////////////////////////////////////////////////////////////////////////////////
/// DISCLAIMER
///
/// Copyright 2014-2018 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
/// @author Jan Christoph Uhde
////////////////////////////////////////////////////////////////////////////////

#include "RocksDBEngine.h"
#include "ApplicationFeatures/ApplicationServer.h"
#include "ApplicationFeatures/RocksDBOptionFeature.h"
#include "Basics/Exceptions.h"
#include "Basics/FileUtils.h"
#include "Basics/ReadLocker.h"
#include "Basics/Result.h"
#include "Basics/RocksDBLogger.h"
#include "Basics/StaticStrings.h"
#include "Basics/Thread.h"
#include "Basics/VelocyPackHelper.h"
#include "Basics/WriteLocker.h"
#include "Basics/build.h"
#include "Cache/CacheManagerFeature.h"
#include "Cache/Manager.h"
#include "Cluster/ServerState.h"
#include "GeneralServer/RestHandlerFactory.h"
#include "IResearch/IResearchCommon.h"
#include "Logger/Logger.h"
#include "ProgramOptions/ProgramOptions.h"
#include "ProgramOptions/Section.h"
#include "Rest/Version.h"
#include "RestHandler/RestHandlerCreator.h"
#include "RestServer/DatabasePathFeature.h"
#include "RestServer/FlushFeature.h"
#include "RestServer/ServerIdFeature.h"
#include "RocksDBEngine/RocksDBBackgroundErrorListener.h"
#include "RocksDBEngine/RocksDBBackgroundThread.h"
#include "RocksDBEngine/RocksDBCollection.h"
#include "RocksDBEngine/RocksDBColumnFamily.h"
#include "RocksDBEngine/RocksDBCommon.h"
#include "RocksDBEngine/RocksDBComparator.h"
#include "RocksDBEngine/RocksDBEventListener.h"
#include "RocksDBEngine/RocksDBIncrementalSync.h"
#include "RocksDBEngine/RocksDBIndex.h"
#include "RocksDBEngine/RocksDBIndexFactory.h"
#include "RocksDBEngine/RocksDBKey.h"
#include "RocksDBEngine/RocksDBLogValue.h"
#include "RocksDBEngine/RocksDBOptimizerRules.h"
#include "RocksDBEngine/RocksDBPrefixExtractor.h"
#include "RocksDBEngine/RocksDBRecoveryManager.h"
#include "RocksDBEngine/RocksDBReplicationManager.h"
#include "RocksDBEngine/RocksDBReplicationTailing.h"
#include "RocksDBEngine/RocksDBRestHandlers.h"
#include "RocksDBEngine/RocksDBSettingsManager.h"
#include "RocksDBEngine/RocksDBSyncThread.h"
#include "RocksDBEngine/RocksDBThrottle.h"
#include "RocksDBEngine/RocksDBTransactionCollection.h"
#include "RocksDBEngine/RocksDBTransactionState.h"
#include "RocksDBEngine/RocksDBTypes.h"
#include "RocksDBEngine/RocksDBUpgrade.h"
#include "RocksDBEngine/RocksDBV8Functions.h"
#include "RocksDBEngine/RocksDBValue.h"
#include "RocksDBEngine/RocksDBWalAccess.h"
#include "Transaction/Context.h"
#include "Transaction/ContextData.h"
#include "Transaction/Manager.h"
#include "Transaction/Options.h"
#include "Transaction/StandaloneContext.h"
#include "VocBase/LogicalView.h"
#include "VocBase/ticks.h"

#include <rocksdb/convenience.h>
#include <rocksdb/db.h>
#include <rocksdb/env.h>
#include <rocksdb/filter_policy.h>
#include <rocksdb/iterator.h>
#include <rocksdb/options.h>
#include <rocksdb/slice_transform.h>
#include <rocksdb/statistics.h>
#include <rocksdb/table.h>
#include <rocksdb/transaction_log.h>
#include <rocksdb/utilities/transaction_db.h>
#include <rocksdb/write_batch.h>

#include <velocypack/Iterator.h>
#include <velocypack/velocypack-aliases.h>

#include <limits>

using namespace arangodb;
using namespace arangodb::application_features;
using namespace arangodb::options;

namespace arangodb {

std::string const RocksDBEngine::EngineName("rocksdb");
std::string const RocksDBEngine::FeatureName("RocksDBEngine");

// static variables for all existing column families
rocksdb::ColumnFamilyHandle* RocksDBColumnFamily::_definitions(nullptr);
rocksdb::ColumnFamilyHandle* RocksDBColumnFamily::_documents(nullptr);
rocksdb::ColumnFamilyHandle* RocksDBColumnFamily::_primary(nullptr);
rocksdb::ColumnFamilyHandle* RocksDBColumnFamily::_edge(nullptr);
rocksdb::ColumnFamilyHandle* RocksDBColumnFamily::_vpack(nullptr);
rocksdb::ColumnFamilyHandle* RocksDBColumnFamily::_geo(nullptr);
rocksdb::ColumnFamilyHandle* RocksDBColumnFamily::_fulltext(nullptr);
std::vector<rocksdb::ColumnFamilyHandle*> RocksDBColumnFamily::_allHandles;

// minimum value for --rocksdb.sync-interval (in ms)
// a value of 0 however means turning off the syncing altogether!
static constexpr uint64_t minSyncInterval = 5;

static constexpr uint64_t databaseIdForGlobalApplier = 0;

// handles for recovery helpers
std::vector<std::shared_ptr<RocksDBRecoveryHelper>> RocksDBEngine::_recoveryHelpers;

RocksDBFilePurgePreventer::RocksDBFilePurgePreventer(RocksDBEngine* engine)
    : _engine(engine) {
  TRI_ASSERT(_engine != nullptr);
  _engine->_purgeLock.readLock();
}

RocksDBFilePurgePreventer::~RocksDBFilePurgePreventer() {
  if (_engine != nullptr) {
    _engine->_purgeLock.unlockRead();
  }
}

RocksDBFilePurgePreventer::RocksDBFilePurgePreventer(RocksDBFilePurgePreventer&& other)
    : _engine(other._engine) {
  // steal engine from other
  other._engine = nullptr;
}

RocksDBFilePurgeEnabler::RocksDBFilePurgeEnabler(RocksDBEngine* engine)
    : _engine(nullptr) {
  TRI_ASSERT(engine != nullptr);

  if (engine->_purgeLock.tryWriteLock()) {
    // we got the lock
    _engine = engine;
  }
}

