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

Support marking snapshots for write-conflict checking - Take 2

Browse Source 
Summary:
D51183 was reverted due to breaking the LITE build.

This diff is the same as D51183 but with a fix for the LITE BUILD(D51693)

Test Plan: run all unit tests

Reviewers: sdong

Reviewed By: sdong

Subscribers: dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D51711
This commit is contained in:
agiardullo 2015-12-08 12:25:48 -08:00
parent ea11923550
commit e5c5f23814
13 changed files with 95 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
db/builder.cc
View File

@ -96,7 +96,8 @@ Status BuildTable(
snapshots.empty() ? 0 : snapshots.back());
CompactionIterator c_iter(iter, internal_comparator.user_comparator(),
&merge, kMaxSequenceNumber, &snapshots, env,
&merge, kMaxSequenceNumber, &snapshots,
kMaxSequenceNumber, env,
true /* internal key corruption is not ok */);
c_iter.SeekToFirst();
for (; c_iter.Valid(); c_iter.Next()) {

9
db/compaction_iterator.cc
View File

@ -13,12 +13,14 @@ namespace rocksdb {
CompactionIterator::CompactionIterator(
InternalIterator* input, const Comparator* cmp, MergeHelper* merge_helper,
SequenceNumber last_sequence, std::vector<SequenceNumber>* snapshots,
Env* env, bool expect_valid_internal_key, Compaction* compaction,
SequenceNumber earliest_write_conflict_snapshot, Env* env,
bool expect_valid_internal_key, Compaction* compaction,
const CompactionFilter* compaction_filter, LogBuffer* log_buffer)
: input_(input),
cmp_(cmp),
merge_helper_(merge_helper),
snapshots_(snapshots),
earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
env_(env),
expect_valid_internal_key_(expect_valid_internal_key),
compaction_(compaction),
@ -200,6 +202,11 @@ void CompactionIterator::NextFromInput() {
ParsedInternalKey next_ikey;
input_->Next();
if (earliest_write_conflict_snapshot_) {
// TODO(agiardullo): to be used in D50295
// adding this if statement to keep CLANG happy in the meantime
}
// Check whether the current key is valid, not corrupt and the same
// as the single delete.
if (input_->Valid() && ParseInternalKey(input_->key(), &next_ikey) &&

4
db/compaction_iterator.h
View File

@ -39,7 +39,8 @@ class CompactionIterator {
public:
CompactionIterator(InternalIterator* input, const Comparator* cmp,
MergeHelper* merge_helper, SequenceNumber last_sequence,
std::vector<SequenceNumber>* snapshots, Env* env,
std::vector<SequenceNumber>* snapshots,
SequenceNumber earliest_write_conflict_snapshot, Env* env,
bool expect_valid_internal_key,
Compaction* compaction = nullptr,
const CompactionFilter* compaction_filter = nullptr,
@ -88,6 +89,7 @@ class CompactionIterator {
const Comparator* cmp_;
MergeHelper* merge_helper_;
const std::vector<SequenceNumber>* snapshots_;
const SequenceNumber earliest_write_conflict_snapshot_;
Env* env_;
bool expect_valid_internal_key_;
Compaction* compaction_;

6
db/compaction_iterator_test.cc
View File

@ -20,9 +20,9 @@ class CompactionIteratorTest : public testing::Test {
nullptr, 0U, false, 0));
iter_.reset(new test::VectorIterator(ks, vs));
iter_->SeekToFirst();
c_iter_.reset(new CompactionIterator(iter_.get(), cmp_, merge_helper_.get(),
last_sequence, &snapshots_,
Env::Default(), false));
c_iter_.reset(new CompactionIterator(
iter_.get(), cmp_, merge_helper_.get(), last_sequence, &snapshots_,
kMaxSequenceNumber, Env::Default(), false));
}
const Comparator* cmp_;

6
db/compaction_job.cc
View File

@ -212,6 +212,7 @@ CompactionJob::CompactionJob(
std::atomic<bool>* shutting_down, LogBuffer* log_buffer,
Directory* db_directory, Directory* output_directory, Statistics* stats,
std::vector<SequenceNumber> existing_snapshots,
SequenceNumber earliest_write_conflict_snapshot,
std::shared_ptr<Cache> table_cache, EventLogger* event_logger,
bool paranoid_file_checks, bool measure_io_stats, const std::string& dbname,
CompactionJobStats* compaction_job_stats)
@ -230,6 +231,7 @@ CompactionJob::CompactionJob(
output_directory_(output_directory),
stats_(stats),
existing_snapshots_(std::move(existing_snapshots)),
earliest_write_conflict_snapshot_(earliest_write_conflict_snapshot),
table_cache_(std::move(table_cache)),
event_logger_(event_logger),
paranoid_file_checks_(paranoid_file_checks),
@ -638,8 +640,8 @@ void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
Status status;
sub_compact->c_iter.reset(new CompactionIterator(
input.get(), cfd->user_comparator(), &merge, versions_->LastSequence(),
&existing_snapshots_, env_, false, sub_compact->compaction,
compaction_filter));
&existing_snapshots_, earliest_write_conflict_snapshot_, env_, false,
sub_compact->compaction, compaction_filter));
auto c_iter = sub_compact->c_iter.get();
c_iter->SeekToFirst();
const auto& c_iter_stats = c_iter->iter_stats();

