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

2013 lines
63 KiB
C++
Raw Blame History

////////////////////////////////////////////////////////////////////////////////
/// DISCLAIMER
///
/// Copyright 2017 ArangoDB 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 Christoph Uhde
////////////////////////////////////////////////////////////////////////////////
#include "RocksDBCollection.h"
#include "Aql/PlanCache.h"
#include "Basics/ReadLocker.h"
#include "Basics/Result.h"
#include "Basics/StaticStrings.h"
#include "Basics/StringUtils.h"
#include "Basics/VelocyPackHelper.h"
#include "Basics/WriteLocker.h"
#include "Cache/CacheManagerFeature.h"
#include "Cache/Common.h"
#include "Cache/Manager.h"
#include "Cache/TransactionalCache.h"
#include "Cluster/ClusterMethods.h"
#include "Indexes/Index.h"
#include "Indexes/IndexIterator.h"
#include "RestServer/DatabaseFeature.h"
#include "RocksDBEngine/RocksDBPrimaryIndex.h"
#include "RocksDBEngine/RocksDBCommon.h"
#include "RocksDBEngine/RocksDBComparator.h"
#include "RocksDBEngine/RocksDBEngine.h"
#include "RocksDBEngine/RocksDBIterators.h"
#include "RocksDBEngine/RocksDBKey.h"
#include "RocksDBEngine/RocksDBLogValue.h"
#include "RocksDBEngine/RocksDBMethods.h"
#include "RocksDBEngine/RocksDBPrimaryIndex.h"
#include "RocksDBEngine/RocksDBSettingsManager.h"
#include "RocksDBEngine/RocksDBTransactionCollection.h"
#include "RocksDBEngine/RocksDBTransactionState.h"
#include "RocksDBEngine/RocksDBValue.h"
#include "StorageEngine/EngineSelectorFeature.h"
#include "StorageEngine/StorageEngine.h"
#include "StorageEngine/TransactionState.h"
#include "Transaction/Helpers.h"
#include "Transaction/StandaloneContext.h"
#include "Utils/CollectionNameResolver.h"
#include "Utils/Events.h"
#include "Utils/OperationOptions.h"
#include "Utils/SingleCollectionTransaction.h"
#include "VocBase/KeyGenerator.h"
#include "VocBase/LocalDocumentId.h"
#include "VocBase/LogicalCollection.h"
#include "VocBase/ManagedDocumentResult.h"
#include "VocBase/ticks.h"
#include "VocBase/voc-types.h"
#include <rocksdb/utilities/transaction.h>
#include <rocksdb/utilities/transaction_db.h>
#include <rocksdb/utilities/write_batch_with_index.h>
#include <velocypack/Iterator.h>
#include <velocypack/velocypack-aliases.h>
using namespace arangodb;
RocksDBCollection::RocksDBCollection(
LogicalCollection& collection,
arangodb::velocypack::Slice const& info
)
: PhysicalCollection(collection, info),
_objectId(basics::VelocyPackHelper::stringUInt64(info, "objectId")),
_numberDocuments(0),
_revisionId(0),
_primaryIndex(nullptr),
_cache(nullptr),
_cachePresent(false),
_cacheEnabled(!collection.system() &&
basics::VelocyPackHelper::readBooleanValue(
info, "cacheEnabled", false) &&
CacheManagerFeature::MANAGER != nullptr) {
TRI_ASSERT(!ServerState::instance()->isCoordinator());
VPackSlice s = info.get("isVolatile");
if (s.isBoolean() && s.getBoolean()) {
THROW_ARANGO_EXCEPTION_MESSAGE(
TRI_ERROR_BAD_PARAMETER,
"volatile collections are unsupported in the RocksDB engine");
}
TRI_ASSERT(_logicalCollection.isAStub() || _objectId != 0);
rocksutils::globalRocksEngine()->addCollectionMapping(
_objectId, _logicalCollection.vocbase().id(), _logicalCollection.id()
);
if (_cacheEnabled) {
createCache();
}
}
RocksDBCollection::RocksDBCollection(
LogicalCollection& collection,
PhysicalCollection const* physical
)
: PhysicalCollection(collection, VPackSlice::emptyObjectSlice()),
_objectId(static_cast<RocksDBCollection const*>(physical)->_objectId),
_numberDocuments(0),
_revisionId(0),
_primaryIndex(nullptr),
_cache(nullptr),
_cachePresent(false),
_cacheEnabled(
static_cast<RocksDBCollection const*>(physical)->_cacheEnabled &&
CacheManagerFeature::MANAGER != nullptr) {
TRI_ASSERT(!ServerState::instance()->isCoordinator());
rocksutils::globalRocksEngine()->addCollectionMapping(
_objectId, _logicalCollection.vocbase().id(), _logicalCollection.id()
);
if (_cacheEnabled) {
createCache();
}
}
RocksDBCollection::~RocksDBCollection() {
if (useCache()) {
try {
destroyCache();
} catch (...) {
}
}
}
std::string const& RocksDBCollection::path() const {
return StaticStrings::Empty; // we do not have any path
}
void RocksDBCollection::setPath(std::string const&) {
// we do not have any path
}
Result RocksDBCollection::updateProperties(VPackSlice const& slice,
bool doSync) {
auto isSys = _logicalCollection.system();
_cacheEnabled = !isSys && basics::VelocyPackHelper::readBooleanValue(
slice, "cacheEnabled", _cacheEnabled) &&
CacheManagerFeature::MANAGER != nullptr;
primaryIndex()->setCacheEnabled(_cacheEnabled);
if (_cacheEnabled) {
createCache();
primaryIndex()->createCache();
} else {
// will do nothing if cache is not present
destroyCache();
primaryIndex()->destroyCache();
TRI_ASSERT(_cache.get() == nullptr);
}
// nothing else to do
return TRI_ERROR_NO_ERROR;
}
arangodb::Result RocksDBCollection::persistProperties() {
// only code path calling this causes these properties to be
// already written in RocksDBEngine::changeCollection()
return Result();
}
PhysicalCollection* RocksDBCollection::clone(LogicalCollection& logical) const {
return new RocksDBCollection(logical, this);
}
/// @brief export properties
void RocksDBCollection::getPropertiesVPack(velocypack::Builder& result) const {
TRI_ASSERT(result.isOpenObject());
result.add("objectId", VPackValue(std::to_string(_objectId)));
result.add("cacheEnabled", VPackValue(_cacheEnabled));
TRI_ASSERT(result.isOpenObject());
}
/// @brief closes an open collection
int RocksDBCollection::close() {
READ_LOCKER(guard, _indexesLock);
for (auto it : _indexes) {
it->unload();
}
return TRI_ERROR_NO_ERROR;
}
void RocksDBCollection::load() {
if (_cacheEnabled) {
createCache();
if (_cachePresent) {
uint64_t numDocs = numberDocuments();
if (numDocs > 0) {
_cache->sizeHint(static_cast<uint64_t>(0.3 * numDocs));
}
}
}
READ_LOCKER(guard, _indexesLock);
for (auto it : _indexes) {
it->load();
}
}
void RocksDBCollection::unload() {
if (useCache()) {
destroyCache();
TRI_ASSERT(!_cachePresent);
}
READ_LOCKER(guard, _indexesLock);
for (auto it : _indexes) {
it->unload();
}
}
TRI_voc_rid_t RocksDBCollection::revision() const { return _revisionId; }
TRI_voc_rid_t RocksDBCollection::revision(transaction::Methods* trx) const {
auto state = RocksDBTransactionState::toState(trx);
auto trxCollection = static_cast<RocksDBTransactionCollection*>(
state->findCollection(_logicalCollection.id())
);
TRI_ASSERT(trxCollection != nullptr);
return trxCollection->revision();
}
uint64_t RocksDBCollection::numberDocuments() const { return _numberDocuments; }
uint64_t RocksDBCollection::numberDocuments(transaction::Methods* trx) const {
TRI_ASSERT(!ServerState::instance()->isCoordinator());
auto state = RocksDBTransactionState::toState(trx);
auto trxCollection = static_cast<RocksDBTransactionCollection*>(
state->findCollection(_logicalCollection.id())
);
TRI_ASSERT(trxCollection != nullptr);
return trxCollection->numberDocuments();
}
/// @brief report extra memory used by indexes etc.
