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

Minimize accessing multiple objects in Version::Get()

Browse Source 
Summary:
One of our profilings shows that Version::Get() sometimes is slow when getting pointer of user comparators or other global objects. In this patch:
(1) we keep pointers of immutable objects in Version to avoid accesses them though option objects or cfd objects
(2) table_reader is directly cached in FileMetaData so that table cache don't have to go through handle first to fetch it
(3) If level 0 has less than 3 files, skip the filtering logic based on SST tables' key range. Smallest and largest key are stored in separated memory locations, which has potential cache misses

Test Plan: make all check

Reviewers: haobo, ljin

Reviewed By: haobo

CC: igor, yhchiang, nkg-, leveldb

Differential Revision: https://reviews.facebook.net/D17739
This commit is contained in:
sdong 2014-04-17 14:07:05 -07:00
parent e37dd216f9
commit fa430bfd04
6 changed files with 69 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
db/db_impl.cc
View File

@ -3219,7 +3219,7 @@ Status DBImpl::GetImpl(const ReadOptions& options,
PERF_TIMER_START(get_from_output_files_time); PERF_TIMER_START(get_from_output_files_time);
sv->current->Get(options, lkey, value, &s, &merge_context, &stats, sv->current->Get(options, lkey, value, &s, &merge_context, &stats,
*cfd->options(), value_found); value_found);
have_stat_update = true; have_stat_update = true;
PERF_TIMER_STOP(get_from_output_files_time); PERF_TIMER_STOP(get_from_output_files_time);
RecordTick(options_.statistics.get(), MEMTABLE_MISS); RecordTick(options_.statistics.get(), MEMTABLE_MISS);
@ -3334,7 +3334,7 @@ std::vector<Status> DBImpl::MultiGet(
// Done // Done
} else { } else {
super_version->current->Get(options, lkey, value, &s, &merge_context, super_version->current->Get(options, lkey, value, &s, &merge_context,
&mgd->stats, *cfd->options()); &mgd->stats);
mgd->have_stat_update = true; mgd->have_stat_update = true;
} }

2
db/db_impl_readonly.cc
View File

@ -67,7 +67,7 @@ Status DBImplReadOnly::Get(const ReadOptions& options,
} else { } else {
Version::GetStats stats; Version::GetStats stats;
super_version->current->Get(options, lkey, value, &s, &merge_context, super_version->current->Get(options, lkey, value, &s, &merge_context,
&stats, *cfd->options()); &stats);
} }
return s; return s;
} }