RocksDBFilePurgeEnabler::~RocksDBFilePurgeEnabler() {
  if (_engine != nullptr) {
    _engine->_purgeLock.unlockWrite();
  }
}

RocksDBFilePurgeEnabler::RocksDBFilePurgeEnabler(RocksDBFilePurgeEnabler&& other)
    : _engine(other._engine) {
  // steal engine from other
  other._engine = nullptr;
}

// create the storage engine
RocksDBEngine::RocksDBEngine(application_features::ApplicationServer& server)
    : StorageEngine(server, EngineName, FeatureName,
                    std::unique_ptr<IndexFactory>(new RocksDBIndexFactory())),
#ifdef USE_ENTERPRISE
      _createShaFiles(true),
#else
      _createShaFiles(false),
#endif
      _db(nullptr),
      _vpackCmp(new RocksDBVPackComparator()),
      _walAccess(new RocksDBWalAccess()),
      _maxTransactionSize(transaction::Options::defaultMaxTransactionSize),
      _intermediateCommitSize(transaction::Options::defaultIntermediateCommitSize),
      _intermediateCommitCount(transaction::Options::defaultIntermediateCommitCount),
      _pruneWaitTime(10.0),
      _pruneWaitTimeInitial(180.0),
      _maxWalArchiveSizeLimit(0),
      _releasedTick(0),
#ifdef _WIN32
      // background syncing is not supported on Windows
      _syncInterval(0),
#else
      _syncInterval(100),
#endif
      _useThrottle(true),
      _useReleasedTick(false),
      _debugLogging(false) {

  startsAfter("BasicsPhase");
  // inherits order from StorageEngine but requires "RocksDBOption" that is used
  // to configure this engine and the MMFiles PersistentIndexFeature
  startsAfter("RocksDBOption");

  server.addFeature(new RocksDBRecoveryManager(server));
}

RocksDBEngine::~RocksDBEngine() { shutdownRocksDBInstance(); }

/// shuts down the RocksDB instance. this is called from unprepare
/// and the dtor
void RocksDBEngine::shutdownRocksDBInstance() noexcept {
  if (_db == nullptr) {
    return;
  }

  // turn off RocksDBThrottle, and release our pointers to it
  if (nullptr != _listener.get()) {
    _listener->StopThread();
  }  // if

  for (rocksdb::ColumnFamilyHandle* h : RocksDBColumnFamily::_allHandles) {
    _db->DestroyColumnFamilyHandle(h);
  }

  // now prune all obsolete WAL files
  try {
    determinePrunableWalFiles(0);
    pruneWalFiles();
  } catch (...) {
    // this is allowed to go wrong on shutdown
    // we must not throw an exception from here
  }

  try {
    // do a final WAL sync here before shutting down
    Result res = RocksDBSyncThread::sync(_db->GetBaseDB());
    if (res.fail()) {
      LOG_TOPIC("14ede", WARN, Logger::ENGINES)
          << "could not sync RocksDB WAL: " << res.errorMessage();
    }

    rocksdb::Status status = _db->Close();

    if (!status.ok()) {
      Result res = rocksutils::convertStatus(status);
      LOG_TOPIC("2b9c1", ERR, Logger::ENGINES)
          << "could not shutdown RocksDB: " << res.errorMessage();
    }
  } catch (...) {
    // this is allowed to go wrong on shutdown
    // we must not throw an exception from here
  }

  delete _db;
  _db = nullptr;
}

// inherited from ApplicationFeature
// ---------------------------------

// add the storage engine's specific options to the global list of options
void RocksDBEngine::collectOptions(std::shared_ptr<options::ProgramOptions> options) {
  options->addSection("rocksdb", "RocksDB engine specific configuration");

  // control transaction size for RocksDB engine
  options->addOption("--rocksdb.max-transaction-size",
                     "transaction size limit (in bytes)",
                     new UInt64Parameter(&_maxTransactionSize));

  options->addOption("--rocksdb.intermediate-commit-size",
                     "an intermediate commit will be performed automatically "
                     "when a transaction "
                     "has accumulated operations of this size (in bytes)",
                     new UInt64Parameter(&_intermediateCommitSize));

  options->addOption("--rocksdb.intermediate-commit-count",
                     "an intermediate commit will be performed automatically "
                     "when this number of "
                     "operations is reached in a transaction",
                     new UInt64Parameter(&_intermediateCommitCount));

  options->addOption("--rocksdb.sync-interval",
                     "interval for automatic, non-requested disk syncs (in "
                     "milliseconds, use 0 to turn automatic syncing off)",
                     new UInt64Parameter(&_syncInterval));

  options->addOption("--rocksdb.wal-file-timeout",
                     "timeout after which unused WAL files are deleted",
                     new DoubleParameter(&_pruneWaitTime));

  options->addOption(
      "--rocksdb.wal-file-timeout-initial",
      "initial timeout after which unused WAL files deletion kicks in after "
      "server start",
      new DoubleParameter(&_pruneWaitTimeInitial),
      arangodb::options::makeFlags(arangodb::options::Flags::Hidden));

  options->addOption("--rocksdb.throttle", "enable write-throttling",
                     new BooleanParameter(&_useThrottle));

  options->addOption("--rocksdb.create-sha-files", "enable generation of sha256 files for each .sst file",
                     new BooleanParameter(&_createShaFiles),
                     arangodb::options::makeFlags(arangodb::options::Flags::Enterprise));

  options->addOption("--rocksdb.debug-logging",
                     "true to enable rocksdb debug logging",
                     new BooleanParameter(&_debugLogging),
                     arangodb::options::makeFlags(arangodb::options::Flags::Hidden));

  options->addOption("--rocksdb.wal-archive-size-limit",
                     "maximum total size (in bytes) of archived WAL files (0 = unlimited)",
                     new UInt64Parameter(&_maxWalArchiveSizeLimit),
                     arangodb::options::makeFlags(arangodb::options::Flags::Hidden));