size_t RocksDBCollection::memory() const { return 0; }
void RocksDBCollection::open(bool /*ignoreErrors*/) {
TRI_ASSERT(_objectId != 0);
RocksDBEngine* engine =
static_cast<RocksDBEngine*>(EngineSelectorFeature::ENGINE);
TRI_ASSERT(engine != nullptr);
if (!engine->inRecovery()) {
loadInitialNumberDocuments();
}
}
void RocksDBCollection::prepareIndexes(
arangodb::velocypack::Slice indexesSlice) {
WRITE_LOCKER(guard, _indexesLock);
TRI_ASSERT(indexesSlice.isArray());
StorageEngine* engine = EngineSelectorFeature::ENGINE;
std::vector<std::shared_ptr<Index>> indexes;
if (indexesSlice.length() == 0 && _indexes.empty()) {
engine->indexFactory().fillSystemIndexes(_logicalCollection, indexes);
} else {
engine->indexFactory().prepareIndexes(
_logicalCollection, indexesSlice, indexes
);
}
for (std::shared_ptr<Index>& idx : indexes) {
addIndex(std::move(idx));
}
if (_indexes[0]->type() != Index::IndexType::TRI_IDX_TYPE_PRIMARY_INDEX ||
(TRI_COL_TYPE_EDGE == _logicalCollection.type() &&
(_indexes[1]->type() != Index::IndexType::TRI_IDX_TYPE_EDGE_INDEX ||
_indexes[2]->type() != Index::IndexType::TRI_IDX_TYPE_EDGE_INDEX))) {
std::string msg = "got invalid indexes for collection '"
+ _logicalCollection.name() + "'";
LOG_TOPIC(ERR, arangodb::Logger::ENGINES) << msg;
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
for (auto it : _indexes) {
LOG_TOPIC(ERR, arangodb::Logger::ENGINES) << "- " << it.get();
}
#endif
THROW_ARANGO_EXCEPTION_MESSAGE(TRI_ERROR_INTERNAL, msg);
}
TRI_ASSERT(!_indexes.empty());
}
static std::shared_ptr<Index> findIndex(
velocypack::Slice const& info,
std::vector<std::shared_ptr<Index>> const& indexes) {
TRI_ASSERT(info.isObject());
auto value = info.get(arangodb::StaticStrings::IndexType); // extract type
if (!value.isString()) {
// Compatibility with old v8-vocindex.
THROW_ARANGO_EXCEPTION_MESSAGE(TRI_ERROR_INTERNAL,
"invalid index type definition");
}
std::string tmp = value.copyString();
arangodb::Index::IndexType const type = arangodb::Index::type(tmp.c_str());
for (auto const& idx : indexes) {
if (idx->type() == type) {
// Only check relevant indexes
if (idx->matchesDefinition(info)) {
// We found an index for this definition.
return idx;
}
}
}
return nullptr;
}
/// @brief Find index by definition
std::shared_ptr<Index> RocksDBCollection::lookupIndex(
velocypack::Slice const& info) const {
READ_LOCKER(guard, _indexesLock);
return findIndex(info, _indexes);
}
std::shared_ptr<Index> RocksDBCollection::createIndex(
transaction::Methods* trx, arangodb::velocypack::Slice const& info,
bool& created) {
// prevent concurrent dropping
bool isLocked =
trx->isLocked(&_logicalCollection, AccessMode::Type::EXCLUSIVE);
CONDITIONAL_WRITE_LOCKER(guard, _exclusiveLock, !isLocked);
std::shared_ptr<Index> idx;
{
WRITE_LOCKER(guard, _indexesLock);
idx = findIndex(info, _indexes);
if (idx) {
created = false; // We already have this index.
return idx;
}
}
RocksDBEngine* engine = static_cast<RocksDBEngine*>(EngineSelectorFeature::ENGINE);
// We are sure that we do not have an index of this type.
// We also hold the lock. Create it
idx = engine->indexFactory().prepareIndexFromSlice(
info, true, _logicalCollection, false
);
if (!idx) {
THROW_ARANGO_EXCEPTION(TRI_ERROR_ARANGO_INDEX_CREATION_FAILED);
}
// we cannot persist primary or edge indexes
TRI_ASSERT(idx->type() != Index::IndexType::TRI_IDX_TYPE_PRIMARY_INDEX);
TRI_ASSERT(idx->type() != Index::IndexType::TRI_IDX_TYPE_EDGE_INDEX);
Result res = fillIndexes(trx, idx);
if (!res.ok()) {
THROW_ARANGO_EXCEPTION(res);
}
// we need to sync the selectivity estimates
res = engine->settingsManager()->sync(false);
if (res.fail()) {
LOG_TOPIC(WARN, Logger::ENGINES) << "could not sync settings: "
<< res.errorMessage();
}
rocksdb::Status s = engine->db()->GetRootDB()->FlushWAL(true);
if (!s.ok()) {
LOG_TOPIC(WARN, Logger::ENGINES) << "could not flush wal: "
<< s.ToString();
}
#if USE_PLAN_CACHE
arangodb::aql::PlanCache::instance()->invalidate(
_logicalCollection->vocbase());
#endif
// Until here no harm is done if something fails. The shared_ptr will
// clean up, if left before
{
WRITE_LOCKER(guard, _indexesLock);
addIndex(idx);
}
auto builder = _logicalCollection.toVelocyPackIgnore(
{"path", "statusString"}, true, /*forPersistence*/ true);
VPackBuilder indexInfo;
idx->toVelocyPack(indexInfo, Index::makeFlags(Index::Serialize::ObjectId));
res = engine->writeCreateCollectionMarker(
_logicalCollection.vocbase().id(),
_logicalCollection.id(),
builder.slice(),
RocksDBLogValue::IndexCreate(
_logicalCollection.vocbase().id(),
_logicalCollection.id(),
indexInfo.slice()
)
);
if (res.fail()) {
// We could not persist the index creation. Better abort
// Remove the Index in the local list again.
size_t i = 0;
WRITE_LOCKER(guard, _indexesLock);
for (auto index : _indexes) {
if (index == idx) {
_indexes.erase(_indexes.begin() + i);
break;
}
++i;
}
idx->drop();
THROW_ARANGO_EXCEPTION(res);
}
created = true;
return idx;
}
/// @brief Restores an index from VelocyPack.
int RocksDBCollection::restoreIndex(transaction::Methods* trx,
velocypack::Slice const& info,
std::shared_ptr<Index>& idx) {
// The coordinator can never get into this state!
TRI_ASSERT(!ServerState::instance()->isCoordinator());
idx.reset(); // Clear it to make sure.
if (!info.isObject()) {
return TRI_ERROR_INTERNAL;
}
// check if we already have this index
auto oldIdx = lookupIndex(info);
if (oldIdx) {
idx = oldIdx;
return TRI_ERROR_NO_ERROR;
}
RocksDBEngine* engine = static_cast<RocksDBEngine*>(EngineSelectorFeature::ENGINE);
// We create a new Index object to make sure that the index
// is not handed out except for a successful case.
std::shared_ptr<Index> newIdx;
try {
newIdx = engine->indexFactory().prepareIndexFromSlice(
info, false, _logicalCollection, false
);
} catch (arangodb::basics::Exception const& e) {
// Something with index creation went wrong.
// Just report.
return e.code();
}
if (!newIdx) { // simon: probably something wrong with ArangoSearch Links
LOG_TOPIC(ERR, Logger::ENGINES) << "index creation failed while restoring";
return TRI_ERROR_ARANGO_INDEX_CREATION_FAILED;
}
TRI_ASSERT(newIdx != nullptr);
auto const id = newIdx->id();
TRI_UpdateTickServer(id);
for (auto& it : _indexes) {
if (it->id() == id) {
// index already exists
idx = it;
return TRI_ERROR_NO_ERROR;
}
}
TRI_ASSERT(newIdx.get()->type() !=
Index::IndexType::TRI_IDX_TYPE_PRIMARY_INDEX);
Result res = fillIndexes(trx, newIdx);
if (!res.ok()) {
return res.errorNumber();
}
// we need to sync the selectivity estimates
res = engine->settingsManager()->sync(false);
if (res.fail()) {
LOG_TOPIC(WARN, Logger::ENGINES) << "could not sync settings: "
<< res.errorMessage();
}
rocksdb::Status s = engine->db()->GetRootDB()->FlushWAL(true);
if (!s.ok()) {
LOG_TOPIC(WARN, Logger::ENGINES) << "could not flush wal: "
<< s.ToString();
}
addIndex(newIdx);
auto builder = _logicalCollection.toVelocyPackIgnore(
{"path", "statusString"}, true, /*forPersistence*/ true);
VPackBuilder indexInfo;
newIdx->toVelocyPack(indexInfo, Index::makeFlags(Index::Serialize::ObjectId));
TRI_ASSERT(engine != nullptr);
res = engine->writeCreateCollectionMarker(
_logicalCollection.vocbase().id(),
_logicalCollection.id(),
builder.slice(),
RocksDBLogValue::IndexCreate(
_logicalCollection.vocbase().id(),
_logicalCollection.id(),
indexInfo.slice()
)
);
if (res.fail()) {
// We could not persist the index creation. Better abort
// Remove the Index in the local list again.