38
db/table_cache.cc
View File

@ -106,19 +106,20 @@ Iterator* TableCache::NewIterator(const ReadOptions& options,
if (table_reader_ptr != nullptr) { if (table_reader_ptr != nullptr) {
*table_reader_ptr = nullptr; *table_reader_ptr = nullptr;
} }
Cache::Handle* handle = file_meta.table_reader_handle; TableReader* table_reader = file_meta.table_reader;
Cache::Handle* handle = nullptr;
Status s; Status s;
if (!handle) { if (table_reader == nullptr) {
s = FindTable(toptions, icomparator, file_meta.number, file_meta.file_size, s = FindTable(toptions, icomparator, file_meta.number, file_meta.file_size,
&handle, nullptr, options.read_tier == kBlockCacheTier); &handle, nullptr, options.read_tier == kBlockCacheTier);
table_reader = GetTableReaderFromHandle(handle);
} }
if (!s.ok()) { if (!s.ok()) {
return NewErrorIterator(s); return NewErrorIterator(s);
} }
TableReader* table_reader = GetTableReaderFromHandle(handle);
Iterator* result = table_reader->NewIterator(options); Iterator* result = table_reader->NewIterator(options);
if (!file_meta.table_reader_handle) { if (handle != nullptr) {
result->RegisterCleanup(&UnrefEntry, cache_, handle); result->RegisterCleanup(&UnrefEntry, cache_, handle);
} }
if (table_reader_ptr != nullptr) { if (table_reader_ptr != nullptr) {
@ -138,17 +139,18 @@ Status TableCache::Get(const ReadOptions& options,
bool (*saver)(void*, const ParsedInternalKey&, bool (*saver)(void*, const ParsedInternalKey&,
const Slice&, bool), const Slice&, bool),
bool* table_io, void (*mark_key_may_exist)(void*)) { bool* table_io, void (*mark_key_may_exist)(void*)) {
Cache::Handle* handle = file_meta.table_reader_handle; TableReader* t = file_meta.table_reader;
Status s; Status s;
if (!handle) { Cache::Handle* handle = nullptr;
if (!t) {
s = FindTable(storage_options_, internal_comparator, file_meta.number, s = FindTable(storage_options_, internal_comparator, file_meta.number,
file_meta.file_size, &handle, table_io, file_meta.file_size, &handle, table_io,
options.read_tier == kBlockCacheTier); options.read_tier == kBlockCacheTier);
t = GetTableReaderFromHandle(handle);
} }
if (s.ok()) { if (s.ok()) {
TableReader* t = GetTableReaderFromHandle(handle);
s = t->Get(options, k, arg, saver, mark_key_may_exist); s = t->Get(options, k, arg, saver, mark_key_may_exist);
if (!file_meta.table_reader_handle) { if (handle != nullptr) {
ReleaseHandle(handle); ReleaseHandle(handle);
} }
} else if (options.read_tier && s.IsIncomplete()) { } else if (options.read_tier && s.IsIncomplete()) {
@ -164,15 +166,16 @@ Status TableCache::GetTableProperties(
const FileMetaData& file_meta, const FileMetaData& file_meta,
std::shared_ptr<const TableProperties>* properties, bool no_io) { std::shared_ptr<const TableProperties>* properties, bool no_io) {
Status s; Status s;
auto table_handle = file_meta.table_reader_handle; auto table_reader = file_meta.table_reader;
// table already been pre-loaded? // table already been pre-loaded?
if (table_handle) { if (table_reader) {
auto table = GetTableReaderFromHandle(table_handle); *properties = table_reader->GetTableProperties();
*properties = table->GetTableProperties();
return s; return s;
} }
bool table_io; bool table_io;
Cache::Handle* table_handle = nullptr;
s = FindTable(toptions, internal_comparator, file_meta.number, s = FindTable(toptions, internal_comparator, file_meta.number,
file_meta.file_size, &table_handle, &table_io, no_io); file_meta.file_size, &table_handle, &table_io, no_io);
if (!s.ok()) { if (!s.ok()) {
@ -190,20 +193,21 @@ bool TableCache::PrefixMayMatch(const ReadOptions& options,
const FileMetaData& file_meta, const FileMetaData& file_meta,
const Slice& internal_prefix, bool* table_io) { const Slice& internal_prefix, bool* table_io) {
bool may_match = true; bool may_match = true;
auto table_handle = file_meta.table_reader_handle; auto table_reader = file_meta.table_reader;
if (table_handle == nullptr) { Cache::Handle* table_handle = nullptr;
if (table_reader == nullptr) {
// Need to get table handle from file number // Need to get table handle from file number
Status s = FindTable(storage_options_, icomparator, file_meta.number, Status s = FindTable(storage_options_, icomparator, file_meta.number,
file_meta.file_size, &table_handle, table_io); file_meta.file_size, &table_handle, table_io);
if (!s.ok()) { if (!s.ok()) {
return may_match; return may_match;
} }
table_reader = GetTableReaderFromHandle(table_handle);
} }
auto table = GetTableReaderFromHandle(table_handle); may_match = table_reader->PrefixMayMatch(internal_prefix);
may_match = table->PrefixMayMatch(internal_prefix);
if (file_meta.table_reader_handle == nullptr) { if (table_handle != nullptr) {
// Need to release handle if it is generated from here. // Need to release handle if it is generated from here.
ReleaseHandle(table_handle); ReleaseHandle(table_handle);
} }

5
db/version_edit.h
View File

@ -32,6 +32,8 @@ struct FileMetaData {
// Needs to be disposed when refs becomes 0. // Needs to be disposed when refs becomes 0.
Cache::Handle* table_reader_handle; Cache::Handle* table_reader_handle;
// Table reader in table_reader_handle
TableReader* table_reader;
FileMetaData(uint64_t number, uint64_t file_size) FileMetaData(uint64_t number, uint64_t file_size)
: refs(0), : refs(0),
@ -39,7 +41,8 @@ struct FileMetaData {
number(number), number(number),
file_size(file_size), file_size(file_size),
being_compacted(false), being_compacted(false),
table_reader_handle(nullptr) {} table_reader_handle(nullptr),
table_reader(nullptr) {}
FileMetaData() : FileMetaData(0, 0) {} FileMetaData() : FileMetaData(0, 0) {}
}; };

