i++) {
|
|
|
|
values.push_back(RandomString(&rnd, value_size));
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
ASSERT_EQ("1", FilesPerLevel(0));
|
|
|
|
// Compaction will do L0=>L1 (trivial move) then move L1 files to L3
|
|
|
|
CompactRangeOptions compact_options;
|
|
|
|
compact_options.change_level = true;
|
|
|
|
compact_options.target_level = 3;
|
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
|
|
|
|
ASSERT_EQ("0,0,0,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(trivial_move, 1);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
// File with keys [ 100 => 199 ]
|
|
|
|
for (int i = 100; i < 200; i++) {
|
|
|
|
values.push_back(RandomString(&rnd, value_size));
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
ASSERT_EQ("1,0,0,1", FilesPerLevel(0));
|
|
|
|
// Compaction will do L0=>L1 L1=>L2 L2=>L3 (3 trivial moves)
|
|
|
|
// then compacte the bottommost level L3=>L3 (non trivial move)
|
|
|
|
compact_options = CompactRangeOptions();
|
|
|
|
compact_options.bottommost_level_compaction =
|
|
|
|
BottommostLevelCompaction::kForce;
|
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
|
|
|
|
ASSERT_EQ("0,0,0,1", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(trivial_move, 4);
|
|
|
|
ASSERT_EQ(non_trivial_move, 1);
|
|
|
|
|
|
|
|
// File with keys [ 200 => 299 ]
|
|
|
|
for (int i = 200; i < 300; i++) {
|
|
|
|
values.push_back(RandomString(&rnd, value_size));
|
|
|
|
ASSERT_OK(Put(Key(i), values[i]));
|
|
|
|
}
|
|
|
|
ASSERT_OK(Flush());
|
|
|
|
|
|
|
|
ASSERT_EQ("1,0,0,1", FilesPerLevel(0));
|
|
|
|
trivial_move = 0;
|
|
|
|
non_trivial_move = 0;
|
|
|
|
compact_options = CompactRangeOptions();
|
|
|
|
compact_options.bottommost_level_compaction =
|
|
|
|
BottommostLevelCompaction::kSkip;
|
|
|
|
// Compaction will do L0=>L1 L1=>L2 L2=>L3 (3 trivial moves)
|
|
|
|
// and will skip bottommost level compaction
|
|
|
|
ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
|
|
|
|
ASSERT_EQ("0,0,0,2", FilesPerLevel(0));
|
|
|
|
ASSERT_EQ(trivial_move, 3);
|
|
|
|
ASSERT_EQ(non_trivial_move, 0);
|
|
|
|
|
|
|
|
for (int i = 0; i < 300; i++) {
|
|
|
|
ASSERT_EQ(Get(Key(i)), values[i]);
|
|
|
|
}
|
|
|
|
|
|
|
|
rocksdb::SyncPoint::GetInstance()->DisableProcessing();
|
|
|
|
}
|
|
|
|
|
2015-08-05 07:19:07 +02:00
|
|
|
INSTANTIATE_TEST_CASE_P(DBCompactionTestWithParam, DBCompactionTestWithParam,
|
Running manual compactions in parallel with other automatic or manual compactions in restricted cases
Summary:
This diff provides a framework for doing manual
compactions in parallel with other compactions. We now have a deque of manual compactions. We also pass manual compactions as an argument from RunManualCompactions down to
BackgroundCompactions, so that RunManualCompactions can be reentrant.
Parallelism is controlled by the two routines
ConflictingManualCompaction to allow/disallow new parallel/manual
compactions based on already existing ManualCompactions. In this diff, by default manual compactions still have to run exclusive of other compactions. However, by setting the compaction option, exclusive_manual_compaction to false, it is possible to run other compactions in parallel with a manual compaction. However, we are still restricted to one manual compaction per column family at a time. All of these restrictions will be relaxed in future diffs.
I will be adding more tests later.