size_t i = 0;
WRITE_LOCKER(guard, _indexesLock);
for (auto index : _indexes) {
if (index == newIdx) {
_indexes.erase(_indexes.begin() + i);
break;
}
++i;
}
newIdx->drop();
return res.errorNumber();
}
idx = newIdx;
return TRI_ERROR_NO_ERROR;
}
/// @brief Drop an index with the given iid.
bool RocksDBCollection::dropIndex(TRI_idx_iid_t iid) {
// usually always called when _exclusiveLock is held
if (iid == 0) {
// invalid index id or primary index
return true;
}
size_t i = 0;
WRITE_LOCKER(guard, _indexesLock);
for (std::shared_ptr<Index> index : _indexes) {
RocksDBIndex* cindex = static_cast<RocksDBIndex*>(index.get());
TRI_ASSERT(cindex != nullptr);
if (iid == cindex->id()) {
int rv = cindex->drop();
if (rv == TRI_ERROR_NO_ERROR) {
// trigger compaction before deleting the object
cindex->cleanup();
_indexes.erase(_indexes.begin() + i);
events::DropIndex("", std::to_string(iid), TRI_ERROR_NO_ERROR);
// toVelocyPackIgnore will take a read lock and we don't need the
// lock anymore, this branch always returns
guard.unlock();
auto engine = static_cast<RocksDBEngine*>(EngineSelectorFeature::ENGINE);
engine->removeIndexMapping(cindex->objectId());
auto builder = _logicalCollection.toVelocyPackIgnore(
{"path", "statusString"}, true, true);
// log this event in the WAL and in the collection meta-data
int res = engine->writeCreateCollectionMarker(
_logicalCollection.vocbase().id(),
_logicalCollection.id(),
builder.slice(),
RocksDBLogValue::IndexDrop(
_logicalCollection.vocbase().id(), _logicalCollection.id(), iid
)
);
return res == TRI_ERROR_NO_ERROR;
}
break;
}
++i;
}
// We tried to remove an index that does not exist
events::DropIndex("", std::to_string(iid), TRI_ERROR_ARANGO_INDEX_NOT_FOUND);
return false;
}
std::unique_ptr<IndexIterator> RocksDBCollection::getAllIterator(transaction::Methods* trx) const {
return std::unique_ptr<IndexIterator>(
new RocksDBAllIndexIterator(&_logicalCollection, trx, primaryIndex())
);
}
std::unique_ptr<IndexIterator> RocksDBCollection::getAnyIterator(
transaction::Methods* trx) const {
return std::unique_ptr<IndexIterator>(
new RocksDBAnyIndexIterator(&_logicalCollection, trx, primaryIndex())
);
}
void RocksDBCollection::invokeOnAllElements(
transaction::Methods* trx,
std::function<bool(LocalDocumentId const&)> callback) {
std::unique_ptr<IndexIterator> cursor(this->getAllIterator(trx));
bool cnt = true;
auto cb = [&](LocalDocumentId token) {
if (cnt) {
cnt = callback(token);
}
};
while (cursor->next(cb, 1000) && cnt) {
}
}
////////////////////////////////////
// -- SECTION DML Operations --
///////////////////////////////////
Result RocksDBCollection::truncate(transaction::Methods* trx,
OperationOptions& options) {
TRI_ASSERT(_objectId != 0);
auto state = RocksDBTransactionState::toState(trx);
RocksDBMethods* mthds = state->rocksdbMethods();
if (state->isExclusiveTransactionOnSingleCollection() &&
state->hasHint(transaction::Hints::Hint::ALLOW_RANGE_DELETE) &&
static_cast<RocksDBEngine*>(EngineSelectorFeature::ENGINE)->canUseRangeDeleteInWal() &&
_numberDocuments >= 32 * 1024) {
// non-transactional truncate optimization. We perform a bunch of
// range deletes and circumwent the normal rocksdb::Transaction.
// no savepoint needed here
TRI_ASSERT(!state->hasOperations()); // not allowed
TRI_IF_FAILURE("RocksDBRemoveLargeRangeOn") {
return Result(TRI_ERROR_DEBUG);
}
RocksDBEngine* engine = rocksutils::globalRocksEngine();
// add the assertion again here, so we are sure we can use RangeDeletes
TRI_ASSERT(engine->canUseRangeDeleteInWal());
rocksdb::DB* db = engine->db()->GetRootDB();
TRI_IF_FAILURE("RocksDBCollection::truncate::forceSync") {
engine->settingsManager()->sync(false);
}
// pre commit sequence needed to place a blocker
rocksdb::SequenceNumber seq = rocksutils::latestSequenceNumber();
auto guard = scopeGuard([&] { // remove blocker afterwards
_meta.removeBlocker(state->id());
});
_meta.placeBlocker(state->id(), seq);
rocksdb::WriteBatch batch;
// delete documents
RocksDBKeyBounds bounds = RocksDBKeyBounds::CollectionDocuments(_objectId);
rocksdb::Status s = batch.DeleteRange(bounds.columnFamily(), bounds.start(), bounds.end());
if (!s.ok()) {
return rocksutils::convertStatus(s);
}
// delete indexes, place estimator blockers
{
READ_LOCKER(guard, _indexesLock);
for (std::shared_ptr<Index> const& idx : _indexes) {
RocksDBIndex* ridx = static_cast<RocksDBIndex*>(idx.get());
bounds = ridx->getBounds();
s = batch.DeleteRange(bounds.columnFamily(), bounds.start(), bounds.end());
if (!s.ok()) {
return rocksutils::convertStatus(s);
}
}
}
// add the log entry so we can recover the correct count
auto log = RocksDBLogValue::CollectionTruncate(trx->vocbase().id(),
_logicalCollection.id(), _objectId);
s = batch.PutLogData(log.slice());
if (!s.ok()) {
return rocksutils::convertStatus(s);
}
rocksdb::WriteOptions wo;
s = db->Write(wo, &batch);
if (!s.ok()) {
return rocksutils::convertStatus(s);
}
seq = db->GetLatestSequenceNumber() - 1; // post commit sequence
uint64_t numDocs = _numberDocuments.exchange(0);
_meta.adjustNumberDocuments(seq, /*revision*/newRevisionId(), - static_cast<int64_t>(numDocs));
{
READ_LOCKER(guard, _indexesLock);
for (std::shared_ptr<Index> const& idx : _indexes) {
idx->afterTruncate(seq); // clears caches / clears links (if applicable)
}
}
guard.fire(); // remove blocker
if (numDocs > 64 * 1024) {
// also compact the ranges in order to speed up all further accesses
compact();
}
TRI_ASSERT(!state->hasOperations()); // not allowed
return Result{};
}
TRI_IF_FAILURE("RocksDBRemoveLargeRangeOff") {
return Result(TRI_ERROR_DEBUG);
}
// normal transactional truncate
RocksDBKeyBounds documentBounds =
RocksDBKeyBounds::CollectionDocuments(_objectId);
rocksdb::Comparator const* cmp =
RocksDBColumnFamily::documents()->GetComparator();
rocksdb::ReadOptions ro = mthds->iteratorReadOptions();
rocksdb::Slice const end = documentBounds.end();
ro.iterate_upper_bound = &end;
// avoid OOM error for truncate by committing earlier
uint64_t const prvICC = state->options().intermediateCommitCount;
state->options().intermediateCommitCount = std::min<uint64_t>(prvICC, 10000);
std::unique_ptr<rocksdb::Iterator> iter =
mthds->NewIterator(ro, documentBounds.columnFamily());
iter->Seek(documentBounds.start());
uint64_t found = 0;
while (iter->Valid() && cmp->Compare(iter->key(), end) < 0) {
++found;
TRI_ASSERT(_objectId == RocksDBKey::objectId(iter->key()));
VPackSlice doc = VPackSlice(iter->value().data());
TRI_ASSERT(doc.isObject());
// To print the WAL we need key and RID
VPackSlice key;
TRI_voc_rid_t rid = 0;
transaction::helpers::extractKeyAndRevFromDocument(doc, key, rid);
TRI_ASSERT(key.isString());
TRI_ASSERT(rid != 0);
RocksDBSavePoint guard(trx, TRI_VOC_DOCUMENT_OPERATION_REMOVE);
state->prepareOperation(_logicalCollection.id(),
rid, // actual revision ID!!
