rocksdb/db/perf_context_test.cc

690 lines
22 KiB
C++
Raw Permalink Normal View History

// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree. An additional grant
// of patent rights can be found in the PATENTS file in the same directory.
//
#include <algorithm>
#include <iostream>
#include <thread>
#include <vector>
#include "rocksdb/db.h"
#include "rocksdb/memtablerep.h"
#include "rocksdb/perf_context.h"
2014-03-14 21:02:20 +01:00
#include "rocksdb/slice_transform.h"
#include "util/histogram.h"
#include "util/instrumented_mutex.h"
#include "util/stop_watch.h"
#include "util/string_util.h"
#include "util/testharness.h"
#include "util/thread_status_util.h"
#include "utilities/merge_operators.h"
bool FLAGS_random_key = false;
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
bool FLAGS_use_set_based_memetable = false;
int FLAGS_total_keys = 100;
int FLAGS_write_buffer_size = 1000000000;
int FLAGS_max_write_buffer_number = 8;
int FLAGS_min_write_buffer_number_to_merge = 7;
bool FLAGS_verbose = false;
// Path to the database on file system
const std::string kDbName = rocksdb::test::TmpDir() + "/perf_context_test";
namespace rocksdb {
std::shared_ptr<DB> OpenDb(bool read_only = false) {
DB* db;
Options options;
options.create_if_missing = true;
options.max_open_files = -1;
options.write_buffer_size = FLAGS_write_buffer_size;
options.max_write_buffer_number = FLAGS_max_write_buffer_number;
options.min_write_buffer_number_to_merge =
FLAGS_min_write_buffer_number_to_merge;
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
if (FLAGS_use_set_based_memetable) {
#ifndef ROCKSDB_LITE
options.prefix_extractor.reset(rocksdb::NewFixedPrefixTransform(0));
options.memtable_factory.reset(NewHashSkipListRepFactory());
#endif // ROCKSDB_LITE
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
}
Status s;
if (!read_only) {
s = DB::Open(options, kDbName, &db);
} else {
s = DB::OpenForReadOnly(options, kDbName, &db);
}
rocksdb: Replace ASSERT* with EXPECT* in functions that does not return void value Summary: gtest does not use exceptions to fail a unit test by design, and `ASSERT*`s are implemented using `return`. As a consequence we cannot use `ASSERT*` in a function that does not return `void` value ([[ https://code.google.com/p/googletest/wiki/AdvancedGuide#Assertion_Placement | 1]]), and have to fix our existing code. This diff does this in a generic way, with no manual changes. In order to detect all existing `ASSERT*` that are used in functions that doesn't return void value, I change the code to generate compile errors for such cases. In `util/testharness.h` I defined `EXPECT*` assertions, the same way as `ASSERT*`, and redefined `ASSERT*` to return `void`. Then executed: ```lang=bash % USE_CLANG=1 make all -j55 -k 2> build.log % perl -naF: -e 'print "-- -number=".$F[1]." ".$F[0]."\n" if /: error:/' \ build.log | xargs -L 1 perl -spi -e 's/ASSERT/EXPECT/g if $. == $number' % make format ``` After that I reverted back change to `ASSERT*` in `util/testharness.h`. But preserved introduced `EXPECT*`, which is the same as `ASSERT*`. This will be deleted once switched to gtest. This diff is independent and contains manual changes only in `util/testharness.h`. Test Plan: Make sure all tests are passing. ```lang=bash % USE_CLANG=1 make check ``` Reviewers: igor, lgalanis, sdong, yufei.zhu, rven, meyering Reviewed By: meyering Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D33333
2015-03-17 04:52:32 +01:00
EXPECT_OK(s);
return std::shared_ptr<DB>(db);
}
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 22:08:00 +01:00
class PerfContextTest : public testing::Test {};
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 22:08:00 +01:00
TEST_F(PerfContextTest, SeekIntoDeletion) {
DestroyDB(kDbName, Options());
auto db = OpenDb();
WriteOptions write_options;
ReadOptions read_options;
for (int i = 0; i < FLAGS_total_keys; ++i) {
std::string key = "k" + ToString(i);
std::string value = "v" + ToString(i);
db->Put(write_options, key, value);
}
for (int i = 0; i < FLAGS_total_keys -1 ; ++i) {
std::string key = "k" + ToString(i);
db->Delete(write_options, key);
}
HistogramImpl hist_get;
HistogramImpl hist_get_time;
for (int i = 0; i < FLAGS_total_keys - 1; ++i) {
std::string key = "k" + ToString(i);
std::string value;
perf_context.Reset();
