andreacavalli/rocksdb
0
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Avoid fsync on the same directory in atomic flush (#4817)

Browse Source 
Summary:
In `DBImpl::AtomicFlushMemTablesToOutputFiles`, we need to call fsync only once
on the same data directory. If two column families share a common directory for
their data, we call fsync only once.
Pull Request resolved: https://github.com/facebook/rocksdb/pull/4817

Differential Revision: D13543689

Pulled By: riversand963

fbshipit-source-id: 4701d77c96a47802fbf6cb9f3337ee65d46b95f5
This commit is contained in:
Yanqin Jin 2019-02-12 12:01:55 -08:00 committed by Facebook Github Bot
parent 62f70f6d14
commit c5a64cffd2
1 changed files with 6 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
db/db_impl_compaction_flush.cc
View File

@ -296,6 +296,7 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
&earliest_write_conflict_snapshot, &snapshot_checker); &earliest_write_conflict_snapshot, &snapshot_checker);
autovector<Directory*> distinct_output_dirs; autovector<Directory*> distinct_output_dirs;
autovector<std::string> distinct_output_dir_paths;
std::vector<FlushJob> jobs; std::vector<FlushJob> jobs;
std::vector<MutableCFOptions> all_mutable_cf_options; std::vector<MutableCFOptions> all_mutable_cf_options;
int num_cfs = static_cast<int>(cfds.size()); int num_cfs = static_cast<int>(cfds.size());
@ -303,18 +304,20 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
for (int i = 0; i < num_cfs; ++i) { for (int i = 0; i < num_cfs; ++i) {
auto cfd = cfds[i]; auto cfd = cfds[i];
Directory* data_dir = GetDataDir(cfd, 0U); Directory* data_dir = GetDataDir(cfd, 0U);
const std::string& curr_path = cfd->ioptions()->cf_paths[0].path;
// Add to distinct output directories if eligible. Use linear search. Since // Add to distinct output directories if eligible. Use linear search. Since
// the number of elements in the vector is not large, performance should be // the number of elements in the vector is not large, performance should be
// tolerable. // tolerable.
bool found = false; bool found = false;
for (const auto dir : distinct_output_dirs) { for (const auto& path : distinct_output_dir_paths) {
if (dir == data_dir) { if (path == curr_path) {
found = true; found = true;
break; break;
} }
} }
if (!found) { if (!found) {
distinct_output_dir_paths.emplace_back(curr_path);
distinct_output_dirs.emplace_back(data_dir); distinct_output_dirs.emplace_back(data_dir);
} }
@ -322,7 +325,7 @@ Status DBImpl::AtomicFlushMemTablesToOutputFiles(
const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.back(); const MutableCFOptions& mutable_cf_options = all_mutable_cf_options.back();
const uint64_t* max_memtable_id = &(bg_flush_args[i].max_memtable_id_); const uint64_t* max_memtable_id = &(bg_flush_args[i].max_memtable_id_);
jobs.emplace_back( jobs.emplace_back(
dbname_, cfds[i], immutable_db_options_, mutable_cf_options, dbname_, cfd, immutable_db_options_, mutable_cf_options,
max_memtable_id, env_options_for_compaction_, versions_.get(), &mutex_, max_memtable_id, env_options_for_compaction_, versions_.get(), &mutex_,
&shutting_down_, snapshot_seqs, earliest_write_conflict_snapshot, &shutting_down_, snapshot_seqs, earliest_write_conflict_snapshot,
snapshot_checker, job_context, log_buffer, directories_.GetDbDir(), snapshot_checker, job_context, log_buffer, directories_.GetDbDir(),