#ifdef USE_ENTERPRISE
  collectEnterpriseOptions(options);
#endif
}

// validate the storage engine's specific options
void RocksDBEngine::validateOptions(std::shared_ptr<options::ProgramOptions> options) {
  transaction::Options::setLimits(_maxTransactionSize, _intermediateCommitSize,
                                  _intermediateCommitCount);
#ifdef USE_ENTERPRISE
  validateEnterpriseOptions(options);
#endif

  if (_syncInterval > 0 && _syncInterval < minSyncInterval) {
    // _syncInterval = 0 means turned off!
    LOG_TOPIC("bbd68", FATAL, arangodb::Logger::CONFIG)
        << "invalid value for --rocksdb.sync-interval. Please use a value "
        << "of at least " << minSyncInterval;
    FATAL_ERROR_EXIT();
  }

#ifdef _WIN32
  if (_syncInterval > 0) {
    LOG_TOPIC("68301", WARN, arangodb::Logger::CONFIG)
        << "automatic syncing of RocksDB WAL via background thread is not "
        << " supported on this platform";
  }
#endif

  if (_pruneWaitTimeInitial < 10) {
    LOG_TOPIC("a9667", WARN, arangodb::Logger::ENGINES)
        << "consider increasing the value for "
           "--rocksdb.wal-file-timeout-initial. "
        << "Replication clients might have trouble to get in sync";
  }
}

// preparation phase for storage engine. can be used for internal setup.
// the storage engine must not start any threads here or write any files
void RocksDBEngine::prepare() {
  // get base path from DatabaseServerFeature
  auto databasePathFeature =
      application_features::ApplicationServer::getFeature<DatabasePathFeature>(
          "DatabasePath");
  _basePath = databasePathFeature->directory();

  TRI_ASSERT(!_basePath.empty());

#ifdef USE_ENTERPRISE
  prepareEnterprise();
#endif
}

void RocksDBEngine::start() {
  // it is already decided that rocksdb is used
  TRI_ASSERT(isEnabled());

  if (ServerState::instance()->isAgent() &&
      !application_features::ApplicationServer::server->options()->processingResult().touched("rocksdb.wal-file-timeout-initial")) {
    // reduce --rocksb.wal-file-timeout-initial to 15 seconds for agency nodes
    // as we probably won't need the WAL for WAL tailing and replication here
    _pruneWaitTimeInitial = 15;
  }

  LOG_TOPIC("107fd", TRACE, arangodb::Logger::ENGINES)
      << "rocksdb version " << rest::Version::getRocksDBVersion()
      << ", supported compression types: " << getCompressionSupport();

  // set the database sub-directory for RocksDB
  auto* databasePathFeature =
      ApplicationServer::getFeature<DatabasePathFeature>("DatabasePath");
  _path = databasePathFeature->subdirectoryName("engine-rocksdb");

  if (!basics::FileUtils::isDirectory(_path)) {
    std::string systemErrorStr;
    long errorNo;

    int res = TRI_CreateRecursiveDirectory(_path.c_str(), errorNo, systemErrorStr);

    if (res == TRI_ERROR_NO_ERROR) {
      LOG_TOPIC("b2958", TRACE, arangodb::Logger::ENGINES)
          << "created RocksDB data directory '" << _path << "'";
    } else {
      LOG_TOPIC("a5ae3", FATAL, arangodb::Logger::ENGINES)
          << "unable to create RocksDB data directory '" << _path
          << "': " << systemErrorStr;
      FATAL_ERROR_EXIT();
    }
  }

  // options imported set by RocksDBOptionFeature
  auto const* opts = ApplicationServer::getFeature<arangodb::RocksDBOptionFeature>(
      "RocksDBOption");

  rocksdb::TransactionDBOptions transactionOptions;
  // number of locks per column_family
  transactionOptions.num_stripes = TRI_numberProcessors();
  transactionOptions.transaction_lock_timeout = opts->_transactionLockTimeout;

  _options.allow_fallocate = opts->_allowFAllocate;
  _options.enable_pipelined_write = opts->_enablePipelinedWrite;
  _options.write_buffer_size = static_cast<size_t>(opts->_writeBufferSize);
  _options.max_write_buffer_number = static_cast<int>(opts->_maxWriteBufferNumber);
  _options.delayed_write_rate = opts->_delayedWriteRate;
  _options.min_write_buffer_number_to_merge =
      static_cast<int>(opts->_minWriteBufferNumberToMerge);
  _options.num_levels = static_cast<int>(opts->_numLevels);
  _options.level_compaction_dynamic_level_bytes = opts->_dynamicLevelBytes;
  _options.max_bytes_for_level_base = opts->_maxBytesForLevelBase;
  _options.max_bytes_for_level_multiplier =
      static_cast<int>(opts->_maxBytesForLevelMultiplier);
  _options.optimize_filters_for_hits = opts->_optimizeFiltersForHits;
  _options.use_direct_reads = opts->_useDirectReads;
  _options.use_direct_io_for_flush_and_compaction = opts->_useDirectIoForFlushAndCompaction;
  // limit the total size of WAL files. This forces the flush of memtables of
  // column families still backed by WAL files. If we would not do this, WAL
  // files may linger around forever and will not get removed
  _options.max_total_wal_size = opts->_maxTotalWalSize;

  if (opts->_walDirectory.empty()) {
    _options.wal_dir = basics::FileUtils::buildFilename(_path, "journals");
  } else {
    _options.wal_dir = opts->_walDirectory;
  }

  LOG_TOPIC("bc82a", TRACE, arangodb::Logger::ENGINES)
      << "initializing RocksDB, path: '" << _path << "', WAL directory '"
      << _options.wal_dir << "'";

  if (opts->_skipCorrupted) {
    _options.wal_recovery_mode = rocksdb::WALRecoveryMode::kSkipAnyCorruptedRecords;
  } else {
    _options.wal_recovery_mode = rocksdb::WALRecoveryMode::kPointInTimeRecovery;
  }

  _options.max_background_jobs = static_cast<int>(opts->_maxBackgroundJobs);
  _options.max_subcompactions = static_cast<int>(opts->_maxSubcompactions);
  _options.use_fsync = opts->_useFSync;

  // only compress levels >= 2
  _options.compression_per_level.resize(_options.num_levels);
  for (int level = 0; level < _options.num_levels; ++level) {
    _options.compression_per_level[level] =
        (((uint64_t)level >= opts->_numUncompressedLevels) ? rocksdb::kSnappyCompression
                                                           : rocksdb::kNoCompression);
  }

  // Number of files to trigger level-0 compaction. A value <0 means that
  // level-0 compaction will not be triggered by number of files at all.
  // Default: 4
  _options.level0_file_num_compaction_trigger =
      static_cast<int>(opts->_level0CompactionTrigger);

