Hawkin
/
arangodb
mirror of https://gitee.com/bigwinds/arangodb
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
arangodb/arangod/RocksDBEngine/RocksDBTransactionState.cpp

626 lines
20 KiB
C++
Raw Blame History

////////////////////////////////////////////////////////////////////////////////
/// DISCLAIMER
///
/// Copyright 2014-2016 ArangoDB GmbH, Cologne, Germany
/// Copyright 2004-2014 triAGENS GmbH, Cologne, Germany
///
/// Licensed under the Apache License, Version 2.0 (the "License");
/// you may not use this file except in compliance with the License.
/// You may obtain a copy of the License at
///
/// http://www.apache.org/licenses/LICENSE-2.0
///
/// Unless required by applicable law or agreed to in writing, software
/// distributed under the License is distributed on an "AS IS" BASIS,
/// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
/// See the License for the specific language governing permissions and
/// limitations under the License.
///
/// Copyright holder is ArangoDB GmbH, Cologne, Germany
///
/// @author Jan Steemann
////////////////////////////////////////////////////////////////////////////////
#include "RocksDBTransactionState.h"
#include "Aql/QueryCache.h"
#include "Basics/Exceptions.h"
#include "Cache/CacheManagerFeature.h"
#include "Cache/Manager.h"
#include "Cache/Transaction.h"
#include "Logger/Logger.h"
#include "RestServer/TransactionManagerFeature.h"
#include "RocksDBEngine/RocksDBCollection.h"
#include "RocksDBEngine/RocksDBCommon.h"
#include "RocksDBEngine/RocksDBEngine.h"
#include "RocksDBEngine/RocksDBLogValue.h"
#include "RocksDBEngine/RocksDBMethods.h"
#include "RocksDBEngine/RocksDBTransactionCollection.h"
#include "StorageEngine/StorageEngine.h"
#include "StorageEngine/TransactionCollection.h"
#include "StorageEngine/TransactionManager.h"
#include "Transaction/Methods.h"
#include "Utils/ExecContext.h"
#include "VocBase/LogicalCollection.h"
#include "VocBase/ticks.h"
#include <rocksdb/options.h>
#include <rocksdb/status.h>
#include <rocksdb/utilities/optimistic_transaction_db.h>
#include <rocksdb/utilities/transaction.h>
#include <rocksdb/utilities/transaction_db.h>
#include <rocksdb/utilities/write_batch_with_index.h>
using namespace arangodb;
// for the RocksDB engine we do not need any additional data
struct RocksDBTransactionData final : public TransactionData {};
/// @brief transaction type
RocksDBTransactionState::RocksDBTransactionState(
TRI_vocbase_t& vocbase,
TRI_voc_tid_t tid,
transaction::Options const& options
): TransactionState(vocbase, tid, options),
_rocksTransaction(nullptr),
_snapshot(nullptr),
_rocksWriteOptions(),
_rocksReadOptions(),
_cacheTx(nullptr),
_numCommits(0),
_numInserts(0),
_numUpdates(0),
_numRemoves(0),
_keys{_arena},
_parallel(false) {}
/// @brief free a transaction container
RocksDBTransactionState::~RocksDBTransactionState() {
cleanupTransaction();
for (auto& it : _keys) {
delete it;
}
}
/// @brief start a transaction
Result RocksDBTransactionState::beginTransaction(transaction::Hints hints) {
LOG_TRX(this, _nestingLevel)
<< "beginning " << AccessMode::typeString(_type) << " transaction";
TRI_ASSERT(!hasHint(transaction::Hints::Hint::NO_USAGE_LOCK) || !AccessMode::isWriteOrExclusive(_type));
if (_nestingLevel == 0) {
// set hints
_hints = hints;
}
Result result = useCollections(_nestingLevel);
if (result.ok()) {
// all valid
