2016-02-10 00:12:00 +01:00
|
|
|
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
|
2017-07-16 01:03:42 +02:00
|
|
|
// This source code is licensed under both the GPLv2 (found in the
|
|
|
|
|
// COPYING file in the root directory) and Apache 2.0 License
|
|
|
|
|
// (found in the LICENSE.Apache file in the root directory).
|
2013-10-16 23:59:46 +02:00
|
|
|
//
|
2014-03-05 19:32:54 +01:00
|
|
|
|
|
|
|
|
#include <sstream>
|
2017-04-06 04:02:00 +02:00
|
|
|
#include "monitoring/perf_context_imp.h"
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
|
2013-10-04 06:49:15 +02:00
|
|
|
namespace rocksdb {
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
|
2017-06-27 00:13:35 +02:00
|
|
|
#if defined(NPERF_CONTEXT) || !defined(ROCKSDB_SUPPORT_THREAD_LOCAL)
|
2017-06-03 02:12:39 +02:00
|
|
|
PerfContext perf_context;
|
2014-04-04 22:11:44 +02:00
|
|
|
#else
|
2017-06-27 00:13:35 +02:00
|
|
|
#if defined(OS_SOLARIS)
|
|
|
|
|
__thread PerfContext perf_context_;
|
|
|
|
|
#else
|
|
|
|
|
__thread PerfContext perf_context;
|
|
|
|
|
#endif
|
2014-04-04 22:11:44 +02:00
|
|
|
#endif
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
|
2017-06-03 02:12:39 +02:00
|
|
|
PerfContext* get_perf_context() {
|
2017-06-27 00:13:35 +02:00
|
|
|
#if defined(NPERF_CONTEXT) || !defined(ROCKSDB_SUPPORT_THREAD_LOCAL)
|
2017-06-03 02:12:39 +02:00
|
|
|
return &perf_context;
|
|
|
|
|
#else
|
2017-06-27 00:13:35 +02:00
|
|
|
#if defined(OS_SOLARIS)
|
|
|
|
|
return &perf_context_;
|
|
|
|
|
#else
|
|
|
|
|
return &perf_context;
|
|
|
|
|
#endif
|
2017-06-03 02:12:39 +02:00
|
|
|
#endif
|
|
|
|
|
}
|
|
|
|
|
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
void PerfContext::Reset() {
|
2017-06-03 02:12:39 +02:00
|
|
|
#ifndef NPERF_CONTEXT
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
user_key_comparison_count = 0;
|
|
|
|
|
block_cache_hit_count = 0;
|
|
|
|
|
block_read_count = 0;
|
|
|
|
|
block_read_byte = 0;
|
|
|
|
|
block_read_time = 0;
|
|
|
|
|
block_checksum_time = 0;
|
|
|
|
|
block_decompress_time = 0;
|
2017-08-18 20:40:36 +02:00
|
|
|
get_read_bytes = 0;
|
|
|
|
|
multiget_read_bytes = 0;
|
|
|
|
|
iter_read_bytes = 0;
|
2013-10-02 19:28:25 +02:00
|
|
|
internal_key_skipped_count = 0;
|
|
|
|
|
internal_delete_skipped_count = 0;
|
2016-11-28 19:12:28 +01:00
|
|
|
internal_recent_skipped_count = 0;
|
|
|
|
|
internal_merge_count = 0;
|
2013-11-18 20:32:54 +01:00
|
|
|
write_wal_time = 0;
|
|
|
|
|
|
|
|
|
|
get_snapshot_time = 0;
|
|
|
|
|
get_from_memtable_time = 0;
|
|
|
|
|
get_from_memtable_count = 0;
|
|
|
|
|
get_post_process_time = 0;
|
|
|
|
|
get_from_output_files_time = 0;
|
2015-02-28 02:06:06 +01:00
|
|
|
seek_on_memtable_time = 0;
|
|
|
|
|
seek_on_memtable_count = 0;
|
2016-11-28 19:12:28 +01:00
|
|
|
next_on_memtable_count = 0;
|
|
|
|
|
prev_on_memtable_count = 0;
|
2013-11-18 20:32:54 +01:00
|
|
|
seek_child_seek_time = 0;
|
|
|
|
|
seek_child_seek_count = 0;
|
|
|
|
|
seek_min_heap_time = 0;
|
|
|
|
|
seek_internal_seek_time = 0;
|
|
|
|
|
find_next_user_entry_time = 0;
|
|
|
|
|
write_pre_and_post_process_time = 0;
|
|
|
|
|
write_memtable_time = 0;
|
2015-06-02 11:07:58 +02:00
|
|
|
write_delay_time = 0;
|
2015-02-10 01:12:31 +01:00
|
|
|
db_mutex_lock_nanos = 0;
|
|
|
|
|
db_condition_wait_nanos = 0;
|
2015-03-03 19:59:36 +01:00
|
|
|
merge_operator_time_nanos = 0;
|
2015-07-09 01:34:48 +02:00
|
|
|
read_index_block_nanos = 0;
|
|
|
|
|
read_filter_block_nanos = 0;
|
|
|
|
|
new_table_block_iter_nanos = 0;
|
2015-07-11 01:09:10 +02:00
|
|
|
new_table_iterator_nanos = 0;
|
2015-07-09 01:34:48 +02:00
|
|
|
block_seek_nanos = 0;
|
|
|
|
|
find_table_nanos = 0;
|
2015-10-07 20:23:20 +02:00
|
|
|
bloom_memtable_hit_count = 0;
|
|
|
|
|
bloom_memtable_miss_count = 0;
|
|
|