Test Plan: Rocksdb regression + new tests + valgrind
Reviewers: igor, anthony, IslamAbdelRahman, kradhakrishnan, yhchiang, sdong
Reviewed By: sdong
Subscribers: yoshinorim, dhruba, leveldb
Differential Revision: https://reviews.facebook.net/D47973
2015-12-14 20:20:34 +01:00
|
|
|
::testing::Values(std::make_tuple(1, true),
|
|
|
|
std::make_tuple(1, false),
|
|
|
|
std::make_tuple(4, true),
|
|
|
|
std::make_tuple(4, false)));
|
2015-12-05 02:03:55 +01:00
|
|
|
|
|
|
|
class CompactionPriTest : public DBTestBase,
|
|
|
|
public testing::WithParamInterface<uint32_t> {
|
|
|
|
public:
|
|
|
|
CompactionPriTest() : DBTestBase("/compaction_pri_test") {
|
|
|
|
compaction_pri_ = GetParam();
|
|
|
|
}
|
|
|
|
|
|
|
|
// Required if inheriting from testing::WithParamInterface<>
|
|
|
|
static void SetUpTestCase() {}
|
|
|
|
static void TearDownTestCase() {}
|
|
|
|
|
|
|
|
uint32_t compaction_pri_;
|
|
|
|
};
|
|
|
|
|
|
|
|
TEST_P(CompactionPriTest, Test) {
|
2016-03-24 20:45:50 +01:00
|
|
|
Options options = CurrentOptions();
|
2015-12-05 02:03:55 +01:00
|
|
|
options.write_buffer_size = 16 * 1024;
|
|
|
|
options.compaction_pri = static_cast<CompactionPri>(compaction_pri_);
|
|
|
|
options.hard_pending_compaction_bytes_limit = 256 * 1024;
|
|
|
|
options.max_bytes_for_level_base = 64 * 1024;
|
|
|
|
options.max_bytes_for_level_multiplier = 4;
|
|
|
|
options.compression = kNoCompression;
|
2016-03-24 20:45:50 +01:00
|
|
|
|
2015-12-05 02:03:55 +01:00
|
|
|
DestroyAndReopen(options);
|
|
|
|
|
|
|
|
Random rnd(301);
|
|
|
|
const int kNKeys = 5000;
|
|
|
|
int keys[kNKeys];
|
|
|
|
for (int i = 0; i < kNKeys; i++) {
|
|
|
|
keys[i] = i;
|
|
|
|
}
|
|
|
|
std::random_shuffle(std::begin(keys), std::end(keys));
|
|
|
|
|
|
|
|
for (int i = 0; i < kNKeys; i++) {
|
|
|
|
ASSERT_OK(Put(Key(keys[i]), RandomString(&rnd, 102)));
|
|
|
|
}
|
|
|
|
|
|
|
|
dbfull()->TEST_WaitForCompact();
|
|
|
|
for (int i = 0; i < kNKeys; i++) {
|
|
|
|
ASSERT_NE("NOT_FOUND", Get(Key(i)));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-02-11 22:45:53 +01:00
|
|
|
INSTANTIATE_TEST_CASE_P(
|
|
|
|
CompactionPriTest, CompactionPriTest,
|
|
|
|
::testing::Values(CompactionPri::kByCompensatedSize,
|
|
|
|
CompactionPri::kOldestLargestSeqFirst,
|
|
|
|
CompactionPri::kOldestSmallestSeqFirst,
|
|
|
|
CompactionPri::kMinOverlappingRatio));
|
2015-12-05 02:03:55 +01:00
|
|
|
|
2015-10-23 00:15:37 +02:00
|
|
|
#endif // !defined(ROCKSDB_LITE)
|
2015-07-21 12:05:57 +02:00
|
|
|
} // namespace rocksdb
|
|
|
|
|
|
|
|
int main(int argc, char** argv) {
|
2015-10-23 00:15:37 +02:00
|
|
|
#if !defined(ROCKSDB_LITE)
|
2015-07-21 12:05:57 +02:00
|
|
|
rocksdb::port::InstallStackTraceHandler();
|
|
|
|
::testing::InitGoogleTest(&argc, argv);
|
|
|
|
return RUN_ALL_TESTS();
|
|
|
|
#else
|
|
|
|
return 0;
|
|
|
|
#endif
|
|
|
|
}
|