if (_nestingLevel == 0) {
updateStatus(transaction::Status::RUNNING);
}
} else {
// something is wrong
if (_nestingLevel == 0) {
updateStatus(transaction::Status::ABORTED);
}
// free what we have got so far
unuseCollections(_nestingLevel);
return result;
}
if (_nestingLevel == 0) {
// register a protector (intentionally empty)
auto data = std::make_unique<RocksDBTransactionData>();
TransactionManagerFeature::manager()->registerTransaction(_id,
std::move(data));
TRI_ASSERT(_rocksTransaction == nullptr);
TRI_ASSERT(_cacheTx == nullptr);
// start cache transaction
_cacheTx =
CacheManagerFeature::MANAGER->beginTransaction(isReadOnlyTransaction());
rocksdb::TransactionDB* db = rocksutils::globalRocksDB();
_rocksReadOptions.prefix_same_as_start = true; // should always be true
if (isReadOnlyTransaction()) {
// server wide replication may insert a snapshot
if (_snapshot == nullptr) {
// we must call ReleaseSnapshot at some point
_snapshot = db->GetSnapshot();
}
TRI_ASSERT(_snapshot != nullptr);
_rocksReadOptions.snapshot = _snapshot;
_rocksMethods.reset(new RocksDBReadOnlyMethods(this));
} else {
TRI_ASSERT(_snapshot == nullptr);
createTransaction();
if (hasHint(transaction::Hints::Hint::SINGLE_OPERATION)) {
_rocksReadOptions.snapshot = _rocksTransaction->GetSnapshot();
} else {
TRI_ASSERT(_options.intermediateCommitCount != UINT64_MAX ||
_options.intermediateCommitSize != UINT64_MAX);
// we must call ReleaseSnapshot at some point
_snapshot = db->GetSnapshot();
_rocksReadOptions.snapshot = _snapshot;
TRI_ASSERT(_snapshot != nullptr);
}
TRI_ASSERT(_rocksReadOptions.snapshot != nullptr);
// under some circumstances we can use untracking Put/Delete methods,
// but we need to be sure this does not cause any lost updates or other
// inconsistencies.
// TODO: enable this optimization once these circumstances are clear
// and fully covered by tests
if (false && isExclusiveTransactionOnSingleCollection()) {
_rocksMethods.reset(new RocksDBTrxUntrackedMethods(this));
} else {
_rocksMethods.reset(new RocksDBTrxMethods(this));
}
}
} else {
TRI_ASSERT(_status == transaction::Status::RUNNING);
}
return result;
}
// create a rocksdb transaction. will only be called for write transactions
void RocksDBTransactionState::createTransaction() {
TRI_ASSERT(!isReadOnlyTransaction());
// start rocks transaction
rocksdb::TransactionDB* db = rocksutils::globalRocksDB();
rocksdb::TransactionOptions trxOpts;
trxOpts.set_snapshot = true;
// unclear performance implications do not use for now
// trxOpts.deadlock_detect = !hasHint(transaction::Hints::Hint::NO_DLD);
TRI_ASSERT(_rocksTransaction == nullptr ||
_rocksTransaction->GetState() == rocksdb::Transaction::COMMITED ||
(_rocksTransaction->GetState() == rocksdb::Transaction::STARTED &&
_rocksTransaction->GetNumKeys() == 0));
_rocksTransaction =
db->BeginTransaction(_rocksWriteOptions, trxOpts, _rocksTransaction);
// add transaction begin marker
if (!hasHint(transaction::Hints::Hint::SINGLE_OPERATION)) {
auto header =
RocksDBLogValue::BeginTransaction(_vocbase.id(), _id);
_rocksTransaction->PutLogData(header.slice());
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
TRI_ASSERT(_numLogdata == 0);
++_numLogdata;
#endif
}
}
void RocksDBTransactionState::cleanupTransaction() noexcept {
delete _rocksTransaction;
_rocksTransaction = nullptr;
if (_cacheTx != nullptr) {
// note: endTransaction() will delete _cacheTrx!