  // Soft limit on number of level-0 files. We start slowing down writes at this
  // point. A value <0 means that no writing slow down will be triggered by
  // number of files in level-0.
  _options.level0_slowdown_writes_trigger = static_cast<int>(opts->_level0SlowdownTrigger);

  // Maximum number of level-0 files.  We stop writes at this point.
  _options.level0_stop_writes_trigger = static_cast<int>(opts->_level0StopTrigger);

  _options.recycle_log_file_num = opts->_recycleLogFileNum;
  _options.compaction_readahead_size = static_cast<size_t>(opts->_compactionReadaheadSize);

#ifdef USE_ENTERPRISE
  configureEnterpriseRocksDBOptions(_options);
  startEnterprise();
#endif

  _options.env->SetBackgroundThreads(static_cast<int>(opts->_numThreadsHigh),
                                     rocksdb::Env::Priority::HIGH);
  _options.env->SetBackgroundThreads(static_cast<int>(opts->_numThreadsLow),
                                     rocksdb::Env::Priority::LOW);

  // intentionally set the RocksDB logger to warning because it will
  // log lots of things otherwise
  if (_debugLogging) {
    _options.info_log_level = rocksdb::InfoLogLevel::DEBUG_LEVEL;
  } else {
    if (!opts->_useFileLogging) {
      // if we don't use file logging but log into ArangoDB's logfile,
      // we only want real errors
      _options.info_log_level = rocksdb::InfoLogLevel::ERROR_LEVEL;
    }
  }

  std::shared_ptr<RocksDBLogger> logger;

  if (!opts->_useFileLogging) {
    // if option "--rocksdb.use-file-logging" is set to false, we will use
    // our own logger that logs to ArangoDB's logfile
    logger = std::make_shared<RocksDBLogger>(_options.info_log_level);
    _options.info_log = logger;

    if (!_debugLogging) {
      logger->disable();
    }  // if
  }

  if (opts->_enableStatistics) {
    _options.statistics = rocksdb::CreateDBStatistics();
    // _options.stats_dump_period_sec = 1;
  }

  rocksdb::BlockBasedTableOptions tableOptions;
  if (opts->_blockCacheSize > 0) {
    tableOptions.block_cache =
        rocksdb::NewLRUCache(opts->_blockCacheSize,
                             static_cast<int>(opts->_blockCacheShardBits),
                             /*strict_capacity_limit*/ opts->_enforceBlockCacheSizeLimit);
    // tableOptions.cache_index_and_filter_blocks =
    // tableOptions.pin_l0_filter_and_index_blocks_in_cache
    // opts->_compactionReadaheadSize > 0;
  } else {
    tableOptions.no_block_cache = true;
  }
  tableOptions.block_size = opts->_tableBlockSize;
  tableOptions.filter_policy.reset(rocksdb::NewBloomFilterPolicy(10, true));
  // use slightly space-optimized format version 3
  tableOptions.format_version = 3;
  tableOptions.block_align = opts->_blockAlignDataBlocks;

  _options.table_factory.reset(rocksdb::NewBlockBasedTableFactory(tableOptions));

  _options.create_if_missing = true;
  _options.create_missing_column_families = true;

  if (opts->_limitOpenFilesAtStartup) {
    _options.max_open_files = 16;
    _options.skip_stats_update_on_db_open = true;
    _options.avoid_flush_during_recovery = true;
  } else {
    _options.max_open_files = -1;
  }

  // WAL_ttl_seconds needs to be bigger than the sync interval of the count
  // manager. Should be several times bigger counter_sync_seconds
  _options.WAL_ttl_seconds = 60 * 60 * 24 * 30;  // we manage WAL file deletion
  // ourselves, don't let RocksDB
  // garbage collect them
  _options.WAL_size_limit_MB = 0;
  _options.memtable_prefix_bloom_size_ratio = 0.2;  // TODO: pick better value?
  // TODO: enable memtable_insert_with_hint_prefix_extractor?
  _options.bloom_locality = 1;

  if (_useThrottle) {
    _listener.reset(new RocksDBThrottle);
    _options.listeners.push_back(_listener);
  } // if

  if (_createShaFiles) {
    _shaListener.reset(new RocksDBEventListener);
    _options.listeners.push_back(_shaListener);
  } // if

  _options.listeners.push_back(std::make_shared<RocksDBBackgroundErrorListener>());

  if (opts->_totalWriteBufferSize > 0) {
    _options.db_write_buffer_size = opts->_totalWriteBufferSize;
  }

  if (!application_features::ApplicationServer::server->options()->processingResult().touched("rocksdb.max-write-buffer-number")) {
    // user hasn't explicitly set the number of write buffers, so we use a default value based
    // on the number of column families
    // this is cfFamilies.size() + 2 ... but _option needs to be set before
    //  building cfFamilies
    // Update max_write_buffer_number above if you change number of families used
    _options.max_write_buffer_number = 7 + 2;
  } else if (_options.max_write_buffer_number < 7 + 2) {
    // user set the value explicitly, and it is lower than recommended
    _options.max_write_buffer_number = 7 + 2;
    LOG_TOPIC("d5c49", WARN, Logger::ENGINES) << "ignoring value for option `--rocksdb.max-write-buffer-number` because it is lower than recommended";
  }

  // cf options for definitons (dbs, collections, views, ...)
  rocksdb::ColumnFamilyOptions definitionsCF(_options);

  // cf options with fixed 8 byte object id prefix for documents
  rocksdb::ColumnFamilyOptions fixedPrefCF(_options);
  fixedPrefCF.prefix_extractor = std::shared_ptr<rocksdb::SliceTransform const>(
      rocksdb::NewFixedPrefixTransform(RocksDBKey::objectIdSize()));

  // construct column family options with prefix containing indexed value
  rocksdb::ColumnFamilyOptions dynamicPrefCF(_options);
  dynamicPrefCF.prefix_extractor = std::make_shared<RocksDBPrefixExtractor>();
  // also use hash-search based SST file format
  rocksdb::BlockBasedTableOptions tblo(tableOptions);
  tblo.index_type = rocksdb::BlockBasedTableOptions::IndexType::kHashSearch;
  dynamicPrefCF.table_factory =
      std::shared_ptr<rocksdb::TableFactory>(rocksdb::NewBlockBasedTableFactory(tblo));