TRI_VOC_DOCUMENT_OPERATION_REMOVE);
LocalDocumentId const docId = RocksDBKey::documentId(iter->key());
auto res = removeDocument(trx, docId, doc, options);
if (res.fail()) { // Failed to remove document in truncate.
return res;
}
bool hasPerformedIntermediateCommit = false;
res = state->addOperation(_logicalCollection.id(), docId.id(),
TRI_VOC_DOCUMENT_OPERATION_REMOVE, hasPerformedIntermediateCommit);
if (res.fail()) { // This should never happen...
return res;
}
guard.finish(hasPerformedIntermediateCommit);
trackWaitForSync(trx, options);
iter->Next();
}
// reset to previous value after truncate is finished
state->options().intermediateCommitCount = prvICC;
if (found > 64 * 1024) {
// also compact the ranges in order to speed up all further accesses
compact();
}
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
if (state->numCommits() == 0) {
// check IN TRANSACTION if documents have been deleted
if (mthds->countInBounds(RocksDBKeyBounds::CollectionDocuments(_objectId), true)) {
THROW_ARANGO_EXCEPTION_MESSAGE(TRI_ERROR_INTERNAL,
"deletion check in collection truncate "
"failed - not all documents have been "
"deleted");
}
}
#endif
TRI_IF_FAILURE("FailAfterAllCommits") {
return Result(TRI_ERROR_DEBUG);
}
TRI_IF_FAILURE("SegfaultAfterAllCommits") {
TRI_SegfaultDebugging("SegfaultAfterAllCommits");
}
return Result{};
}
LocalDocumentId RocksDBCollection::lookupKey(transaction::Methods* trx,
VPackSlice const& key) const {
TRI_ASSERT(key.isString());
return primaryIndex()->lookupKey(trx, StringRef(key));
}
bool RocksDBCollection::lookupRevision(transaction::Methods* trx,
VPackSlice const& key,
TRI_voc_rid_t& revisionId) const {
TRI_ASSERT(key.isString());
LocalDocumentId documentId;
revisionId = 0;
// lookup the revision id in the primary index
if (!primaryIndex()->lookupRevision(trx, StringRef(key), documentId, revisionId)) {
// document not found
TRI_ASSERT(revisionId == 0);
return false;
}
// document found, but revisionId may not have been present in the primary index
// this can happen for "older" collections
TRI_ASSERT(documentId.isSet());
// now look up the revision id in the actual document data
return readDocumentWithCallback(trx, documentId, [&revisionId](LocalDocumentId const&, VPackSlice doc) {
revisionId = transaction::helpers::extractRevFromDocument(doc);
});
}
Result RocksDBCollection::read(transaction::Methods* trx,
arangodb::StringRef const& key,
ManagedDocumentResult& result, bool) {
LocalDocumentId const documentId = primaryIndex()->lookupKey(trx, key);
if (documentId.isSet()) {
return lookupDocumentVPack(documentId, trx, result, true);
}
// not found
return Result(TRI_ERROR_ARANGO_DOCUMENT_NOT_FOUND);
}
// read using a token!
bool RocksDBCollection::readDocument(transaction::Methods* trx,
LocalDocumentId const& documentId,
ManagedDocumentResult& result) const {
if (documentId.isSet()) {
auto res = lookupDocumentVPack(documentId, trx, result, true);
return res.ok();
}
return false;
}
// read using a token!
bool RocksDBCollection::readDocumentWithCallback(
transaction::Methods* trx, LocalDocumentId const& documentId,
IndexIterator::DocumentCallback const& cb) const {
if (documentId.isSet()) {
return lookupDocumentVPack(documentId, trx, cb, true).ok();
}
return false;
}
Result RocksDBCollection::insert(
arangodb::transaction::Methods* trx,
arangodb::velocypack::Slice const slice,
arangodb::ManagedDocumentResult& mdr, OperationOptions& options,
TRI_voc_tick_t& resultMarkerTick, bool, TRI_voc_tick_t& revisionId,
std::function<Result(void)> callbackDuringLock) {
// store the tick that was used for writing the document
// note that we don't need it for this engine
resultMarkerTick = 0;
LocalDocumentId const documentId = LocalDocumentId::create();
auto isEdgeCollection = (TRI_COL_TYPE_EDGE == _logicalCollection.type());
transaction::BuilderLeaser builder(trx);
Result res(newObjectForInsert(trx, slice, isEdgeCollection,
*builder.get(), options.isRestore, revisionId));
if (res.fail()) {
return res;
}
VPackSlice newSlice = builder->slice();
if (options.overwrite) {
// special optimization for the overwrite case:
// in case the operation is a RepSert, we will first check if the specified
// primary key exists. we can abort this low-level insert early, before any
// modification to the data has been done. this saves us from creating a RocksDB
// transaction SavePoint.
// if we don't do the check here, we will always create a SavePoint first and
// insert the new document. when then inserting the key for the primary index and
// then detecting a unique constraint violation, the transaction would be rolled
// back to the SavePoint state, which will rebuild *all* data in the WriteBatch
// up to the SavePoint. this can be super-expensive for bigger transactions.
// to keep things simple, we are not checking for unique constraint violations
// in secondary indexes here, but defer it to the regular index insertion check
VPackSlice keySlice = newSlice.get(StaticStrings::KeyString);
if (keySlice.isString()) {
LocalDocumentId const documentId = primaryIndex()->lookupKey(trx, StringRef(keySlice));
if (documentId.isSet()) {
if (options.indexOperationMode == Index::OperationMode::internal) {
// need to return the key of the conflict document
return Result(TRI_ERROR_ARANGO_UNIQUE_CONSTRAINT_VIOLATED, keySlice.copyString());
}
return Result(TRI_ERROR_ARANGO_UNIQUE_CONSTRAINT_VIOLATED);
}
}
}
RocksDBSavePoint guard(trx, TRI_VOC_DOCUMENT_OPERATION_INSERT);
auto state = RocksDBTransactionState::toState(trx);
state->prepareOperation(
_logicalCollection.id(), revisionId, TRI_VOC_DOCUMENT_OPERATION_INSERT
);
res = insertDocument(trx, documentId, newSlice, options);
if (res.ok()) {
trackWaitForSync(trx, options);
if (options.silent) {
mdr.clear();
} else {
mdr.setManaged(newSlice.begin(), documentId);
TRI_ASSERT(!mdr.empty());
}
bool hasPerformedIntermediateCommit = false;
auto result = state->addOperation(
_logicalCollection.id(), revisionId, TRI_VOC_DOCUMENT_OPERATION_INSERT,
hasPerformedIntermediateCommit
);
if (result.ok() && callbackDuringLock != nullptr) {
result = callbackDuringLock();
}
if (result.fail()) {
THROW_ARANGO_EXCEPTION(result);
}
guard.finish(hasPerformedIntermediateCommit);
}
return res;
}
Result RocksDBCollection::update(
arangodb::transaction::Methods* trx,
arangodb::velocypack::Slice const newSlice, ManagedDocumentResult& mdr,
OperationOptions& options, TRI_voc_tick_t& resultMarkerTick, bool,
TRI_voc_rid_t& prevRev, ManagedDocumentResult& previous,
arangodb::velocypack::Slice const key,
std::function<Result(void)> callbackDuringLock) {
resultMarkerTick = 0;
LocalDocumentId const documentId = LocalDocumentId::create();
auto isEdgeCollection = (TRI_COL_TYPE_EDGE == _logicalCollection.type());
Result res = this->read(trx, key, previous, /*lock*/false);
if (res.fail()) {
return res;
}
TRI_ASSERT(!previous.empty());
LocalDocumentId const oldDocumentId = previous.localDocumentId();
VPackSlice oldDoc(previous.vpack());
TRI_voc_rid_t const oldRevisionId =
transaction::helpers::extractRevFromDocument(oldDoc);
prevRev = oldRevisionId;
// Check old revision:
if (!options.ignoreRevs) {
TRI_voc_rid_t expectedRev = 0;
if (newSlice.isObject()) {
expectedRev = TRI_ExtractRevisionId(newSlice);
}
int result = checkRevision(trx, expectedRev, prevRev);
if (result != TRI_ERROR_NO_ERROR) {
return Result(result);
}
}