CacheManagerFeature::MANAGER->endTransaction(_cacheTx);
_cacheTx = nullptr;
}
if (_snapshot != nullptr) {
rocksdb::TransactionDB* db = rocksutils::globalRocksDB();
db->ReleaseSnapshot(_snapshot);
_snapshot = nullptr;
}
}
arangodb::Result RocksDBTransactionState::internalCommit() {
TRI_ASSERT(_rocksTransaction != nullptr);
ExecContext const* exe = ExecContext::CURRENT;
if (!isReadOnlyTransaction() && exe != nullptr) {
bool cancelRW = !ServerState::writeOpsEnabled() && !exe->isSuperuser();
if (exe->isCanceled() || cancelRW) {
return Result(TRI_ERROR_ARANGO_READ_ONLY, "server is in read-only mode");
}
}
Result result;
if (hasOperations()) {
// we are actually going to attempt a commit
if (!hasHint(transaction::Hints::Hint::SINGLE_OPERATION)) {
// add custom commit marker to increase WAL tailing reliability
auto logValue =
RocksDBLogValue::CommitTransaction(_vocbase.id(), id());
_rocksTransaction->PutLogData(logValue.slice());
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
_numLogdata++;
#endif
}
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
uint64_t x = _numInserts + _numRemoves + _numUpdates;
if (hasHint(transaction::Hints::Hint::SINGLE_OPERATION)) {
TRI_ASSERT(x <= 1 && _numLogdata == x);
} else {
if (_numLogdata != (2 + _numRemoves)) {
LOG_TOPIC(ERR, Logger::FIXME) << "_numInserts " << _numInserts << " "
<< "_numRemoves " << _numRemoves << " "
<< "_numUpdates " << _numUpdates << " "
<< "_numLogdata " << _numLogdata;
}
// begin transaction + commit transaction + n doc removes
TRI_ASSERT(_numLogdata == (2 + _numRemoves));
}
#endif
// set wait for sync flag if required
if (waitForSync()) {
_rocksWriteOptions.sync = true;
_rocksTransaction->SetWriteOptions(_rocksWriteOptions);
}
// prepare for commit on each collection, e.g. place blockers for estimators
rocksdb::SequenceNumber preCommitSeq =
rocksutils::globalRocksDB()->GetLatestSequenceNumber();
for (auto& trxCollection : _collections) {
RocksDBTransactionCollection* collection =
static_cast<RocksDBTransactionCollection*>(trxCollection);
collection->prepareCommit(id(), preCommitSeq);
}
bool committed = false;
auto cleanupCollectionTransactions = [this, &committed]() -> void {
// if we didn't commit, make sure we remove blockers, etc.
if (!committed) {
for (auto& trxCollection : _collections) {
RocksDBTransactionCollection* collection =
static_cast<RocksDBTransactionCollection*>(trxCollection);
collection->abortCommit(id());
}
}
};
TRI_DEFER(cleanupCollectionTransactions());
++_numCommits;
result = rocksutils::convertStatus(_rocksTransaction->Commit());
rocksdb::SequenceNumber latestSeq =
rocksutils::globalRocksDB()->GetLatestSequenceNumber();
if (result.ok()) {
for (auto& trxCollection : _collections) {
RocksDBTransactionCollection* collection =
static_cast<RocksDBTransactionCollection*>(trxCollection);
// we need this in case of an intermediate commit. The number of
// initial documents is adjusted and numInserts / removes is set to 0
// index estimator updates are buffered
collection->commitCounts(id(), latestSeq);
committed = true;
}
}
} else {
TRI_ASSERT(_rocksTransaction->GetNumKeys() == 0 &&
_rocksTransaction->GetNumPuts() == 0 &&
_rocksTransaction->GetNumDeletes() == 0);
for (auto& trxCollection : _collections) {
RocksDBTransactionCollection* collection =
static_cast<RocksDBTransactionCollection*>(trxCollection);
// We get here if we have filled indexes. So let us commit counts and
// any buffered index estimator updates
collection->commitCounts(id(), 0);
}
// don't write anything if the transaction is empty
result = rocksutils::convertStatus(_rocksTransaction->Rollback());
}
return result;
}
/// @brief commit a transaction
Result RocksDBTransactionState::commitTransaction(
transaction::Methods* activeTrx) {
LOG_TRX(this, _nestingLevel)
<< "committing " << AccessMode::typeString(_type) << " transaction";
TRI_ASSERT(_status == transaction::Status::RUNNING);
TRI_IF_FAILURE("TransactionWriteCommitMarker") {
return Result(TRI_ERROR_DEBUG);
}
arangodb::Result res;