55
db/version_set.cc
View File

@ -483,6 +483,17 @@ bool BySmallestKey(FileMetaData* a, FileMetaData* b,
Version::Version(ColumnFamilyData* cfd, VersionSet* vset, Version::Version(ColumnFamilyData* cfd, VersionSet* vset,
uint64_t version_number) uint64_t version_number)
: cfd_(cfd), : cfd_(cfd),
internal_comparator_((cfd == nullptr) ? nullptr
: &cfd->internal_comparator()),
user_comparator_((cfd == nullptr)
? nullptr
: internal_comparator_->user_comparator()),
table_cache_((cfd == nullptr) ? nullptr : cfd->table_cache()),
merge_operator_((cfd == nullptr) ? nullptr
: cfd->options()->merge_operator.get()),
info_log_((cfd == nullptr) ? nullptr : cfd->options()->info_log.get()),
db_statistics_((cfd == nullptr) ? nullptr
: cfd->options()->statistics.get()),
vset_(vset), vset_(vset),
next_(this), next_(this),
prev_(this), prev_(this),
@ -504,27 +515,22 @@ void Version::Get(const ReadOptions& options,
Status* status, Status* status,
MergeContext* merge_context, MergeContext* merge_context,
GetStats* stats, GetStats* stats,
const Options& db_options,
bool* value_found) { bool* value_found) {
Slice ikey = k.internal_key(); Slice ikey = k.internal_key();
Slice user_key = k.user_key(); Slice user_key = k.user_key();
const Comparator* ucmp = cfd_->internal_comparator().user_comparator();
auto merge_operator = db_options.merge_operator.get();
auto logger = db_options.info_log.get();
assert(status->ok() || status->IsMergeInProgress()); assert(status->ok() || status->IsMergeInProgress());
Saver saver; Saver saver;
saver.state = status->ok()? kNotFound : kMerge; saver.state = status->ok()? kNotFound : kMerge;
saver.ucmp = ucmp; saver.ucmp = user_comparator_;
saver.user_key = user_key; saver.user_key = user_key;
saver.value_found = value_found; saver.value_found = value_found;
saver.value = value; saver.value = value;
saver.merge_operator = merge_operator; saver.merge_operator = merge_operator_;
saver.merge_context = merge_context; saver.merge_context = merge_context;
saver.logger = logger; saver.logger = info_log_;
saver.didIO = false; saver.didIO = false;
saver.statistics = db_options.statistics.get(); saver.statistics = db_statistics_;
stats->seek_file = nullptr; stats->seek_file = nullptr;
stats->seek_file_level = -1; stats->seek_file_level = -1;
@ -555,7 +561,7 @@ void Version::Get(const ReadOptions& options,
// On Level-n (n>=1), files are sorted. // On Level-n (n>=1), files are sorted.
// Binary search to find earliest index whose largest key >= ikey. // Binary search to find earliest index whose largest key >= ikey.
// We will also stop when the file no longer overlaps ikey // We will also stop when the file no longer overlaps ikey
start_index = FindFile(cfd_->internal_comparator(), files_[level], ikey); start_index = FindFile(*internal_comparator_, files_[level], ikey);
} }
// Traverse each relevant file to find the desired key // Traverse each relevant file to find the desired key
@ -564,8 +570,10 @@ void Version::Get(const ReadOptions& options,
#endif #endif
for (uint32_t i = start_index; i < num_files; ++i) { for (uint32_t i = start_index; i < num_files; ++i) {
FileMetaData* f = files[i]; FileMetaData* f = files[i];
if (ucmp->Compare(user_key, f->smallest.user_key()) < 0 || // Skip key range filtering for levle 0 if there are few level 0 files.
ucmp->Compare(user_key, f->largest.user_key()) > 0) { if ((level > 0 || num_files > 2) &&
(user_comparator_->Compare(user_key, f->smallest.user_key()) < 0 ||
user_comparator_->Compare(user_key, f->largest.user_key()) > 0)) {
// Only process overlapping files. // Only process overlapping files.
if (level > 0) { if (level > 0) {
// If on Level-n (n>=1) then the files are sorted. // If on Level-n (n>=1) then the files are sorted.
@ -581,8 +589,8 @@ void Version::Get(const ReadOptions& options,
// Sanity check to make sure that the files are correctly sorted // Sanity check to make sure that the files are correctly sorted
if (prev_file) { if (prev_file) {
if (level != 0) { if (level != 0) {
int comp_sign = cfd_->internal_comparator().Compare( int comp_sign =
prev_file->largest, f->smallest); internal_comparator_->Compare(prev_file->largest, f->smallest);
assert(comp_sign < 0); assert(comp_sign < 0);
} else { } else {
// level == 0, the current file cannot be newer than the previous one. // level == 0, the current file cannot be newer than the previous one.
@ -596,9 +604,8 @@ void Version::Get(const ReadOptions& options,
prev_file = f; prev_file = f;
#endif #endif
bool tableIO = false; bool tableIO = false;
*status = cfd_->table_cache()->Get(options, cfd_->internal_comparator(), *status = table_cache_->Get(options, *internal_comparator_, *f, ikey,
*f, ikey, &saver, SaveValue, &tableIO, &saver, SaveValue, &tableIO, MarkKeyMayExist);
MarkKeyMayExist);
// TODO: examine the behavior for corrupted key // TODO: examine the behavior for corrupted key
if (!status->ok()) { if (!status->ok()) {
return; return;
@ -643,12 +650,12 @@ void Version::Get(const ReadOptions& options,
if (kMerge == saver.state) { if (kMerge == saver.state) {
// merge_operands are in saver and we hit the beginning of the key history // merge_operands are in saver and we hit the beginning of the key history
// do a final merge of nullptr and operands; // do a final merge of nullptr and operands;
if (merge_operator->FullMerge(user_key, nullptr, if (merge_operator_->FullMerge(user_key, nullptr,
saver.merge_context->GetOperands(), saver.merge_context->GetOperands(), value,
value, logger)) { info_log_)) {
*status = Status::OK(); *status = Status::OK();
} else { } else {
RecordTick(db_options.statistics.get(), NUMBER_MERGE_FAILURES); RecordTick(db_statistics_, NUMBER_MERGE_FAILURES);
*status = Status::Corruption("could not perform end-of-key merge for ", *status = Status::Corruption("could not perform end-of-key merge for ",
user_key); user_key);
} }
@ -1458,6 +1465,12 @@ class VersionSet::Builder {
base_->vset_->storage_options_, cfd_->internal_comparator(), base_->vset_->storage_options_, cfd_->internal_comparator(),
file_meta->number, file_meta->file_size, file_meta->number, file_meta->file_size,
&file_meta->table_reader_handle, &table_io, false); &file_meta->table_reader_handle, &table_io, false);
if (file_meta->table_reader_handle != nullptr) {
// Load table_reader
file_meta->table_reader =
cfd_->table_cache()->GetTableReaderFromHandle(
file_meta->table_reader_handle);
}
} }
} }
} }