if (newSlice.length() <= 1) {
// shortcut. no need to do anything
mdr = previous;
TRI_ASSERT(!mdr.empty());
trackWaitForSync(trx, options);
return Result();
}
// merge old and new values
TRI_voc_rid_t revisionId;
transaction::BuilderLeaser builder(trx);
res = mergeObjectsForUpdate(trx, oldDoc, newSlice, isEdgeCollection,
options.mergeObjects, options.keepNull, *builder.get(),
options.isRestore, revisionId);
if (res.fail()) {
return res;
}
if (_isDBServer) {
// Need to check that no sharding keys have changed:
if (arangodb::shardKeysChanged(
_logicalCollection,
oldDoc,
builder->slice(),
false
)) {
return Result(TRI_ERROR_CLUSTER_MUST_NOT_CHANGE_SHARDING_ATTRIBUTES);
}
}
VPackSlice const newDoc(builder->slice());
auto state = RocksDBTransactionState::toState(trx);
RocksDBSavePoint guard(trx, TRI_VOC_DOCUMENT_OPERATION_UPDATE);
// add possible log statement under guard
state->prepareOperation(
_logicalCollection.id(), revisionId, TRI_VOC_DOCUMENT_OPERATION_UPDATE
);
res = updateDocument(trx, oldDocumentId, oldDoc, documentId, newDoc, options);
if (res.ok()) {
trackWaitForSync(trx, options);
if (options.silent) {
mdr.clear();
} else {
mdr.setManaged(newDoc.begin(), documentId);
TRI_ASSERT(!mdr.empty());
}
bool hasPerformedIntermediateCommit = false;
auto result = state->addOperation(
_logicalCollection.id(), revisionId, TRI_VOC_DOCUMENT_OPERATION_UPDATE,
hasPerformedIntermediateCommit
);
if (result.ok() && callbackDuringLock != nullptr) {
result = callbackDuringLock();
}
if (result.fail()) {
THROW_ARANGO_EXCEPTION(result);
}
guard.finish(hasPerformedIntermediateCommit);
}
return res;
}
Result RocksDBCollection::replace(
transaction::Methods* trx, arangodb::velocypack::Slice const newSlice,
ManagedDocumentResult& mdr, OperationOptions& options,
TRI_voc_tick_t& resultMarkerTick, bool, TRI_voc_rid_t& prevRev,
ManagedDocumentResult& previous,
std::function<Result(void)> callbackDuringLock) {
resultMarkerTick = 0;
LocalDocumentId const documentId = LocalDocumentId::create();
auto isEdgeCollection = (TRI_COL_TYPE_EDGE == _logicalCollection.type());
// get the previous revision
VPackSlice key = newSlice.get(StaticStrings::KeyString);
if (key.isNone()) {
return Result(TRI_ERROR_ARANGO_DOCUMENT_HANDLE_BAD);
}
// get the previous revision
Result res = this->read(trx, key, previous, /*lock*/false);
if (res.fail()) {
return res;
}
TRI_ASSERT(!previous.empty());
LocalDocumentId const oldDocumentId = previous.localDocumentId();
VPackSlice oldDoc(previous.vpack());
TRI_voc_rid_t oldRevisionId =
transaction::helpers::extractRevFromDocument(oldDoc);
prevRev = oldRevisionId;
// Check old revision:
if (!options.ignoreRevs) {
TRI_voc_rid_t expectedRev = 0;
if (newSlice.isObject()) {
expectedRev = TRI_ExtractRevisionId(newSlice);
}
int res = checkRevision(trx, expectedRev, prevRev);
if (res != TRI_ERROR_NO_ERROR) {
return Result(res);
}
}
// merge old and new values
TRI_voc_rid_t revisionId;
transaction::BuilderLeaser builder(trx);
res = newObjectForReplace(trx, oldDoc, newSlice,
isEdgeCollection, *builder.get(), options.isRestore,
revisionId);
if (res.fail()) {
return res;
}
if (_isDBServer) {
// Need to check that no sharding keys have changed:
if (arangodb::shardKeysChanged(
_logicalCollection,
oldDoc,
builder->slice(),
false
)) {
return Result(TRI_ERROR_CLUSTER_MUST_NOT_CHANGE_SHARDING_ATTRIBUTES);
}
}
VPackSlice const newDoc(builder->slice());
auto state = RocksDBTransactionState::toState(trx);
RocksDBSavePoint guard(trx, TRI_VOC_DOCUMENT_OPERATION_REPLACE);
// add possible log statement under guard
state->prepareOperation(
_logicalCollection.id(), revisionId, TRI_VOC_DOCUMENT_OPERATION_REPLACE
);
Result opResult = updateDocument(trx, oldDocumentId, oldDoc, documentId, newDoc, options);
if (opResult.ok()) {
trackWaitForSync(trx, options);
if (options.silent) {
mdr.clear();
} else {
mdr.setManaged(newDoc.begin(), documentId);
TRI_ASSERT(!mdr.empty());
}
bool hasPerformedIntermediateCommit = false;
auto result = state->addOperation(
_logicalCollection.id(), revisionId, TRI_VOC_DOCUMENT_OPERATION_REPLACE,
hasPerformedIntermediateCommit
);
if (result.ok() && callbackDuringLock != nullptr) {
result = callbackDuringLock();
}
if (result.fail()) {
THROW_ARANGO_EXCEPTION(result);
}
guard.finish(hasPerformedIntermediateCommit);
}
return opResult;
}
Result RocksDBCollection::remove(
arangodb::transaction::Methods* trx, arangodb::velocypack::Slice slice,
arangodb::ManagedDocumentResult& previous, OperationOptions& options,
TRI_voc_tick_t& resultMarkerTick, bool, TRI_voc_rid_t& prevRev,
TRI_voc_rid_t& revisionId, std::function<Result(void)> callbackDuringLock) {
// store the tick that was used for writing the document
// note that we don't need it for this engine
resultMarkerTick = 0;
prevRev = 0;
revisionId = newRevisionId();
VPackSlice key;
if (slice.isString()) {
key = slice;
} else {
key = slice.get(StaticStrings::KeyString);
}
TRI_ASSERT(!key.isNone());
// get the previous revision
Result res = this->read(trx, key, previous, /*lock*/false);
if (res.fail()) {
return res;
}
TRI_ASSERT(!previous.empty());
LocalDocumentId const oldDocumentId = previous.localDocumentId();
VPackSlice oldDoc(previous.vpack());
TRI_voc_rid_t oldRevisionId =
arangodb::transaction::helpers::extractRevFromDocument(oldDoc);
prevRev = oldRevisionId;
// Check old revision:
if (!options.ignoreRevs && slice.isObject()) {
TRI_voc_rid_t expectedRevisionId = TRI_ExtractRevisionId(slice);
int res = checkRevision(trx, expectedRevisionId, oldRevisionId);
if (res != TRI_ERROR_NO_ERROR) {
return Result(res);
}
}
auto state = RocksDBTransactionState::toState(trx);
RocksDBSavePoint guard(trx, TRI_VOC_DOCUMENT_OPERATION_REMOVE);
// add possible log statement under guard
state->prepareOperation(
_logicalCollection.id(), oldRevisionId, TRI_VOC_DOCUMENT_OPERATION_REMOVE
);
res = removeDocument(trx, oldDocumentId, oldDoc, options);
if (res.ok()) {
trackWaitForSync(trx, options);
bool hasPerformedIntermediateCommit = false;
res = state->addOperation(
_logicalCollection.id(), revisionId, TRI_VOC_DOCUMENT_OPERATION_REMOVE,
hasPerformedIntermediateCommit
);
if (res.ok() && callbackDuringLock != nullptr) {
res = callbackDuringLock();
}
if (res.fail()) {
THROW_ARANGO_EXCEPTION(res);
}
guard.finish(hasPerformedIntermediateCommit);
}
return res;
}
void RocksDBCollection::deferDropCollection(
std::function<bool(LogicalCollection&)> const& /*callback*/
) {
// nothing to do here
}
/// @brief return engine-specific figures
void RocksDBCollection::figuresSpecific(
std::shared_ptr<arangodb::velocypack::Builder>& builder) {
rocksdb::TransactionDB* db = rocksutils::globalRocksDB();
RocksDBKeyBounds bounds = RocksDBKeyBounds::CollectionDocuments(_objectId);
rocksdb::Range r(bounds.start(), bounds.end());
uint64_t out = 0;
db->GetApproximateSizes(
RocksDBColumnFamily::documents(), &r, 1, &out,
static_cast<uint8_t>(
rocksdb::DB::SizeApproximationFlags::INCLUDE_MEMTABLES |
rocksdb::DB::SizeApproximationFlags::INCLUDE_FILES));
builder->add("documentsSize", VPackValue(out));
bool cacheInUse = useCache();
builder->add("cacheInUse", VPackValue(cacheInUse));
if (cacheInUse) {
builder->add("cacheSize", VPackValue(_cache->size()));
builder->add("cacheUsage", VPackValue(_cache->usage()));
auto hitRates = _cache->hitRates();
double rate = hitRates.first;
rate = std::isnan(rate) ? 0.0 : rate;