if (_nestingLevel == 0) {
if (_rocksTransaction != nullptr) {
res = internalCommit();
}
if (res.ok()) {
updateStatus(transaction::Status::COMMITTED);
cleanupTransaction(); // deletes trx
} else {
abortTransaction(activeTrx); // deletes trx
}
TRI_ASSERT(!_rocksTransaction && !_cacheTx && !_snapshot);
}
unuseCollections(_nestingLevel);
return res;
}
/// @brief abort and rollback a transaction
Result RocksDBTransactionState::abortTransaction(
transaction::Methods* activeTrx) {
LOG_TRX(this, _nestingLevel)
<< "aborting " << AccessMode::typeString(_type) << " transaction";
TRI_ASSERT(_status == transaction::Status::RUNNING);
Result result;
if (_nestingLevel == 0) {
if (_rocksTransaction != nullptr) {
rocksdb::Status status = _rocksTransaction->Rollback();
result = rocksutils::convertStatus(status);
}
cleanupTransaction(); // deletes trx
updateStatus(transaction::Status::ABORTED);
if (hasOperations()) {
// must clean up the query cache because the transaction
// may have queried something via AQL that is now rolled back
clearQueryCache();
}
TRI_ASSERT(!_rocksTransaction && !_cacheTx && !_snapshot);
}
unuseCollections(_nestingLevel);
return result;
}
void RocksDBTransactionState::prepareOperation(TRI_voc_cid_t cid, TRI_voc_rid_t rid,
TRI_voc_document_operation_e operationType) {
TRI_ASSERT(!isReadOnlyTransaction());
bool singleOp = hasHint(transaction::Hints::Hint::SINGLE_OPERATION);
if (singleOp) {
// singleOp => no modifications yet
TRI_ASSERT(_rocksTransaction->GetNumPuts() == 0 &&
_rocksTransaction->GetNumDeletes() == 0);
switch (operationType) {
case TRI_VOC_DOCUMENT_OPERATION_INSERT:
case TRI_VOC_DOCUMENT_OPERATION_UPDATE:
case TRI_VOC_DOCUMENT_OPERATION_REPLACE: {
auto logValue =
RocksDBLogValue::SinglePut(_vocbase.id(), cid);
_rocksTransaction->PutLogData(logValue.slice());
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
TRI_ASSERT(_numLogdata == 0);
_numLogdata++;
#endif
break;
}
case TRI_VOC_DOCUMENT_OPERATION_REMOVE: {
TRI_ASSERT(rid != 0);
auto logValue =
RocksDBLogValue::SingleRemoveV2(_vocbase.id(), cid, rid);
_rocksTransaction->PutLogData(logValue.slice());
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
TRI_ASSERT(_numLogdata == 0);
_numLogdata++;
#endif
} break;
case TRI_VOC_DOCUMENT_OPERATION_UNKNOWN:
break;
}
} else {
if (operationType == TRI_VOC_DOCUMENT_OPERATION_REMOVE) {
RocksDBLogValue logValue = RocksDBLogValue::DocumentRemoveV2(rid);
_rocksTransaction->PutLogData(logValue.slice());
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
++_numLogdata;
#endif
}
}
}
/// @brief add an operation for a transaction collection
Result RocksDBTransactionState::addOperation(
TRI_voc_cid_t cid, TRI_voc_rid_t revisionId,
TRI_voc_document_operation_e operationType) {
size_t currentSize = _rocksTransaction->GetWriteBatch()->GetWriteBatch()->GetDataSize();
if (currentSize > _options.maxTransactionSize) {
// we hit the transaction size limit
std::string message =
"aborting transaction because maximal transaction size limit of " +
std::to_string(_options.maxTransactionSize) + " bytes is reached";
return Result(Result(TRI_ERROR_RESOURCE_LIMIT, message));
}
auto collection =
static_cast<RocksDBTransactionCollection*>(findCollection(cid));
if (collection == nullptr) {
std::string message = "collection '" + std::to_string(cid) +
"' not found in transaction state";
THROW_ARANGO_EXCEPTION_MESSAGE(TRI_ERROR_INTERNAL, message);
}
// should not fail or fail with exception
collection->addOperation(operationType, revisionId);
// clear the query cache for this collection
if (arangodb::aql::QueryCache::instance()->mayBeActive()) {
arangodb::aql::QueryCache::instance()->invalidate(
&_vocbase, collection->collectionName()
);
}
switch (operationType) {
case TRI_VOC_DOCUMENT_OPERATION_UNKNOWN:
break;
case TRI_VOC_DOCUMENT_OPERATION_INSERT:
++_numInserts;
break;
case TRI_VOC_DOCUMENT_OPERATION_UPDATE:
case TRI_VOC_DOCUMENT_OPERATION_REPLACE:
++_numUpdates;
break;
case TRI_VOC_DOCUMENT_OPERATION_REMOVE:
++_numRemoves;
break;
}