StopWatchNano timer(Env::Default());
timer.Start();
auto status = db->Get(read_options, key, &value);
auto elapsed_nanos = timer.ElapsedNanos();
ASSERT_TRUE(status.IsNotFound());
hist_get.Add(perf_context.user_key_comparison_count);
hist_get_time.Add(elapsed_nanos);
}
if (FLAGS_verbose) {
std::cout << "Get user key comparison: \n" << hist_get.ToString()
<< "Get time: \n" << hist_get_time.ToString();
}
{
HistogramImpl hist_seek_to_first;
std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
perf_context.Reset();
StopWatchNano timer(Env::Default(), true);
iter->SeekToFirst();
hist_seek_to_first.Add(perf_context.user_key_comparison_count);
auto elapsed_nanos = timer.ElapsedNanos();
if (FLAGS_verbose) {
std::cout << "SeekToFirst uesr key comparison: \n"
<< hist_seek_to_first.ToString()
<< "ikey skipped: " << perf_context.internal_key_skipped_count
<< "\n"
<< "idelete skipped: "
<< perf_context.internal_delete_skipped_count << "\n"
<< "elapsed: " << elapsed_nanos << "\n";
}
}
HistogramImpl hist_seek;
for (int i = 0; i < FLAGS_total_keys; ++i) {
std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
std::string key = "k" + ToString(i);
perf_context.Reset();
StopWatchNano timer(Env::Default(), true);
iter->Seek(key);
auto elapsed_nanos = timer.ElapsedNanos();
hist_seek.Add(perf_context.user_key_comparison_count);
if (FLAGS_verbose) {
std::cout << "seek cmp: " << perf_context.user_key_comparison_count
<< " ikey skipped " << perf_context.internal_key_skipped_count
<< " idelete skipped "
<< perf_context.internal_delete_skipped_count
<< " elapsed: " << elapsed_nanos << "ns\n";
}
perf_context.Reset();
ASSERT_TRUE(iter->Valid());
StopWatchNano timer2(Env::Default(), true);
iter->Next();
auto elapsed_nanos2 = timer2.ElapsedNanos();
if (FLAGS_verbose) {
std::cout << "next cmp: " << perf_context.user_key_comparison_count
<< "elapsed: " << elapsed_nanos2 << "ns\n";
}
}
if (FLAGS_verbose) {
std::cout << "Seek uesr key comparison: \n" << hist_seek.ToString();
}
}
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 22:08:00 +01:00
TEST_F(PerfContextTest, StopWatchNanoOverhead) {
// profile the timer cost by itself!
const int kTotalIterations = 1000000;
std::vector<uint64_t> timings(kTotalIterations);
StopWatchNano timer(Env::Default(), true);
for (auto& timing : timings) {
timing = timer.ElapsedNanos(true /* reset */);
}
HistogramImpl histogram;
for (const auto timing : timings) {
histogram.Add(timing);
}
if (FLAGS_verbose) {
std::cout << histogram.ToString();
}
}
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 22:08:00 +01:00
TEST_F(PerfContextTest, StopWatchOverhead) {
// profile the timer cost by itself!
const int kTotalIterations = 1000000;
uint64_t elapsed = 0;
std::vector<uint64_t> timings(kTotalIterations);
StopWatch timer(Env::Default(), nullptr, 0, &elapsed);
for (auto& timing : timings) {
timing = elapsed;
}
HistogramImpl histogram;
uint64_t prev_timing = 0;
for (const auto timing : timings) {
histogram.Add(timing - prev_timing);
prev_timing = timing;
}
if (FLAGS_verbose) {
std::cout << histogram.ToString();
}
}
void ProfileQueries(bool enabled_time = false) {
DestroyDB(kDbName, Options()); // Start this test with a fresh DB
auto db = OpenDb();
WriteOptions write_options;
ReadOptions read_options;
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
HistogramImpl hist_put;
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
HistogramImpl hist_get;
HistogramImpl hist_get_snapshot;
HistogramImpl hist_get_memtable;
HistogramImpl hist_get_files;
HistogramImpl hist_get_post_process;
HistogramImpl hist_num_memtable_checked;
HistogramImpl hist_mget;
HistogramImpl hist_mget_snapshot;
HistogramImpl hist_mget_memtable;
HistogramImpl hist_mget_files;
HistogramImpl hist_mget_post_process;
HistogramImpl hist_mget_num_memtable_checked;
HistogramImpl hist_write_pre_post;
HistogramImpl hist_write_wal_time;
HistogramImpl hist_write_memtable_time;
uint64_t total_db_mutex_nanos = 0;
if (FLAGS_verbose) {
std::cout << "Inserting " << FLAGS_total_keys << " key/value pairs\n...\n";
}
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
std::vector<int> keys;
const int kFlushFlag = -1;
for (int i = 0; i < FLAGS_total_keys; ++i) {
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
keys.push_back(i);
if (i == FLAGS_total_keys / 2) {
// Issuing a flush in the middle.