builder->add("cacheLifeTimeHitRate", VPackValue(rate));
rate = hitRates.second;
rate = std::isnan(rate) ? 0.0 : rate;
builder->add("cacheWindowedHitRate", VPackValue(rate));
} else {
builder->add("cacheSize", VPackValue(0));
builder->add("cacheUsage", VPackValue(0));
}
}
void RocksDBCollection::addIndex(std::shared_ptr<arangodb::Index> idx) {
// LOCKED from the outside
// primary index must be added at position 0
TRI_ASSERT(ServerState::instance()->isRunningInCluster() ||
idx->type() != arangodb::Index::TRI_IDX_TYPE_PRIMARY_INDEX ||
_indexes.empty());
auto const id = idx->id();
for (auto const& it : _indexes) {
if (it->id() == id) {
// already have this particular index. do not add it again
return;
}
}
TRI_UpdateTickServer(static_cast<TRI_voc_tick_t>(id));
_indexes.emplace_back(idx);
if (idx->type() == Index::TRI_IDX_TYPE_PRIMARY_INDEX) {
TRI_ASSERT(idx->id() == 0);
_primaryIndex = static_cast<RocksDBPrimaryIndex*>(idx.get());
}
}
template<typename WriteBatchType, typename MethodsType>
static arangodb::Result fillIndex(transaction::Methods* trx,
RocksDBIndex* ridx,
std::unique_ptr<IndexIterator> it,
WriteBatchType& batch,
RocksDBCollection* rcol) {
auto state = RocksDBTransactionState::toState(trx);
// fillindex can be non transactional, we just need to clean up
rocksdb::DB* db = rocksutils::globalRocksDB()->GetRootDB();
TRI_ASSERT(db != nullptr);
uint64_t numDocsWritten = 0;
// write batch will be reset every x documents
MethodsType batched(state, &batch);
arangodb::Result res;
auto cb = [&](LocalDocumentId const& documentId, VPackSlice slice) {
if (res.ok()) {
res = ridx->insertInternal(trx, &batched, documentId, slice,
Index::OperationMode::normal);
if (res.ok()) {
numDocsWritten++;
}
}
};
rocksdb::WriteOptions wo;
bool hasMore = true;
while (hasMore && res.ok()) {
hasMore = it->nextDocument(cb, 250);
if (TRI_VOC_COL_STATUS_DELETED == it->collection()->status()
|| it->collection()->deleted()) {
res = TRI_ERROR_INTERNAL;
} else if (application_features::ApplicationServer::isStopping()) {
res = TRI_ERROR_SHUTTING_DOWN;
}
if (res.ok()) {
rocksdb::Status s = db->Write(wo, batch.GetWriteBatch());
if (!s.ok()) {
res = rocksutils::convertStatus(s, rocksutils::StatusHint::index);
break;
}
}
batch.Clear();
}
// we will need to remove index elements created before an error
// occurred, this needs to happen since we are non transactional
if (res.fail()) {
RocksDBKeyBounds bounds = ridx->getBounds();
arangodb::Result res2 = rocksutils::removeLargeRange(rocksutils::globalRocksDB(), bounds,
true, /*useRangeDel*/numDocsWritten > 25000);
if (res2.fail()) {
LOG_TOPIC(WARN, Logger::ENGINES) << "was not able to roll-back "
<< "index creation: " << res2.errorMessage();
}
}
return res;
}
/// non-transactional: fill index with existing documents
/// from this collection
arangodb::Result RocksDBCollection::fillIndexes(
transaction::Methods* trx, std::shared_ptr<arangodb::Index> added) {
TRI_ASSERT(trx->state()->collection(
_logicalCollection.id(), AccessMode::Type::EXCLUSIVE
));
std::unique_ptr<IndexIterator> it(new RocksDBAllIndexIterator(
&_logicalCollection, trx, primaryIndex()
));
RocksDBIndex* ridx = static_cast<RocksDBIndex*>(added.get());
if (ridx->unique()) {
// unique index. we need to keep track of all our changes because we need to avoid
// duplicate index keys. must therefore use a WriteBatchWithIndex
rocksdb::WriteBatchWithIndex batch(ridx->columnFamily()->GetComparator(), 32 * 1024 * 1024);
return fillIndex<rocksdb::WriteBatchWithIndex, RocksDBBatchedWithIndexMethods>(
trx, ridx, std::move(it), batch, this);
} else {
// non-unique index. all index keys will be unique anyway because they contain the document id
// we can therefore get away with a cheap WriteBatch
rocksdb::WriteBatch batch(32 * 1024 * 1024);
return fillIndex<rocksdb::WriteBatch, RocksDBBatchedMethods>(
trx, ridx, std::move(it), batch, this);
}
return Result();
}
Result RocksDBCollection::insertDocument(
arangodb::transaction::Methods* trx, LocalDocumentId const& documentId,
VPackSlice const& doc, OperationOptions& options) const {
// Coordinator doesn't know index internals
TRI_ASSERT(!ServerState::instance()->isCoordinator());
TRI_ASSERT(trx->state()->isRunning());
RocksDBKeyLeaser key(trx);
key->constructDocument(_objectId, documentId);
blackListKey(key->string().data(), static_cast<uint32_t>(key->string().size()));
RocksDBMethods* mthds = RocksDBTransactionState::toMethods(trx);
// disable indexing in this transaction if we are allowed to
IndexingDisabler disabler(mthds, trx->isSingleOperationTransaction());
Result res = mthds->Put(RocksDBColumnFamily::documents(), key.ref(),
rocksdb::Slice(reinterpret_cast<char const*>(doc.begin()),
static_cast<size_t>(doc.byteSize())));
if (res.fail()) {
return res;
}
READ_LOCKER(guard, _indexesLock);
for (std::shared_ptr<Index> const& idx : _indexes) {
RocksDBIndex* rIdx = static_cast<RocksDBIndex*>(idx.get());
Result tmpres = rIdx->insertInternal(trx, mthds, documentId, doc,
options.indexOperationMode);
if (tmpres.fail()) {
if (tmpres.is(TRI_ERROR_OUT_OF_MEMORY)) {
// in case of OOM return immediately
return tmpres;
}
if (tmpres.is(TRI_ERROR_ARANGO_UNIQUE_CONSTRAINT_VIOLATED) || res.ok()) {
// "prefer" unique constraint violated over other errors
res.reset(tmpres);
}
}
}
return res;
}
Result RocksDBCollection::removeDocument(
arangodb::transaction::Methods* trx, LocalDocumentId const& documentId,
VPackSlice const& doc, OperationOptions& options) const {
// Coordinator doesn't know index internals
TRI_ASSERT(!ServerState::instance()->isCoordinator());
TRI_ASSERT(trx->state()->isRunning());
TRI_ASSERT(_objectId != 0);
RocksDBKeyLeaser key(trx);
key->constructDocument(_objectId, documentId);
blackListKey(key->string().data(), static_cast<uint32_t>(key->string().size()));
RocksDBMethods* mthd = RocksDBTransactionState::toMethods(trx);
// disable indexing in this transaction if we are allowed to
IndexingDisabler disabler(mthd, trx->isSingleOperationTransaction());
Result res = mthd->SingleDelete(RocksDBColumnFamily::documents(), key.ref());
if (res.fail()) {
return res;
}
/*LOG_TOPIC(ERR, Logger::ENGINES)
<< "Delete rev: " << revisionId << " trx: " << trx->state()->id()
<< " seq: " << mthd->sequenceNumber()
<< " objectID " << _objectId << " name: " << _logicalCollection->name();*/
Result resInner;
READ_LOCKER(guard, _indexesLock);
for (std::shared_ptr<Index> const& idx : _indexes) {
Result tmpres = idx->remove(trx, documentId, doc, options.indexOperationMode);
if (tmpres.fail()) {
if (tmpres.is(TRI_ERROR_OUT_OF_MEMORY)) {
// in case of OOM return immediately
return tmpres;
}
// for other errors, set result
res.reset(tmpres);
}
}
return res;
}
Result RocksDBCollection::updateDocument(
transaction::Methods* trx, LocalDocumentId const& oldDocumentId,
VPackSlice const& oldDoc, LocalDocumentId const& newDocumentId,
VPackSlice const& newDoc, OperationOptions& options) const {
// Coordinator doesn't know index internals
TRI_ASSERT(!ServerState::instance()->isCoordinator());
TRI_ASSERT(trx->state()->isRunning());
TRI_ASSERT(_objectId != 0);
RocksDBMethods* mthd = RocksDBTransactionState::toMethods(trx);
// We NEED to do the PUT first, otherwise WAL tailing breaks
RocksDBKeyLeaser newKey(trx);
newKey->constructDocument(_objectId, newDocumentId);
// TODO: given that this should have a unique revision ID, do
// we really need to blacklist the new key?