  // velocypack based index variants with custom comparator
  rocksdb::ColumnFamilyOptions vpackFixedPrefCF(fixedPrefCF);
  rocksdb::BlockBasedTableOptions tblo2(tableOptions);
  tblo2.filter_policy.reset();  // intentionally no bloom filter here
  vpackFixedPrefCF.table_factory =
      std::shared_ptr<rocksdb::TableFactory>(rocksdb::NewBlockBasedTableFactory(tblo2));
  vpackFixedPrefCF.comparator = _vpackCmp.get();

  // create column families
  std::vector<rocksdb::ColumnFamilyDescriptor> cfFamilies;
  // no prefix families for default column family (Has to be there)
  cfFamilies.emplace_back(rocksdb::kDefaultColumnFamilyName,
                          definitionsCF);                   // 0
  cfFamilies.emplace_back("Documents", fixedPrefCF);        // 1
  cfFamilies.emplace_back("PrimaryIndex", fixedPrefCF);     // 2
  cfFamilies.emplace_back("EdgeIndex", dynamicPrefCF);      // 3
  cfFamilies.emplace_back("VPackIndex", vpackFixedPrefCF);  // 4
  cfFamilies.emplace_back("GeoIndex", fixedPrefCF);         // 5
  cfFamilies.emplace_back("FulltextIndex", fixedPrefCF);    // 6

  TRI_ASSERT(static_cast<int>(_options.max_write_buffer_number) >= static_cast<int>(cfFamilies.size()));
  // Update max_write_buffer_number above if you change number of families used

  std::vector<rocksdb::ColumnFamilyHandle*> cfHandles;
  size_t const numberOfColumnFamilies = RocksDBColumnFamily::minNumberOfColumnFamilies;
  bool dbExisted = false;
  {
    rocksdb::Options testOptions;
    testOptions.create_if_missing = false;
    testOptions.create_missing_column_families = false;
    testOptions.env = _options.env;
    std::vector<std::string> existingColumnFamilies;
    rocksdb::Status status =
        rocksdb::DB::ListColumnFamilies(testOptions, _path, &existingColumnFamilies);
    if (!status.ok()) {
      // check if we have found the database directory or not
      Result res = rocksutils::convertStatus(status);
      if (res.errorNumber() != TRI_ERROR_ARANGO_IO_ERROR) {
        // not an I/O error. so we better report the error and abort here
        LOG_TOPIC("74b7f", FATAL, arangodb::Logger::STARTUP)
            << "unable to initialize RocksDB engine: " << status.ToString();
        FATAL_ERROR_EXIT();
      }
    }

    if (status.ok()) {
      dbExisted = true;
      // we were able to open the database.
      // now check which column families are present in the db
      std::string names;
      for (auto const& it : existingColumnFamilies) {
        if (!names.empty()) {
          names.append(", ");
        }
        names.append(it);
      }

      LOG_TOPIC("528b8", DEBUG, arangodb::Logger::STARTUP)
          << "found existing column families: " << names;

      for (auto const& it : cfFamilies) {
        auto it2 = std::find(existingColumnFamilies.begin(),
                             existingColumnFamilies.end(), it.name);

        if (it2 == existingColumnFamilies.end()) {
          LOG_TOPIC("d9df8", FATAL, arangodb::Logger::STARTUP)
              << "column family '" << it.name << "' is missing in database"
              << ". if you are upgrading from an earlier alpha or beta version "
                 "of ArangoDB 3.2, "
              << "it is required to restart with a new database directory and "
                 "re-import data";
          FATAL_ERROR_EXIT();
        }
      }

      if (existingColumnFamilies.size() < numberOfColumnFamilies) {
        LOG_TOPIC("e99ec", FATAL, arangodb::Logger::STARTUP)
            << "unexpected number of column families found in database ("
            << cfHandles.size() << "). "
            << "expecting at least " << numberOfColumnFamilies
            << ". if you are upgrading from an earlier alpha or beta version "
               "of ArangoDB 3.2, "
            << "it is required to restart with a new database directory and "
               "re-import data";
        FATAL_ERROR_EXIT();
      }
    }
  }

  rocksdb::Status status =
      rocksdb::TransactionDB::Open(_options, transactionOptions, _path,
                                   cfFamilies, &cfHandles, &_db);

  if (!status.ok()) {
    std::string error;
    if (status.IsIOError()) {
      error =
          "; Maybe your filesystem doesn't provide required features? (Cifs? "
          "NFS?)";
    }

    LOG_TOPIC("fe3df", FATAL, arangodb::Logger::STARTUP)
        << "unable to initialize RocksDB engine: " << status.ToString() << error;
    FATAL_ERROR_EXIT();
  }
  if (cfFamilies.size() != cfHandles.size()) {
    LOG_TOPIC("ffc6d", FATAL, arangodb::Logger::STARTUP)
        << "unable to initialize RocksDB column families";
    FATAL_ERROR_EXIT();
  }
  if (cfHandles.size() < numberOfColumnFamilies) {
    LOG_TOPIC("e572e", FATAL, arangodb::Logger::STARTUP)
        << "unexpected number of column families found in database. "
        << "got " << cfHandles.size() << ", expecting at least " << numberOfColumnFamilies;
    FATAL_ERROR_EXIT();
  }

  // give throttle access to families
  if (_useThrottle) {
    _listener->SetFamilies(cfHandles);
  }

  // set our column families
  RocksDBColumnFamily::_definitions = cfHandles[0];
  RocksDBColumnFamily::_documents = cfHandles[1];
  RocksDBColumnFamily::_primary = cfHandles[2];
  RocksDBColumnFamily::_edge = cfHandles[3];
  RocksDBColumnFamily::_vpack = cfHandles[4];
  RocksDBColumnFamily::_geo = cfHandles[5];
  RocksDBColumnFamily::_fulltext = cfHandles[6];
  RocksDBColumnFamily::_allHandles = cfHandles;
  TRI_ASSERT(RocksDBColumnFamily::_definitions->GetID() == 0);

  // will crash the process if version does not match
  arangodb::rocksdbStartupVersionCheck(_db, dbExisted);

  // only enable logger after RocksDB start
  if (logger != nullptr) {
    logger->enable();
  }

  if (opts->_limitOpenFilesAtStartup) {
    _db->SetDBOptions({{"max_open_files", "-1"}});
  }

  {
    auto feature = application_features::ApplicationServer::getFeature<FlushFeature>("Flush");
    TRI_ASSERT(feature);
    _useReleasedTick = feature->isEnabled();
  }