keys.push_back(kFlushFlag);
}
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
}
if (FLAGS_random_key) {
std::random_shuffle(keys.begin(), keys.end());
}
#ifndef NDEBUG
ThreadStatusUtil::TEST_SetStateDelay(ThreadStatus::STATE_MUTEX_WAIT, 1U);
#endif
int num_mutex_waited = 0;
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
for (const int i : keys) {
if (i == kFlushFlag) {
FlushOptions fo;
db->Flush(fo);
continue;
}
std::string key = "k" + ToString(i);
std::string value = "v" + ToString(i);
std::vector<std::string> values;
perf_context.Reset();
db->Put(write_options, key, value);
if (++num_mutex_waited > 3) {
#ifndef NDEBUG
ThreadStatusUtil::TEST_SetStateDelay(ThreadStatus::STATE_MUTEX_WAIT, 0U);
#endif
}
hist_write_pre_post.Add(perf_context.write_pre_and_post_process_time);
hist_write_wal_time.Add(perf_context.write_wal_time);
hist_write_memtable_time.Add(perf_context.write_memtable_time);
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
hist_put.Add(perf_context.user_key_comparison_count);
total_db_mutex_nanos += perf_context.db_mutex_lock_nanos;
}
#ifndef NDEBUG
ThreadStatusUtil::TEST_SetStateDelay(ThreadStatus::STATE_MUTEX_WAIT, 0U);
#endif
for (const int i : keys) {
if (i == kFlushFlag) {
continue;
}
std::string key = "k" + ToString(i);
std::string expected_value = "v" + ToString(i);
std::string value;
std::vector<Slice> multiget_keys = {Slice(key)};
std::vector<std::string> values;
perf_context.Reset();
ASSERT_OK(db->Get(read_options, key, &value));
ASSERT_EQ(expected_value, value);
hist_get_snapshot.Add(perf_context.get_snapshot_time);
hist_get_memtable.Add(perf_context.get_from_memtable_time);
hist_get_files.Add(perf_context.get_from_output_files_time);
hist_num_memtable_checked.Add(perf_context.get_from_memtable_count);
hist_get_post_process.Add(perf_context.get_post_process_time);
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
hist_get.Add(perf_context.user_key_comparison_count);
perf_context.Reset();
db->MultiGet(read_options, multiget_keys, &values);
hist_mget_snapshot.Add(perf_context.get_snapshot_time);
hist_mget_memtable.Add(perf_context.get_from_memtable_time);
hist_mget_files.Add(perf_context.get_from_output_files_time);
hist_mget_num_memtable_checked.Add(perf_context.get_from_memtable_count);
hist_mget_post_process.Add(perf_context.get_post_process_time);
hist_mget.Add(perf_context.user_key_comparison_count);
}
if (FLAGS_verbose) {
std::cout << "Put uesr key comparison: \n" << hist_put.ToString()
<< "Get uesr key comparison: \n" << hist_get.ToString()
<< "MultiGet uesr key comparison: \n" << hist_get.ToString();
std::cout << "Put(): Pre and Post Process Time: \n"
<< hist_write_pre_post.ToString() << " Writing WAL time: \n"
<< hist_write_wal_time.ToString() << "\n"
<< " Writing Mem Table time: \n"
<< hist_write_memtable_time.ToString() << "\n"
<< " Total DB mutex nanos: \n" << total_db_mutex_nanos << "\n";
std::cout << "Get(): Time to get snapshot: \n"
<< hist_get_snapshot.ToString()
<< " Time to get value from memtables: \n"
<< hist_get_memtable.ToString() << "\n"
<< " Time to get value from output files: \n"
<< hist_get_files.ToString() << "\n"
<< " Number of memtables checked: \n"
<< hist_num_memtable_checked.ToString() << "\n"
<< " Time to post process: \n" << hist_get_post_process.ToString()
<< "\n";
std::cout << "MultiGet(): Time to get snapshot: \n"
<< hist_mget_snapshot.ToString()
<< " Time to get value from memtables: \n"
<< hist_mget_memtable.ToString() << "\n"