7
db/compaction_job.h
View File

@ -58,6 +58,7 @@ class CompactionJob {
Directory* db_directory, Directory* output_directory,
Statistics* stats,
std::vector<SequenceNumber> existing_snapshots,
SequenceNumber earliest_write_conflict_snapshot,
std::shared_ptr<Cache> table_cache, EventLogger* event_logger,
bool paranoid_file_checks, bool measure_io_stats,
const std::string& dbname,
@ -134,6 +135,12 @@ class CompactionJob {
// entirely within s1 and s2, then the earlier version of k1 can be safely
// deleted because that version is not visible in any snapshot.
std::vector<SequenceNumber> existing_snapshots_;
// This is the earliest snapshot that could be used for write-conflict
// checking by a transaction. For any user-key newer than this snapshot, we
// should make sure not to remove evidence that a write occured.
SequenceNumber earliest_write_conflict_snapshot_;
std::shared_ptr<Cache> table_cache_;
EventLogger* event_logger_;

10
db/compaction_job_test.cc
View File

@ -243,11 +243,11 @@ class CompactionJobTest : public testing::Test {
LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL, db_options_.info_log.get());
mutex_.Lock();
EventLogger event_logger(db_options_.info_log.get());
CompactionJob compaction_job(0, &compaction, db_options_, env_options_,
versions_.get(), &shutting_down_, &log_buffer,
nullptr, nullptr, nullptr, snapshots,
table_cache_, &event_logger, false, false,
dbname_, &compaction_job_stats_);
CompactionJob compaction_job(
0, &compaction, db_options_, env_options_, versions_.get(),
&shutting_down_, &log_buffer, nullptr, nullptr, nullptr, snapshots,
kMaxSequenceNumber, table_cache_, &event_logger, false, false, dbname_,
&compaction_job_stats_);
VerifyInitializationOfCompactionJobStats(compaction_job_stats_);

28
db/db_impl.cc
View File

@ -1779,12 +1779,16 @@ Status DBImpl::CompactFilesImpl(
// deletion compaction currently not allowed in CompactFiles.
assert(!c->deletion_compaction());
SequenceNumber earliest_write_conflict_snapshot;
std::vector<SequenceNumber> snapshot_seqs =
snapshots_.GetAll(&earliest_write_conflict_snapshot);
assert(is_snapshot_supported_ || snapshots_.empty());
CompactionJob compaction_job(
job_context->job_id, c.get(), db_options_, env_options_, versions_.get(),
&shutting_down_, log_buffer, directories_.GetDbDir(),
directories_.GetDataDir(c->output_path_id()), stats_, snapshots_.GetAll(),
table_cache_, &event_logger_,
directories_.GetDataDir(c->output_path_id()), stats_, snapshot_seqs,
earliest_write_conflict_snapshot, table_cache_, &event_logger_,
c->mutable_cf_options()->paranoid_file_checks,
c->mutable_cf_options()->compaction_measure_io_stats, dbname_,
nullptr); // Here we pass a nullptr for CompactionJobStats because
@ -2868,12 +2872,17 @@ Status DBImpl::BackgroundCompaction(bool* made_progress,
int output_level __attribute__((unused)) = c->output_level();
TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:NonTrivial",
&output_level);
SequenceNumber earliest_write_conflict_snapshot;
std::vector<SequenceNumber> snapshot_seqs =
snapshots_.GetAll(&earliest_write_conflict_snapshot);
assert(is_snapshot_supported_ || snapshots_.empty());
CompactionJob compaction_job(
job_context->job_id, c.get(), db_options_, env_options_,
versions_.get(), &shutting_down_, log_buffer, directories_.GetDbDir(),
directories_.GetDataDir(c->output_path_id()), stats_,
snapshots_.GetAll(), table_cache_, &event_logger_,
directories_.GetDataDir(c->output_path_id()), stats_, snapshot_seqs,
earliest_write_conflict_snapshot, table_cache_, &event_logger_,
c->mutable_cf_options()->paranoid_file_checks,
c->mutable_cf_options()->compaction_measure_io_stats, dbname_,
&compaction_job_stats);
@ -3784,7 +3793,13 @@ Status DBImpl::NewIterators(
return Status::OK();
}
const Snapshot* DBImpl::GetSnapshot() {
const Snapshot* DBImpl::GetSnapshot() { return GetSnapshotImpl(false); }
const Snapshot* DBImpl::GetSnapshotForWriteConflictBoundary() {
return GetSnapshotImpl(true);
}
const Snapshot* DBImpl::GetSnapshotImpl(bool is_write_conflict_boundary) {
int64_t unix_time = 0;
env_->GetCurrentTime(&unix_time); // Ignore error
SnapshotImpl* s = new SnapshotImpl;
@ -3795,7 +3810,8 @@ const Snapshot* DBImpl::GetSnapshot() {
delete s;
return nullptr;
}
return snapshots_.New(s, versions_->LastSequence(), unix_time);
return snapshots_.New(s, versions_->LastSequence(), unix_time,
is_write_conflict_boundary);
}
void DBImpl::ReleaseSnapshot(const Snapshot* s) {