blackListKey(newKey->string().data(),
static_cast<uint32_t>(newKey->string().size()));
// disable indexing in this transaction if we are allowed to
IndexingDisabler disabler(mthd, trx->isSingleOperationTransaction());
Result res = mthd->Put(RocksDBColumnFamily::documents(), newKey.ref(),
rocksdb::Slice(reinterpret_cast<char const*>(newDoc.begin()),
static_cast<size_t>(newDoc.byteSize())));
if (res.fail()) {
return res;
}
RocksDBKeyLeaser oldKey(trx);
oldKey->constructDocument(_objectId, oldDocumentId);
blackListKey(oldKey->string().data(),
static_cast<uint32_t>(oldKey->string().size()));
res = mthd->SingleDelete(RocksDBColumnFamily::documents(), oldKey.ref());
if (res.fail()) {
return res;
}
READ_LOCKER(guard, _indexesLock);
for (std::shared_ptr<Index> const& idx : _indexes) {
RocksDBIndex* rIdx = static_cast<RocksDBIndex*>(idx.get());
Result tmpres = rIdx->updateInternal(trx, mthd, oldDocumentId, oldDoc, newDocumentId,
newDoc, options.indexOperationMode);
if (tmpres.fail()) {
if (tmpres.is(TRI_ERROR_OUT_OF_MEMORY)) {
// in case of OOM return immediately
return tmpres;
}
res.reset(tmpres);
}
}
return res;
}
arangodb::Result RocksDBCollection::lookupDocumentVPack(
LocalDocumentId const& documentId, transaction::Methods* trx,
arangodb::ManagedDocumentResult& mdr, bool withCache) const {
TRI_ASSERT(trx->state()->isRunning());
TRI_ASSERT(_objectId != 0);
RocksDBKeyLeaser key(trx);
key->constructDocument(_objectId, documentId);
bool lockTimeout = false;
if (withCache && useCache()) {
TRI_ASSERT(_cache != nullptr);
// check cache first for fast path
auto f = _cache->find(key->string().data(),
static_cast<uint32_t>(key->string().size()));
if (f.found()) {
std::string* value = mdr.setManaged(documentId);
value->append(reinterpret_cast<char const*>(f.value()->value()),
f.value()->valueSize());
return TRI_ERROR_NO_ERROR;
}
if (f.result().errorNumber() == TRI_ERROR_LOCK_TIMEOUT) {
// assuming someone is currently holding a write lock, which
// is why we cannot access the TransactionalBucket.
lockTimeout = true; // we skip the insert in this case
}
}
RocksDBMethods* mthd = RocksDBTransactionState::toMethods(trx);
std::string* value = mdr.setManaged(documentId);
Result res = mthd->Get(RocksDBColumnFamily::documents(), key.ref(), value);
if (res.ok()) {
if (withCache && useCache() && !lockTimeout) {
TRI_ASSERT(_cache != nullptr);
// write entry back to cache
auto entry = cache::CachedValue::construct(
key->string().data(), static_cast<uint32_t>(key->string().size()),
value->data(), static_cast<uint64_t>(value->size()));
if (entry) {
Result status = _cache->insert(entry);
if (status.errorNumber() == TRI_ERROR_LOCK_TIMEOUT) {
// the writeLock uses cpu_relax internally, so we can try yield
std::this_thread::yield();
status = _cache->insert(entry);
}
if (status.fail()) {
delete entry;
}
}
}
} else {
LOG_TOPIC(DEBUG, Logger::ENGINES)
<< "NOT FOUND rev: " << documentId.id() << " trx: " << trx->state()->id()
<< " seq: " << mthd->sequenceNumber()
<< " objectID " << _objectId << " name: " << _logicalCollection.name();
mdr.clear();
}
return res;
}
arangodb::Result RocksDBCollection::lookupDocumentVPack(
LocalDocumentId const& documentId, transaction::Methods* trx,
IndexIterator::DocumentCallback const& cb, bool withCache) const {
TRI_ASSERT(trx->state()->isRunning());
TRI_ASSERT(_objectId != 0);
RocksDBKeyLeaser key(trx);
key->constructDocument(_objectId, documentId);
bool lockTimeout = false;
if (withCache && useCache()) {
TRI_ASSERT(_cache != nullptr);
// check cache first for fast path
auto f = _cache->find(key->string().data(),
static_cast<uint32_t>(key->string().size()));
if (f.found()) {
cb(documentId,
VPackSlice(reinterpret_cast<char const*>(f.value()->value())));
return TRI_ERROR_NO_ERROR;
}
if (f.result().errorNumber() == TRI_ERROR_LOCK_TIMEOUT) {
// assuming someone is currently holding a write lock, which
// is why we cannot access the TransactionalBucket.
lockTimeout = true; // we skip the insert in this case
}
}
rocksdb::PinnableSlice ps;
auto state = RocksDBTransactionState::toState(trx);
RocksDBMethods* mthd = state->rocksdbMethods();
Result res = mthd->Get(RocksDBColumnFamily::documents(), key.ref(), &ps);
if (res.ok()) {
if (withCache && useCache() && !lockTimeout) {
TRI_ASSERT(_cache != nullptr);
// write entry back to cache
auto entry = cache::CachedValue::construct(
key->string().data(), static_cast<uint32_t>(key->string().size()),
ps.data(), static_cast<uint64_t>(ps.size()));
if (entry) {
auto status = _cache->insert(entry);
if (status.errorNumber() == TRI_ERROR_LOCK_TIMEOUT) {
// the writeLock uses cpu_relax internally, so we can try yield
std::this_thread::yield();
status = _cache->insert(entry);
}
if (status.fail()) {
delete entry;
}
}
}
cb(documentId, VPackSlice(ps.data()));
} else {
LOG_TOPIC(DEBUG, Logger::ENGINES)
<< "NOT FOUND rev: " << documentId.id() << " trx: " << trx->state()->id()
<< " seq: " << mthd->sequenceNumber()
<< " objectID " << _objectId << " name: " << _logicalCollection.name();
}
return res;
}
/// may never be called unless recovery is finished
void RocksDBCollection::adjustNumberDocuments(TRI_voc_rid_t revId,
int64_t adjustment) {
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
RocksDBEngine* engine =
static_cast<RocksDBEngine*>(EngineSelectorFeature::ENGINE);
TRI_ASSERT(engine != nullptr);
TRI_ASSERT(!engine->inRecovery());
#endif
if (revId != 0) {
_revisionId = revId;
}
if (adjustment < 0) {
TRI_ASSERT(_numberDocuments >= static_cast<uint64_t>(-adjustment));
_numberDocuments -= static_cast<uint64_t>(-adjustment);
} else if (adjustment > 0) {
_numberDocuments += static_cast<uint64_t>(adjustment);
}
}
/// load the number of docs from storage, use careful
void RocksDBCollection::loadInitialNumberDocuments() {
RocksDBCollectionMeta::DocCount count = _meta.currentCount();
TRI_ASSERT(count._added >= count._removed);
_numberDocuments = count._added - count._removed;
_revisionId = count._revisionId;
}
/// @brief write locks a collection, with a timeout
int RocksDBCollection::lockWrite(double timeout) {
uint64_t waitTime = 0; // indicates that time is uninitialized
double startTime = 0.0;
while (true) {
TRY_WRITE_LOCKER(locker, _exclusiveLock);
if (locker.isLocked()) {
// keep lock and exit loop
locker.steal();
return TRI_ERROR_NO_ERROR;
}
double now = TRI_microtime();
if (waitTime == 0) { // initialize times
// set end time for lock waiting
if (timeout <= 0.0) {
timeout = defaultLockTimeout;
}
startTime = now;
waitTime = 1;
}
if (now > startTime + timeout) {
LOG_TOPIC(TRACE, arangodb::Logger::ENGINES)
<< "timed out after " << timeout
<< " s waiting for write-lock on collection '"