<< " Time to get value from output files: \n"
<< hist_mget_files.ToString() << "\n"
<< " Number of memtables checked: \n"
<< hist_mget_num_memtable_checked.ToString() << "\n"
<< " Time to post process: \n"
<< hist_mget_post_process.ToString() << "\n";
}
if (enabled_time) {
ASSERT_GT(hist_get.Average(), 0);
ASSERT_GT(hist_get_snapshot.Average(), 0);
ASSERT_GT(hist_get_memtable.Average(), 0);
ASSERT_GT(hist_get_files.Average(), 0);
ASSERT_GT(hist_get_post_process.Average(), 0);
ASSERT_GT(hist_num_memtable_checked.Average(), 0);
ASSERT_GT(hist_mget.Average(), 0);
ASSERT_GT(hist_mget_snapshot.Average(), 0);
ASSERT_GT(hist_mget_memtable.Average(), 0);
ASSERT_GT(hist_mget_files.Average(), 0);
ASSERT_GT(hist_mget_post_process.Average(), 0);
ASSERT_GT(hist_mget_num_memtable_checked.Average(), 0);
#ifndef NDEBUG
ASSERT_GT(total_db_mutex_nanos, 2000U);
#endif
}
db.reset();
db = OpenDb(true);
hist_get.Clear();
hist_get_snapshot.Clear();
hist_get_memtable.Clear();
hist_get_files.Clear();
hist_get_post_process.Clear();
hist_num_memtable_checked.Clear();
hist_mget.Clear();
hist_mget_snapshot.Clear();
hist_mget_memtable.Clear();
hist_mget_files.Clear();
hist_mget_post_process.Clear();
hist_mget_num_memtable_checked.Clear();
for (const int i : keys) {
if (i == kFlushFlag) {
continue;
}
std::string key = "k" + ToString(i);
std::string expected_value = "v" + ToString(i);
std::string value;
std::vector<Slice> multiget_keys = {Slice(key)};
std::vector<std::string> values;
perf_context.Reset();
ASSERT_OK(db->Get(read_options, key, &value));
ASSERT_EQ(expected_value, value);
hist_get_snapshot.Add(perf_context.get_snapshot_time);
hist_get_memtable.Add(perf_context.get_from_memtable_time);
hist_get_files.Add(perf_context.get_from_output_files_time);
hist_num_memtable_checked.Add(perf_context.get_from_memtable_count);
hist_get_post_process.Add(perf_context.get_post_process_time);
hist_get.Add(perf_context.user_key_comparison_count);
perf_context.Reset();
db->MultiGet(read_options, multiget_keys, &values);
hist_mget_snapshot.Add(perf_context.get_snapshot_time);
hist_mget_memtable.Add(perf_context.get_from_memtable_time);
hist_mget_files.Add(perf_context.get_from_output_files_time);
hist_mget_num_memtable_checked.Add(perf_context.get_from_memtable_count);
hist_mget_post_process.Add(perf_context.get_post_process_time);
hist_mget.Add(perf_context.user_key_comparison_count);
}
if (FLAGS_verbose) {
std::cout << "ReadOnly Get uesr key comparison: \n" << hist_get.ToString()
<< "ReadOnly MultiGet uesr key comparison: \n"
<< hist_mget.ToString();
std::cout << "ReadOnly Get(): Time to get snapshot: \n"
<< hist_get_snapshot.ToString()
<< " Time to get value from memtables: \n"
<< hist_get_memtable.ToString() << "\n"
<< " Time to get value from output files: \n"
<< hist_get_files.ToString() << "\n"
<< " Number of memtables checked: \n"
<< hist_num_memtable_checked.ToString() << "\n"
<< " Time to post process: \n" << hist_get_post_process.ToString()
<< "\n";
std::cout << "ReadOnly MultiGet(): Time to get snapshot: \n"
<< hist_mget_snapshot.ToString()
<< " Time to get value from memtables: \n"
<< hist_mget_memtable.ToString() << "\n"
<< " Time to get value from output files: \n"
<< hist_mget_files.ToString() << "\n"
<< " Number of memtables checked: \n"
<< hist_mget_num_memtable_checked.ToString() << "\n"
<< " Time to post process: \n"
<< hist_mget_post_process.ToString() << "\n";
}
if (enabled_time) {
ASSERT_GT(hist_get.Average(), 0);
ASSERT_GT(hist_get_memtable.Average(), 0);