|
|
bloom_sst_hit_count = 0;
|
|
|
|
|
bloom_sst_miss_count = 0;
|
2017-04-04 20:12:47 +02:00
|
|
|
|
|
|
|
|
env_new_sequential_file_nanos = 0;
|
|
|
|
|
env_new_random_access_file_nanos = 0;
|
|
|
|
|
env_new_writable_file_nanos = 0;
|
|
|
|
|
env_reuse_writable_file_nanos = 0;
|
|
|
|
|
env_new_random_rw_file_nanos = 0;
|
|
|
|
|
env_new_directory_nanos = 0;
|
|
|
|
|
env_file_exists_nanos = 0;
|
|
|
|
|
env_get_children_nanos = 0;
|
|
|
|
|
env_get_children_file_attributes_nanos = 0;
|
|
|
|
|
env_delete_file_nanos = 0;
|
|
|
|
|
env_create_dir_nanos = 0;
|
|
|
|
|
env_create_dir_if_missing_nanos = 0;
|
|
|
|
|
env_delete_dir_nanos = 0;
|
|
|
|
|
env_get_file_size_nanos = 0;
|
|
|
|
|
env_get_file_modification_time_nanos = 0;
|
|
|
|
|
env_rename_file_nanos = 0;
|
|
|
|
|
env_link_file_nanos = 0;
|
|
|
|
|
env_lock_file_nanos = 0;
|
|
|
|
|
env_unlock_file_nanos = 0;
|
|
|
|
|
env_new_logger_nanos = 0;
|
2014-04-08 19:58:07 +02:00
|
|
|
#endif
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
}
|
|
|
|
|
|
2016-02-04 01:04:10 +01:00
|
|
|
#define PERF_CONTEXT_OUTPUT(counter) \
|
2016-02-01 22:41:13 +01:00
|
|
|
if (!exclude_zero_counters || (counter > 0)) { \
|
|
|
|
|
ss << #counter << " = " << counter << ", "; \
|
|
|
|
|
}
|
2014-03-05 19:32:54 +01:00
|
|
|
|
2016-02-01 22:41:13 +01:00
|
|
|
std::string PerfContext::ToString(bool exclude_zero_counters) const {
|
2017-06-03 02:12:39 +02:00
|
|
|
#ifdef NPERF_CONTEXT
|
2014-04-08 19:58:07 +02:00
|
|
|
return "";
|
|
|
|
|
#else
|
2014-03-05 19:32:54 +01:00
|
|
|
std::ostringstream ss;
|
2016-02-04 01:04:10 +01:00
|
|
|
PERF_CONTEXT_OUTPUT(user_key_comparison_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(block_cache_hit_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(block_read_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(block_read_byte);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(block_read_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(block_checksum_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(block_decompress_time);
|
2017-08-18 20:40:36 +02:00
|
|
|
PERF_CONTEXT_OUTPUT(get_read_bytes);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(multiget_read_bytes);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(iter_read_bytes);
|
2016-02-04 01:04:10 +01:00
|
|
|
PERF_CONTEXT_OUTPUT(internal_key_skipped_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(internal_delete_skipped_count);
|
2016-11-28 19:12:28 +01:00
|
|
|
PERF_CONTEXT_OUTPUT(internal_recent_skipped_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(internal_merge_count);
|
2016-02-04 01:04:10 +01:00
|
|
|
PERF_CONTEXT_OUTPUT(write_wal_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(get_snapshot_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(get_from_memtable_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(get_from_memtable_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(get_post_process_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(get_from_output_files_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(seek_on_memtable_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(seek_on_memtable_count);
|
2016-11-28 19:12:28 +01:00
|
|
|
PERF_CONTEXT_OUTPUT(next_on_memtable_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(prev_on_memtable_count);
|
2016-02-04 01:04:10 +01:00
|
|
|