8
db/db_impl.h
View File

@ -243,6 +243,12 @@ class DBImpl : public DB {
#endif // ROCKSDB_LITE
// Similar to GetSnapshot(), but also lets the db know that this snapshot
// will be used for transaction write-conflict checking. The DB can then
// make sure not to compact any keys that would prevent a write-conflict from
// being detected.
const Snapshot* GetSnapshotForWriteConflictBoundary();
// checks if all live files exist on file system and that their file sizes
// match to our in-memory records
virtual Status CheckConsistency();
@ -560,6 +566,8 @@ class DBImpl : public DB {
// helper function to call after some of the logs_ were synced
void MarkLogsSynced(uint64_t up_to, bool synced_dir, const Status& status);
const Snapshot* GetSnapshotImpl(bool is_write_conflict_boundary);
// table_cache_ provides its own synchronization
std::shared_ptr<Cache> table_cache_;

3
db/snapshot_impl.cc
View File

@ -12,6 +12,9 @@ namespace rocksdb {
ManagedSnapshot::ManagedSnapshot(DB* db) : db_(db),
snapshot_(db->GetSnapshot()) {}
ManagedSnapshot::ManagedSnapshot(DB* db, const Snapshot* _snapshot)
: db_(db), snapshot_(_snapshot) {}
ManagedSnapshot::~ManagedSnapshot() {
if (snapshot_) {
db_->ReleaseSnapshot(snapshot_);

23
db/snapshot_impl.h
View File

@ -34,6 +34,9 @@ class SnapshotImpl : public Snapshot {
SnapshotList* list_; // just for sanity checks
int64_t unix_time_;
// Will this snapshot be used by a Transaction to do write-conflict checking?
bool is_write_conflict_boundary_;
};
class SnapshotList {
@ -50,9 +53,10 @@ class SnapshotList {
SnapshotImpl* newest() const { assert(!empty()); return list_.prev_; }
const SnapshotImpl* New(SnapshotImpl* s, SequenceNumber seq,
uint64_t unix_time) {
uint64_t unix_time, bool is_write_conflict_boundary) {
s->number_ = seq;
s->unix_time_ = unix_time;
s->is_write_conflict_boundary_ = is_write_conflict_boundary;
s->list_ = this;
s->next_ = &list_;
s->prev_ = list_.prev_;
@ -71,14 +75,29 @@ class SnapshotList {
}
// retrieve all snapshot numbers. They are sorted in ascending order.
std::vector<SequenceNumber> GetAll() {
std::vector<SequenceNumber> GetAll(
SequenceNumber* oldest_write_conflict_snapshot = nullptr) {
std::vector<SequenceNumber> ret;
if (oldest_write_conflict_snapshot != nullptr) {
*oldest_write_conflict_snapshot = kMaxSequenceNumber;
}
if (empty()) {
return ret;
}
SnapshotImpl* s = &list_;
while (s->next_ != &list_) {
ret.push_back(s->next_->number_);
if (oldest_write_conflict_snapshot != nullptr &&
*oldest_write_conflict_snapshot != kMaxSequenceNumber &&
s->next_->is_write_conflict_boundary_) {
// If this is the first write-conflict boundary snapshot in the list,
// it is the oldest
*oldest_write_conflict_snapshot = s->next_->number_;
}
s = s->next_;
}
return ret;

3
include/rocksdb/snapshot.h
View File

@ -33,6 +33,9 @@ class ManagedSnapshot {
public:
explicit ManagedSnapshot(DB* db);
// Instead of creating a snapshot, take ownership of the input snapshot.
ManagedSnapshot(DB* db, const Snapshot* _snapshot);
~ManagedSnapshot();
const Snapshot* snapshot();

7
utilities/transactions/transaction_base.cc
View File

@ -7,6 +7,7 @@
#include "utilities/transactions/transaction_base.h"
#include "db/db_impl.h"
#include "db/column_family.h"
#include "rocksdb/comparator.h"
#include "rocksdb/db.h"
@ -35,7 +36,11 @@ void TransactionBaseImpl::Clear() {
}
void TransactionBaseImpl::SetSnapshot() {
snapshot_.reset(new ManagedSnapshot(db_));
assert(dynamic_cast<DBImpl*>(db_) != nullptr);
auto db_impl = reinterpret_cast<DBImpl*>(db_);
const Snapshot* snapshot = db_impl->GetSnapshotForWriteConflictBoundary();
snapshot_.reset(new ManagedSnapshot(db_, snapshot));
snapshot_needed_ = false;
snapshot_notifier_ = nullptr;
}