  // useReleasedTick should be true on DB servers and single servers
  TRI_ASSERT((arangodb::ServerState::instance()->isCoordinator() || arangodb::ServerState::instance()->isAgent()) || _useReleasedTick);

  if (_syncInterval > 0) {
    _syncThread.reset(new RocksDBSyncThread(this, std::chrono::milliseconds(_syncInterval)));
    if (!_syncThread->start()) {
      LOG_TOPIC("63919", FATAL, Logger::ENGINES)
          << "could not start rocksdb sync thread";
      FATAL_ERROR_EXIT();
    }
  }

  TRI_ASSERT(_db != nullptr);
  _settingsManager.reset(new RocksDBSettingsManager(_db));
  _replicationManager.reset(new RocksDBReplicationManager());

  _settingsManager->retrieveInitialValues();

  double const counterSyncSeconds = 2.5;
  _backgroundThread.reset(new RocksDBBackgroundThread(this, counterSyncSeconds));
  if (!_backgroundThread->start()) {
    LOG_TOPIC("a5e96", FATAL, Logger::ENGINES)
        << "could not start rocksdb counter manager";
    FATAL_ERROR_EXIT();
  }

  if (!systemDatabaseExists()) {
    addSystemDatabase();
  }
}

void RocksDBEngine::beginShutdown() {
  TRI_ASSERT(isEnabled());

  // block the creation of new replication contexts
  if (_replicationManager != nullptr) {
    _replicationManager->beginShutdown();
  }

  //signal the event listener that we are going to shut down soon
  if (_shaListener != nullptr) {
     _shaListener->beginShutdown();
  } // if
}

void RocksDBEngine::stop() {
  TRI_ASSERT(isEnabled());

  // in case we missed the beginShutdown somehow, call it again
  replicationManager()->beginShutdown();
  replicationManager()->dropAll();

  if (_backgroundThread) {
    // stop the press
    _backgroundThread->beginShutdown();

    if (_settingsManager) {
      _settingsManager->sync(true);
    }

    // wait until background thread stops
    while (_backgroundThread->isRunning()) {
      std::this_thread::yield();
    }
    _backgroundThread.reset();
  }

  if (_syncThread) {
    // _syncThread may be a nullptr, in case automatic syncing is turned off
    _syncThread->beginShutdown();

    // wait until sync thread stops
    while (_syncThread->isRunning()) {
      std::this_thread::yield();
    }
    _syncThread.reset();
  }
}

void RocksDBEngine::unprepare() {
  TRI_ASSERT(isEnabled());
  shutdownRocksDBInstance();
}

std::unique_ptr<transaction::Manager> RocksDBEngine::createTransactionManager() {
  return std::make_unique<transaction::Manager>(/*keepData*/ false);
}

std::unique_ptr<transaction::ContextData> RocksDBEngine::createTransactionContextData() {
  return std::unique_ptr<transaction::ContextData>(nullptr); // not used by rocksdb
}

std::unique_ptr<TransactionState> RocksDBEngine::createTransactionState(
    TRI_vocbase_t& vocbase, TRI_voc_tid_t tid, transaction::Options const& options) {
  return std::make_unique<RocksDBTransactionState>(vocbase, tid, options);
}

std::unique_ptr<TransactionCollection> RocksDBEngine::createTransactionCollection(
    TransactionState& state, TRI_voc_cid_t cid, AccessMode::Type accessType, int nestingLevel) {
  return std::unique_ptr<TransactionCollection>(
      new RocksDBTransactionCollection(&state, cid, accessType, nestingLevel));
}

void RocksDBEngine::addParametersForNewCollection(VPackBuilder& builder, VPackSlice info) {
  if (!info.hasKey("objectId")) {
    builder.add("objectId", VPackValue(std::to_string(TRI_NewTickServer())));
  }
  if (!info.hasKey("cacheEnabled") || !info.get("cacheEnabled").isBool()) {
    builder.add("cacheEnabled", VPackValue(false));
  }
}

// create storage-engine specific collection
std::unique_ptr<PhysicalCollection> RocksDBEngine::createPhysicalCollection(
    LogicalCollection& collection, velocypack::Slice const& info) {
  return std::make_unique<RocksDBCollection>(collection, info);
}

// inventory functionality
// -----------------------

void RocksDBEngine::getDatabases(arangodb::velocypack::Builder& result) {
  LOG_TOPIC("a9cc7", TRACE, Logger::STARTUP) << "getting existing databases";

  rocksdb::ReadOptions readOptions;
  std::unique_ptr<rocksdb::Iterator> iter(
      _db->NewIterator(readOptions, RocksDBColumnFamily::definitions()));
  result.openArray();
  auto rSlice = rocksDBSlice(RocksDBEntryType::Database);
  for (iter->Seek(rSlice); iter->Valid() && iter->key().starts_with(rSlice); iter->Next()) {
    auto slice = VPackSlice(reinterpret_cast<uint8_t const*>(iter->value().data()));

    //// check format id
    VPackSlice idSlice = slice.get("id");
    if (!idSlice.isString()) {
      LOG_TOPIC("099d7", ERR, arangodb::Logger::STARTUP)
          << "found invalid database declaration with non-string id: "
          << slice.toJson();
      THROW_ARANGO_EXCEPTION(TRI_ERROR_ARANGO_ILLEGAL_PARAMETER_FILE);
    }

    // deleted
    if (arangodb::basics::VelocyPackHelper::getBooleanValue(slice, "deleted", false)) {
      TRI_voc_tick_t id = static_cast<TRI_voc_tick_t>(
          basics::StringUtils::uint64(idSlice.copyString()));

      // database is deleted, skip it!
      LOG_TOPIC("43cbc", DEBUG, arangodb::Logger::STARTUP) << "found dropped database " << id;

      dropDatabase(id);
      continue;
    }

    // name
    VPackSlice nameSlice = slice.get("name");
    if (!nameSlice.isString()) {
      LOG_TOPIC("96ffc", ERR, arangodb::Logger::STARTUP)
          << "found invalid database declaration with non-string name: "
          << slice.toJson();
      THROW_ARANGO_EXCEPTION(TRI_ERROR_ARANGO_ILLEGAL_PARAMETER_FILE);
    }

    result.add(slice);
  }
  result.close();
}

void RocksDBEngine::getCollectionInfo(TRI_vocbase_t& vocbase, TRI_voc_cid_t cid,
                                      arangodb::velocypack::Builder& builder,
                                      bool includeIndexes, TRI_voc_tick_t maxTick) {
  builder.openObject();