PERF_CONTEXT_OUTPUT(seek_child_seek_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(seek_child_seek_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(seek_min_heap_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(seek_internal_seek_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(find_next_user_entry_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(write_pre_and_post_process_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(write_memtable_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(db_mutex_lock_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(db_condition_wait_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(merge_operator_time_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(write_delay_time);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(read_index_block_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(read_filter_block_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(new_table_block_iter_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(new_table_iterator_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(block_seek_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(find_table_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(bloom_memtable_hit_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(bloom_memtable_miss_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(bloom_sst_hit_count);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(bloom_sst_miss_count);
|
2017-04-04 20:12:47 +02:00
|
|
|
PERF_CONTEXT_OUTPUT(env_new_sequential_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_new_random_access_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_new_writable_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_reuse_writable_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_new_random_rw_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_new_directory_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_file_exists_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_get_children_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_get_children_file_attributes_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_delete_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_create_dir_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_create_dir_if_missing_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_delete_dir_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_get_file_size_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_get_file_modification_time_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_rename_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_link_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_lock_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_unlock_file_nanos);
|
|
|
|
|
PERF_CONTEXT_OUTPUT(env_new_logger_nanos);
|
2014-03-05 19:32:54 +01:00
|
|
|
return ss.str();
|
2014-04-08 19:58:07 +02:00
|
|
|
#endif
|
2014-03-05 19:32:54 +01:00
|
|
|
}
|
|
|
|
|
|
[RocksDB] Added nano second stopwatch and new perf counters to track block read cost
Summary: The pupose of this diff is to expose per user-call level precise timing of block read, so that we can answer questions like: a Get() costs me 100ms, is that somehow related to loading blocks from file system, or sth else? We will answer that with EXACTLY how many blocks have been read, how much time was spent on transfering the bytes from os, how much time was spent on checksum verification and how much time was spent on block decompression, just for that one Get. A nano second stopwatch was introduced to track time with higher precision. The cost/precision of the stopwatch is also measured in unit-test. On my dev box, retrieving one time instance costs about 30ns, on average. The deviation of timing results is good enough to track 100ns-1us level events. And the overhead could be safely ignored for 100us level events (10000 instances/s), for example, a viewstate thrift call.
Test Plan: perf_context_test, also testing with viewstate shadow traffic.
Reviewers: dhruba
Reviewed By: dhruba
CC: leveldb, xjin
Differential Revision: https://reviews.facebook.net/D12351
2013-06-04 08:09:15 +02:00
|
|
|
}
|