ASSERT_GT(hist_get_files.Average(), 0);
ASSERT_GT(hist_num_memtable_checked.Average(), 0);
// In read-only mode Get(), no super version operation is needed
ASSERT_EQ(hist_get_post_process.Average(), 0);
ASSERT_EQ(hist_get_snapshot.Average(), 0);
ASSERT_GT(hist_mget.Average(), 0);
ASSERT_GT(hist_mget_snapshot.Average(), 0);
ASSERT_GT(hist_mget_memtable.Average(), 0);
ASSERT_GT(hist_mget_files.Average(), 0);
ASSERT_GT(hist_mget_post_process.Average(), 0);
ASSERT_GT(hist_mget_num_memtable_checked.Average(), 0);
}
}
#ifndef ROCKSDB_LITE
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 22:08:00 +01:00
TEST_F(PerfContextTest, KeyComparisonCount) {
SetPerfLevel(kEnableCount);
ProfileQueries();
SetPerfLevel(kDisable);
ProfileQueries();
SetPerfLevel(kEnableTime);
ProfileQueries(true);
}
#endif // ROCKSDB_LITE
// make perf_context_test
// export ROCKSDB_TESTS=PerfContextTest.SeekKeyComparison
// For one memtable:
// ./perf_context_test --write_buffer_size=500000 --total_keys=10000
// For two memtables:
// ./perf_context_test --write_buffer_size=250000 --total_keys=10000
// Specify --random_key=1 to shuffle the key before insertion
// Results show that, for sequential insertion, worst-case Seek Key comparison
// is close to the total number of keys (linear), when there is only one
// memtable. When there are two memtables, even the avg Seek Key comparison
// starts to become linear to the input size.
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 22:08:00 +01:00
TEST_F(PerfContextTest, SeekKeyComparison) {
DestroyDB(kDbName, Options());
auto db = OpenDb();
WriteOptions write_options;
ReadOptions read_options;
if (FLAGS_verbose) {
std::cout << "Inserting " << FLAGS_total_keys << " key/value pairs\n...\n";
}
std::vector<int> keys;
for (int i = 0; i < FLAGS_total_keys; ++i) {
keys.push_back(i);
}
if (FLAGS_random_key) {
std::random_shuffle(keys.begin(), keys.end());
}
HistogramImpl hist_put_time;
HistogramImpl hist_wal_time;
HistogramImpl hist_time_diff;
SetPerfLevel(kEnableTime);
StopWatchNano timer(Env::Default());
for (const int i : keys) {
std::string key = "k" + ToString(i);
std::string value = "v" + ToString(i);
perf_context.Reset();
timer.Start();
db->Put(write_options, key, value);
auto put_time = timer.ElapsedNanos();
hist_put_time.Add(put_time);
hist_wal_time.Add(perf_context.write_wal_time);
hist_time_diff.Add(put_time - perf_context.write_wal_time);
}
if (FLAGS_verbose) {
std::cout << "Put time:\n" << hist_put_time.ToString() << "WAL time:\n"
<< hist_wal_time.ToString() << "time diff:\n"
<< hist_time_diff.ToString();
}
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
HistogramImpl hist_seek;
HistogramImpl hist_next;
for (int i = 0; i < FLAGS_total_keys; ++i) {
std::string key = "k" + ToString(i);
std::string value = "v" + ToString(i);
std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
perf_context.Reset();
iter->Seek(key);
ASSERT_TRUE(iter->Valid());
ASSERT_EQ(iter->value().ToString(), value);
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
hist_seek.Add(perf_context.user_key_comparison_count);
}
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
std::unique_ptr<Iterator> iter(db->NewIterator(read_options));
for (iter->SeekToFirst(); iter->Valid();) {
perf_context.Reset();
iter->Next();
hist_next.Add(perf_context.user_key_comparison_count);
}
if (FLAGS_verbose) {
std::cout << "Seek:\n" << hist_seek.ToString() << "Next:\n"
<< hist_next.ToString();
}
}