9
db/version_set.h
View File

@ -88,8 +88,7 @@ class Version {
int seek_file_level; int seek_file_level;
}; };
void Get(const ReadOptions&, const LookupKey& key, std::string* val, void Get(const ReadOptions&, const LookupKey& key, std::string* val,
Status* status, MergeContext* merge_context, Status* status, MergeContext* merge_context, GetStats* stats,
GetStats* stats, const Options& db_option,
bool* value_found = nullptr); bool* value_found = nullptr);
// Adds "stats" into the current state. Returns true if a new // Adds "stats" into the current state. Returns true if a new
@ -230,6 +229,12 @@ class Version {
void UpdateFilesBySize(); void UpdateFilesBySize();
ColumnFamilyData* cfd_; // ColumnFamilyData to which this Version belongs ColumnFamilyData* cfd_; // ColumnFamilyData to which this Version belongs
const InternalKeyComparator* internal_comparator_;
const Comparator* user_comparator_;
TableCache* table_cache_;
const MergeOperator* merge_operator_;
Logger* info_log_;
Statistics* db_statistics_;
VersionSet* vset_; // VersionSet to which this Version belongs VersionSet* vset_; // VersionSet to which this Version belongs
Version* next_; // Next version in linked list Version* next_; // Next version in linked list
Version* prev_; // Previous version in linked list Version* prev_; // Previous version in linked list