<< _logicalCollection.name() << "'";
return TRI_ERROR_LOCK_TIMEOUT;
}
if (now - startTime < 0.001) {
std::this_thread::yield();
} else {
std::this_thread::sleep_for(std::chrono::microseconds(waitTime));
if (waitTime < 32) {
waitTime *= 2;
}
}
}
}
/// @brief write unlocks a collection
int RocksDBCollection::unlockWrite() {
_exclusiveLock.unlockWrite();
return TRI_ERROR_NO_ERROR;
}
/// @brief read locks a collection, with a timeout
int RocksDBCollection::lockRead(double timeout) {
uint64_t waitTime = 0; // indicates that time is uninitialized
double startTime = 0.0;
while (true) {
TRY_READ_LOCKER(locker, _exclusiveLock);
if (locker.isLocked()) {
// keep lock and exit loop
locker.steal();
return TRI_ERROR_NO_ERROR;
}
double now = TRI_microtime();
if (waitTime == 0) { // initialize times
// set end time for lock waiting
if (timeout <= 0.0) {
timeout = defaultLockTimeout;
}
startTime = now;
waitTime = 1;
}
if (now > startTime + timeout) {
LOG_TOPIC(TRACE, arangodb::Logger::ENGINES)
<< "timed out after " << timeout
<< " s waiting for read-lock on collection '"
<< _logicalCollection.name() << "'";
return TRI_ERROR_LOCK_TIMEOUT;
}
if (now - startTime < 0.001) {
std::this_thread::yield();
} else {
std::this_thread::sleep_for(std::chrono::microseconds(waitTime));
if (waitTime < 32) {
waitTime *= 2;
}
}
}
}
/// @brief read unlocks a collection
int RocksDBCollection::unlockRead() {
_exclusiveLock.unlockRead();
return TRI_ERROR_NO_ERROR;
}
// rescans the collection to update document count
uint64_t RocksDBCollection::recalculateCounts() {
RocksDBEngine* engine = rocksutils::globalRocksEngine();
rocksdb::TransactionDB* db = engine->db();
const rocksdb::Snapshot* snapshot = nullptr;
// start transaction to get a collection lock
TRI_vocbase_t& vocbase = _logicalCollection.vocbase();
if (!vocbase.use()) { // someone dropped the database
return numberDocuments();
}
auto useGuard = scopeGuard([&] {
if (snapshot) {
db->ReleaseSnapshot(snapshot);
}
vocbase.release();
});
TRI_vocbase_col_status_e status;
int res = vocbase.useCollection(&_logicalCollection, status);
if (res != TRI_ERROR_NO_ERROR) {
THROW_ARANGO_EXCEPTION(res);
}
auto collGuard = scopeGuard([&] {
vocbase.releaseCollection(&_logicalCollection);
});
uint64_t snapNumberOfDocuments = 0;
{
// fetch number docs and snapshot under exclusive lock
// this should enable us to correct the count later
auto lockGuard = scopeGuard([this] { unlockWrite(); });
res = lockWrite(transaction::Options::defaultLockTimeout);
if (res != TRI_ERROR_NO_ERROR) {
lockGuard.cancel();
THROW_ARANGO_EXCEPTION(res);
}
snapNumberOfDocuments = numberDocuments();
snapshot = engine->db()->GetSnapshot();
TRI_ASSERT(snapshot);
}
// count documents
auto bounds = RocksDBKeyBounds::CollectionDocuments(_objectId);
rocksdb::Slice upper(bounds.end());
rocksdb::ReadOptions ro;
ro.prefix_same_as_start = true;
ro.iterate_upper_bound = &upper;
ro.verify_checksums = false;
ro.fill_cache = false;
rocksdb::ColumnFamilyHandle* cf = bounds.columnFamily();
std::unique_ptr<rocksdb::Iterator> it(db->NewIterator(ro, cf));
std::size_t count = 0;
it->Seek(bounds.start());
while (it->Valid() && it->key().compare(upper) < 0) {
++count;
it->Next();
}
int64_t adjustment = snapNumberOfDocuments - count;
if (adjustment != 0) {
LOG_TOPIC(WARN, Logger::REPLICATION) << "inconsistent collection count detected, "
<< "an offet of " << adjustment << " will be applied";
adjustNumberDocuments(static_cast<TRI_voc_rid_t>(0), adjustment);
}
return numberDocuments();
}
void RocksDBCollection::compact() {
rocksdb::TransactionDB* db = rocksutils::globalRocksDB();
rocksdb::CompactRangeOptions opts;
RocksDBKeyBounds bounds = RocksDBKeyBounds::CollectionDocuments(_objectId);
rocksdb::Slice b = bounds.start(), e = bounds.end();
db->CompactRange(opts, bounds.columnFamily(), &b, &e);
READ_LOCKER(guard, _indexesLock);
for (std::shared_ptr<Index> i : _indexes) {
RocksDBIndex* index = static_cast<RocksDBIndex*>(i.get());
index->cleanup();
}
}
void RocksDBCollection::estimateSize(velocypack::Builder& builder) {
TRI_ASSERT(!builder.isOpenObject() && !builder.isOpenArray());
rocksdb::TransactionDB* db = rocksutils::globalRocksDB();
RocksDBKeyBounds bounds = RocksDBKeyBounds::CollectionDocuments(_objectId);
rocksdb::Range r(bounds.start(), bounds.end());
uint64_t out = 0, total = 0;
db->GetApproximateSizes(
RocksDBColumnFamily::documents(), &r, 1, &out,
static_cast<uint8_t>(
rocksdb::DB::SizeApproximationFlags::INCLUDE_MEMTABLES |
rocksdb::DB::SizeApproximationFlags::INCLUDE_FILES));
total += out;
builder.openObject();
builder.add("documents", VPackValue(out));
builder.add("indexes", VPackValue(VPackValueType::Object));
READ_LOCKER(guard, _indexesLock);
for (std::shared_ptr<Index> i : _indexes) {
RocksDBIndex* index = static_cast<RocksDBIndex*>(i.get());
out = index->memory();
builder.add(std::to_string(index->id()), VPackValue(out));
total += out;
}
builder.close();
builder.add("total", VPackValue(total));
builder.close();
}
void RocksDBCollection::createCache() const {
if (!_cacheEnabled || _cachePresent || _logicalCollection.isAStub() ||
ServerState::instance()->isCoordinator()) {
// we leave this if we do not need the cache
// or if cache already created
return;
}
TRI_ASSERT(_cacheEnabled);
TRI_ASSERT(_cache.get() == nullptr);
TRI_ASSERT(CacheManagerFeature::MANAGER != nullptr);
LOG_TOPIC(DEBUG, Logger::CACHE) << "Creating document cache";
_cache = CacheManagerFeature::MANAGER->createCache(
cache::CacheType::Transactional);
_cachePresent = (_cache.get() != nullptr);
TRI_ASSERT(_cacheEnabled);
}
void RocksDBCollection::destroyCache() const {
if (!_cachePresent) {
return;
}
TRI_ASSERT(CacheManagerFeature::MANAGER != nullptr);
// must have a cache...
TRI_ASSERT(_cache.get() != nullptr);
LOG_TOPIC(DEBUG, Logger::CACHE) << "Destroying document cache";
CacheManagerFeature::MANAGER->destroyCache(_cache);
_cache.reset();
_cachePresent = false;
}
// blacklist given key from transactional cache
void RocksDBCollection::blackListKey(char const* data, std::size_t len) const {
if (useCache()) {
TRI_ASSERT(_cache != nullptr);
bool blacklisted = false;
while (!blacklisted) {
auto status = _cache->blacklist(data, static_cast<uint32_t>(len));
if (status.ok()) {
blacklisted = true;
} else if (status.errorNumber() == TRI_ERROR_SHUTTING_DOWN) {
destroyCache();
break;
}
}
}
}
void RocksDBCollection::trackWaitForSync(arangodb::transaction::Methods* trx,
OperationOptions& options) {
if (_logicalCollection.waitForSync() && !options.isRestore) {
options.waitForSync = true;
}
if (options.waitForSync) {
trx->state()->waitForSync(true);
}
}
Reference in New Issue View Git Blame Copy Permalink