// perform an intermediate commit if necessary
return checkIntermediateCommit(currentSize);
}
RocksDBMethods* RocksDBTransactionState::rocksdbMethods() {
TRI_ASSERT(_rocksMethods);
return _rocksMethods.get();
}
void RocksDBTransactionState::donateSnapshot(rocksdb::Snapshot const* snap) {
TRI_ASSERT(_snapshot == nullptr);
TRI_ASSERT(isReadOnlyTransaction());
TRI_ASSERT(_status == transaction::Status::CREATED);
_snapshot = snap;
}
rocksdb::Snapshot const* RocksDBTransactionState::stealSnapshot() {
TRI_ASSERT(_snapshot != nullptr);
TRI_ASSERT(isReadOnlyTransaction());
TRI_ASSERT(_status == transaction::Status::RUNNING);
rocksdb::Snapshot const* snap = _snapshot;
_snapshot = nullptr;
return snap;
}
uint64_t RocksDBTransactionState::sequenceNumber() const {
if (_snapshot != nullptr) {
return static_cast<uint64_t>(_snapshot->GetSequenceNumber());
} else if (_rocksTransaction) {
return static_cast<uint64_t>(
_rocksTransaction->GetSnapshot()->GetSequenceNumber());
}
TRI_ASSERT(false);
THROW_ARANGO_EXCEPTION_MESSAGE(TRI_ERROR_INTERNAL, "No snapshot set");
}
Result RocksDBTransactionState::triggerIntermediateCommit() {
TRI_IF_FAILURE("FailBeforeIntermediateCommit") {
THROW_ARANGO_EXCEPTION(TRI_ERROR_DEBUG);
}
TRI_IF_FAILURE("SegfaultBeforeIntermediateCommit") {
TRI_SegfaultDebugging("SegfaultBeforeIntermediateCommit");
}
TRI_ASSERT(!hasHint(transaction::Hints::Hint::SINGLE_OPERATION));
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
LOG_TOPIC(DEBUG, Logger::ROCKSDB) << "INTERMEDIATE COMMIT!";
#endif
Result res = internalCommit();
if (res.fail()) {
// FIXME: do we abort the transaction ?
return res;
}
TRI_IF_FAILURE("FailAfterIntermediateCommit") {
THROW_ARANGO_EXCEPTION(TRI_ERROR_DEBUG);
}
TRI_IF_FAILURE("SegfaultAfterIntermediateCommit") {
TRI_SegfaultDebugging("SegfaultAfterIntermediateCommit");
}
_numInserts = 0;
_numUpdates = 0;
_numRemoves = 0;
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
_numLogdata = 0;
#endif
createTransaction();
return TRI_ERROR_NO_ERROR;
}
Result RocksDBTransactionState::checkIntermediateCommit(uint64_t newSize) {
auto numOperations = _numInserts + _numUpdates + _numRemoves;
// perform an intermediate commit
// this will be done if either the "number of operations" or the
// "transaction size" counters have reached their limit
if (_options.intermediateCommitCount <= numOperations ||
_options.intermediateCommitSize <= newSize) {
return triggerIntermediateCommit();
}
return TRI_ERROR_NO_ERROR;
}
/// @brief temporarily lease a Builder object
RocksDBKey* RocksDBTransactionState::leaseRocksDBKey() {
if (_keys.empty()) {
// create a new key and return it
return new RocksDBKey();
}
// re-use an existing builder
RocksDBKey* k = _keys.back();
_keys.pop_back();
return k;
}
/// @brief return a temporary RocksDBKey object
void RocksDBTransactionState::returnRocksDBKey(RocksDBKey* key) {
try {
// put key back into our vector of keys
_keys.emplace_back(key);
} catch (...) {
// no harm done. just wipe the key
delete key;
}
}
void RocksDBTransactionState::trackIndexInsert(TRI_voc_cid_t cid,
TRI_idx_iid_t idxId,
uint64_t hash) {
auto col = findCollection(cid);
if (col != nullptr) {
static_cast<RocksDBTransactionCollection*>(col)->trackIndexInsert(idxId,
hash);
} else {
TRI_ASSERT(false);
}
}
void RocksDBTransactionState::trackIndexRemove(TRI_voc_cid_t cid,
TRI_idx_iid_t idxId,
uint64_t hash) {
auto col = findCollection(cid);
if (col != nullptr) {
static_cast<RocksDBTransactionCollection*>(col)->trackIndexRemove(idxId,
hash);
} else {
TRI_ASSERT(false);
}
}
/// @brief constructor, leases a builder
RocksDBKeyLeaser::RocksDBKeyLeaser(transaction::Methods* trx)
: _rtrx(RocksDBTransactionState::toState(trx)),
_parallel(_rtrx->inParallelMode()),
_key(_parallel ? &_internal : _rtrx->leaseRocksDBKey()) {
TRI_ASSERT(_key != nullptr);
}
/// @brief destructor
RocksDBKeyLeaser::~RocksDBKeyLeaser() {
if (!_parallel && _key != nullptr) {
_rtrx->returnRocksDBKey(_key);
}
}
Reference in New Issue View Git Blame Copy Permalink