TEST_F(PerfContextTest, DBMutexLockCounter) {
int stats_code[] = {0, static_cast<int>(DB_MUTEX_WAIT_MICROS)};
for (PerfLevel perf_level :
{PerfLevel::kEnableTimeExceptForMutex, PerfLevel::kEnableTime}) {
for (int c = 0; c < 2; ++c) {
InstrumentedMutex mutex(nullptr, Env::Default(), stats_code[c]);
mutex.Lock();
std::thread child_thread([&] {
SetPerfLevel(perf_level);
perf_context.Reset();
ASSERT_EQ(perf_context.db_mutex_lock_nanos, 0);
mutex.Lock();
mutex.Unlock();
if (perf_level == PerfLevel::kEnableTimeExceptForMutex ||
stats_code[c] != DB_MUTEX_WAIT_MICROS) {
ASSERT_EQ(perf_context.db_mutex_lock_nanos, 0);
} else {
// increment the counter only when it's a DB Mutex
ASSERT_GT(perf_context.db_mutex_lock_nanos, 0);
}
});
Env::Default()->SleepForMicroseconds(100);
mutex.Unlock();
child_thread.join();
}
}
}
TEST_F(PerfContextTest, FalseDBMutexWait) {
SetPerfLevel(kEnableTime);
int stats_code[] = {0, static_cast<int>(DB_MUTEX_WAIT_MICROS)};
for (int c = 0; c < 2; ++c) {
InstrumentedMutex mutex(nullptr, Env::Default(), stats_code[c]);
InstrumentedCondVar lock(&mutex);
perf_context.Reset();
mutex.Lock();
lock.TimedWait(100);
mutex.Unlock();
if (stats_code[c] == static_cast<int>(DB_MUTEX_WAIT_MICROS)) {
// increment the counter only when it's a DB Mutex
ASSERT_GT(perf_context.db_condition_wait_nanos, 0);
} else {
ASSERT_EQ(perf_context.db_condition_wait_nanos, 0);
}
}
}
TEST_F(PerfContextTest, ToString) {
perf_context.Reset();
perf_context.block_read_count = 12345;
std::string zero_included = perf_context.ToString();
ASSERT_NE(std::string::npos, zero_included.find("= 0"));
ASSERT_NE(std::string::npos, zero_included.find("= 12345"));
std::string zero_excluded = perf_context.ToString(true);
ASSERT_EQ(std::string::npos, zero_excluded.find("= 0"));
ASSERT_NE(std::string::npos, zero_excluded.find("= 12345"));
}
TEST_F(PerfContextTest, MergeOperatorTime) {
DestroyDB(kDbName, Options());
DB* db;
Options options;
options.create_if_missing = true;
options.merge_operator = MergeOperators::CreateStringAppendOperator();
Status s = DB::Open(options, kDbName, &db);
EXPECT_OK(s);
std::string val;
ASSERT_OK(db->Merge(WriteOptions(), "k1", "val1"));
ASSERT_OK(db->Merge(WriteOptions(), "k1", "val2"));
ASSERT_OK(db->Merge(WriteOptions(), "k1", "val3"));
ASSERT_OK(db->Merge(WriteOptions(), "k1", "val4"));
SetPerfLevel(kEnableTime);
perf_context.Reset();
ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
EXPECT_GT(perf_context.merge_operator_time_nanos, 0);
ASSERT_OK(db->Flush(FlushOptions()));
perf_context.Reset();
ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
EXPECT_GT(perf_context.merge_operator_time_nanos, 0);
ASSERT_OK(db->CompactRange(CompactRangeOptions(), nullptr, nullptr));
perf_context.Reset();
ASSERT_OK(db->Get(ReadOptions(), "k1", &val));
EXPECT_GT(perf_context.merge_operator_time_nanos, 0);
delete db;
}
}
int main(int argc, char** argv) {
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 22:08:00 +01:00
::testing::InitGoogleTest(&argc, argv);
for (int i = 1; i < argc; i++) {
int n;
char junk;
if (sscanf(argv[i], "--write_buffer_size=%d%c", &n, &junk) == 1) {
FLAGS_write_buffer_size = n;
}
if (sscanf(argv[i], "--total_keys=%d%c", &n, &junk) == 1) {
FLAGS_total_keys = n;
}
if (sscanf(argv[i], "--random_key=%d%c", &n, &junk) == 1 &&
(n == 0 || n == 1)) {
FLAGS_random_key = n;
}