  // read collection info from database
  RocksDBKey key;

  key.constructCollection(vocbase.id(), cid);

  rocksdb::PinnableSlice value;
  rocksdb::ReadOptions options;
  rocksdb::Status res =
      _db->Get(options, RocksDBColumnFamily::definitions(), key.string(), &value);
  auto result = rocksutils::convertStatus(res);

  if (result.errorNumber() != TRI_ERROR_NO_ERROR) {
    THROW_ARANGO_EXCEPTION(result);
  }

  VPackSlice fullParameters = RocksDBValue::data(value);

  builder.add("parameters", fullParameters);

  if (includeIndexes) {
    // dump index information
    VPackSlice indexes = fullParameters.get("indexes");
    builder.add(VPackValue("indexes"));
    builder.openArray();

    if (indexes.isArray()) {
      for (auto const idx : VPackArrayIterator(indexes)) {
        // This is only allowed to contain user-defined indexes.
        // So we have to exclude Primary + Edge Types
        auto type = idx.get(StaticStrings::IndexType);
        TRI_ASSERT(type.isString());

        if (!type.isEqualString("primary") && !type.isEqualString("edge")) {
          builder.add(idx);
        }
      }
    }

    builder.close();
  }

  builder.close();
}

int RocksDBEngine::getCollectionsAndIndexes(TRI_vocbase_t& vocbase,
                                            arangodb::velocypack::Builder& result,
                                            bool wasCleanShutdown, bool isUpgrade) {
  rocksdb::ReadOptions readOptions;
  std::unique_ptr<rocksdb::Iterator> iter(
      _db->NewIterator(readOptions, RocksDBColumnFamily::definitions()));

  result.openArray();

  auto rSlice = rocksDBSlice(RocksDBEntryType::Collection);

  for (iter->Seek(rSlice); iter->Valid() && iter->key().starts_with(rSlice); iter->Next()) {
    if (vocbase.id() != RocksDBKey::databaseId(iter->key())) {
      continue;
    }

    auto slice = VPackSlice(reinterpret_cast<uint8_t const*>(iter->value().data()));

    if (arangodb::basics::VelocyPackHelper::readBooleanValue(slice, StaticStrings::DataSourceDeleted,
                                                             false)) {
      continue;
    }

    result.add(slice);
  }

  result.close();

  return TRI_ERROR_NO_ERROR;
}

int RocksDBEngine::getViews(TRI_vocbase_t& vocbase, arangodb::velocypack::Builder& result) {
  auto bounds = RocksDBKeyBounds::DatabaseViews(vocbase.id());
  rocksdb::Slice upper = bounds.end();
  rocksdb::ColumnFamilyHandle* cf = RocksDBColumnFamily::definitions();

  rocksdb::ReadOptions ro;
  ro.iterate_upper_bound = &upper;

  std::unique_ptr<rocksdb::Iterator> iter(_db->NewIterator(ro, cf));
  result.openArray();
  for (iter->Seek(bounds.start()); iter->Valid(); iter->Next()) {
    TRI_ASSERT(iter->key().compare(bounds.end()) < 0);
    auto slice = VPackSlice(reinterpret_cast<uint8_t const*>(iter->value().data()));

    LOG_TOPIC("e3bcd", TRACE, Logger::VIEWS) << "got view slice: " << slice.toJson();

    if (arangodb::basics::VelocyPackHelper::readBooleanValue(slice, StaticStrings::DataSourceDeleted,
                                                             false)) {
      continue;
    }

    result.add(slice);
  }

  result.close();

  return TRI_ERROR_NO_ERROR;
}

std::string RocksDBEngine::versionFilename(TRI_voc_tick_t id) const {
  return _basePath + TRI_DIR_SEPARATOR_CHAR + "VERSION-" + std::to_string(id);
}

void RocksDBEngine::cleanupReplicationContexts() {
  if (_replicationManager != nullptr) {
    _replicationManager->dropAll();
  }
}

VPackBuilder RocksDBEngine::getReplicationApplierConfiguration(TRI_vocbase_t& vocbase,
                                                               int& status) {
  RocksDBKey key;

  key.constructReplicationApplierConfig(vocbase.id());

  return getReplicationApplierConfiguration(key, status);
}

VPackBuilder RocksDBEngine::getReplicationApplierConfiguration(int& status) {
  RocksDBKey key;
  key.constructReplicationApplierConfig(databaseIdForGlobalApplier);
  return getReplicationApplierConfiguration(key, status);
}

VPackBuilder RocksDBEngine::getReplicationApplierConfiguration(RocksDBKey const& key,
                                                               int& status) {
  rocksdb::PinnableSlice value;

  auto db = rocksutils::globalRocksDB();
  auto opts = rocksdb::ReadOptions();
  auto s = db->Get(opts, RocksDBColumnFamily::definitions(), key.string(), &value);
  if (!s.ok()) {
    status = TRI_ERROR_FILE_NOT_FOUND;
    return arangodb::velocypack::Builder();
  }

  status = TRI_ERROR_NO_ERROR;
  VPackBuilder builder;
  builder.add(RocksDBValue::data(value));
  return builder;
}

int RocksDBEngine::removeReplicationApplierConfiguration(TRI_vocbase_t& vocbase) {
  RocksDBKey key;

  key.constructReplicationApplierConfig(vocbase.id());

  return removeReplicationApplierConfiguration(key);
}

int RocksDBEngine::removeReplicationApplierConfiguration() {
  RocksDBKey key;
  key.constructReplicationApplierConfig(databaseIdForGlobalApplier);
  return removeReplicationApplierConfiguration(key);
}

int RocksDBEngine::removeReplicationApplierConfiguration(RocksDBKey const& key) {
  auto status = rocksutils::globalRocksDBRemove(RocksDBColumnFamily::definitions(),
                                                key.string());
  if (!status.ok()) {
    return status.errorNumber();
  }

  return TRI_ERROR_NO_ERROR;
}

int RocksDBEngine::saveReplicationApplierConfiguration(TRI_vocbase_t& vocbase,
                                                       velocypack::Slice slice, bool doSync) {
  RocksDBKey key;

  key.constructReplicationApplierConfig(vocbase.id());