[RocbsDB] Add an option to enable set based memtable for perf_context_test Summary: as title. Some result: -- Sequential insertion of 1M key/value with stock skip list (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=0 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 8.0179 StdDev: 176.34 Min: 0.0000 Median: 2.5555 Max: 88933.0000 Percentiles: P50: 2.56 P75: 2.83 P99: 58.21 P99.9: 133.62 P99.99: 987.50 Get uesr key comparison: Count: 1000000 Average: 43.4465 StdDev: 379.03 Min: 2.0000 Median: 36.0195 Max: 88939.0000 Percentiles: P50: 36.02 P75: 43.66 P99: 112.98 P99.9: 824.84 P99.99: 7615.38 real 0m21.345s user 0m14.723s sys 0m5.677s -- Sequential insertion of 1M key/value with set based memtable (all in on memtable) time ./perf_context_test --total_keys=1000000 --use_set_based_memetable=1 Inserting 1000000 key/value pairs ... Put uesr key comparison: Count: 1000000 Average: 61.5022 StdDev: 6.49 Min: 0.0000 Median: 62.4295 Max: 71.0000 Percentiles: P50: 62.43 P75: 66.61 P99: 71.00 P99.9: 71.00 P99.99: 71.00 Get uesr key comparison: Count: 1000000 Average: 29.3810 StdDev: 3.20 Min: 1.0000 Median: 29.1801 Max: 34.0000 Percentiles: P50: 29.18 P75: 32.06 P99: 34.00 P99.9: 34.00 P99.99: 34.00 real 0m28.875s user 0m21.699s sys 0m5.749s Worst case comparison for a Put is 88933 (skiplist) vs 71 (set based memetable) Of course, there's other in-efficiency in set based memtable implementation, which lead to the overall worst performance. However, P99 behavior advantage is very very obvious. Test Plan: ./perf_context_test and viewstate shadow testing Reviewers: dhruba Reviewed By: dhruba CC: leveldb Differential Revision: https://reviews.facebook.net/D13095
2013-09-25 23:35:01 +02:00
if (sscanf(argv[i], "--use_set_based_memetable=%d%c", &n, &junk) == 1 &&
(n == 0 || n == 1)) {
FLAGS_use_set_based_memetable = n;
}
if (sscanf(argv[i], "--verbose=%d%c", &n, &junk) == 1 &&
(n == 0 || n == 1)) {
FLAGS_verbose = n;
}
}
if (FLAGS_verbose) {
std::cout << kDbName << "\n";
}
rocksdb: switch to gtest Summary: Our existing test notation is very similar to what is used in gtest. It makes it easy to adopt what is different. In this diff I modify existing [[ https://code.google.com/p/googletest/wiki/Primer#Test_Fixtures:_Using_the_Same_Data_Configuration_for_Multiple_Te | test fixture ]] classes to inherit from `testing::Test`. Also for unit tests that use fixture class, `TEST` is replaced with `TEST_F` as required in gtest. There are several custom `main` functions in our existing tests. To make this transition easier, I modify all `main` functions to fallow gtest notation. But eventually we can remove them and use implementation of `main` that gtest provides. ```lang=bash % cat ~/transform #!/bin/sh files=$(git ls-files '*test\.cc') for file in $files do if grep -q "rocksdb::test::RunAllTests()" $file then if grep -Eq '^class \w+Test {' $file then perl -pi -e 's/^(class \w+Test) {/${1}: public testing::Test {/g' $file perl -pi -e 's/^(TEST)/${1}_F/g' $file fi perl -pi -e 's/(int main.*\{)/${1}::testing::InitGoogleTest(&argc, argv);/g' $file perl -pi -e 's/rocksdb::test::RunAllTests/RUN_ALL_TESTS/g' $file fi done % sh ~/transform % make format ``` Second iteration of this diff contains only scripted changes. Third iteration contains manual changes to fix last errors and make it compilable. Test Plan: Build and notice no errors. ```lang=bash % USE_CLANG=1 make check -j55 ``` Tests are still testing. Reviewers: meyering, sdong, rven, igor Reviewed By: igor Subscribers: dhruba, leveldb Differential Revision: https://reviews.facebook.net/D35157
2015-03-17 22:08:00 +01:00
return RUN_ALL_TESTS();
}