  return saveReplicationApplierConfiguration(key, slice, doSync);
}

int RocksDBEngine::saveReplicationApplierConfiguration(arangodb::velocypack::Slice slice,
                                                       bool doSync) {
  RocksDBKey key;
  key.constructReplicationApplierConfig(databaseIdForGlobalApplier);
  return saveReplicationApplierConfiguration(key, slice, doSync);
}

int RocksDBEngine::saveReplicationApplierConfiguration(RocksDBKey const& key,
                                                       arangodb::velocypack::Slice slice,
                                                       bool doSync) {
  auto value = RocksDBValue::ReplicationApplierConfig(slice);

  auto status = rocksutils::globalRocksDBPut(RocksDBColumnFamily::definitions(),
                                             key.string(), value.string());
  if (!status.ok()) {
    return status.errorNumber();
  }

  return TRI_ERROR_NO_ERROR;
}

// database, collection and index management
// -----------------------------------------

std::unique_ptr<TRI_vocbase_t> RocksDBEngine::openDatabase(arangodb::velocypack::Slice const& args,
                                                           bool isUpgrade, bool isVersionCheck, int& status) {
  VPackSlice idSlice = args.get("id");
  TRI_voc_tick_t id =
      static_cast<TRI_voc_tick_t>(basics::StringUtils::uint64(idSlice.copyString()));
  std::string const name = args.get("name").copyString();

  status = TRI_ERROR_NO_ERROR;

  return openExistingDatabase(id, name, true, isUpgrade, isVersionCheck);
}

std::unique_ptr<TRI_vocbase_t> RocksDBEngine::createDatabase(
    TRI_voc_tick_t id, arangodb::velocypack::Slice const& args, int& status) {
  status = TRI_ERROR_NO_ERROR;

  return std::make_unique<TRI_vocbase_t>(TRI_VOCBASE_TYPE_NORMAL, id,
                                         args.get("name").copyString());
}

int RocksDBEngine::writeCreateDatabaseMarker(TRI_voc_tick_t id, VPackSlice const& slice) {
  Result res = writeDatabaseMarker(id, slice, RocksDBLogValue::DatabaseCreate(id));
  return res.errorNumber();
}

Result RocksDBEngine::writeDatabaseMarker(TRI_voc_tick_t id, VPackSlice const& slice,
                                          RocksDBLogValue&& logValue) {
  RocksDBKey key;
  key.constructDatabase(id);
  auto value = RocksDBValue::Database(slice);
  rocksdb::WriteOptions wo;

  // Write marker + key into RocksDB inside one batch
  rocksdb::WriteBatch batch;
  batch.PutLogData(logValue.slice());
  batch.Put(RocksDBColumnFamily::definitions(), key.string(), value.string());
  rocksdb::Status res = _db->GetRootDB()->Write(wo, &batch);
  return rocksutils::convertStatus(res);
}

int RocksDBEngine::writeCreateCollectionMarker(TRI_voc_tick_t databaseId,
                                               TRI_voc_cid_t cid, VPackSlice const& slice,
                                               RocksDBLogValue&& logValue) {
  rocksdb::DB* db = _db->GetRootDB();

  RocksDBKey key;
  key.constructCollection(databaseId, cid);
  auto value = RocksDBValue::Collection(slice);

  rocksdb::WriteOptions wo;
  // Write marker + key into RocksDB inside one batch
  rocksdb::WriteBatch batch;
  if (logValue.slice().size() > 0) {
    batch.PutLogData(logValue.slice());
  }
  batch.Put(RocksDBColumnFamily::definitions(), key.string(), value.string());
  rocksdb::Status res = db->Write(wo, &batch);

  auto result = rocksutils::convertStatus(res);
  return result.errorNumber();
}

void RocksDBEngine::prepareDropDatabase(TRI_vocbase_t& vocbase,
                                        bool useWriteMarker, int& status) {
  VPackBuilder builder;

  builder.openObject();
  builder.add("id", velocypack::Value(std::to_string(vocbase.id())));
  builder.add("name", velocypack::Value(vocbase.name()));
  builder.add("deleted", VPackValue(true));
  builder.close();

  auto log = RocksDBLogValue::DatabaseDrop(vocbase.id());
  auto res = writeDatabaseMarker(vocbase.id(), builder.slice(), std::move(log));

  status = res.errorNumber();
}

Result RocksDBEngine::dropDatabase(TRI_vocbase_t& database) {
  replicationManager()->drop(&database);

  return dropDatabase(database.id());
}

void RocksDBEngine::waitUntilDeletion(TRI_voc_tick_t /* id */, bool /* force */, int& status) {
  // can delete databases instantly
  status = TRI_ERROR_NO_ERROR;
}

// current recovery state
RecoveryState RocksDBEngine::recoveryState() noexcept {
  return RocksDBRecoveryManager::instance()->recoveryState();
}

// current recovery tick
TRI_voc_tick_t RocksDBEngine::recoveryTick() noexcept {
  return TRI_voc_tick_t(RocksDBRecoveryManager::instance()->recoveryTick());
}

void RocksDBEngine::recoveryDone(TRI_vocbase_t& vocbase) {}

std::string RocksDBEngine::createCollection(TRI_vocbase_t& vocbase,
                                            LogicalCollection const& collection) {
  const TRI_voc_cid_t cid = collection.id();
  TRI_ASSERT(cid != 0);

  auto builder = collection.toVelocyPackIgnore(
      {"path", "statusString"},
      LogicalDataSource::makeFlags(LogicalDataSource::Serialize::Detailed,
                                   LogicalDataSource::Serialize::ForPersistence,
                                   LogicalDataSource::Serialize::IncludeInProgress));
  TRI_UpdateTickServer(static_cast<TRI_voc_tick_t>(cid));

  int res =
      writeCreateCollectionMarker(vocbase.id(), cid, builder.slice(),
                                  RocksDBLogValue::CollectionCreate(vocbase.id(), cid));

  if (res != TRI_ERROR_NO_ERROR) {
    THROW_ARANGO_EXCEPTION(res);
  }

#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
  auto* rcoll = toRocksDBCollection(collection.getPhysical());

  TRI_ASSERT(rcoll->numberDocuments() == 0);
#endif

  return std::string();  // no need to return a path
}

arangodb::Result RocksDBEngine::persistCollection(TRI_vocbase_t& vocbase,
                                                  LogicalCollection const& collection) {
  return {};
}

arangodb::Result RocksDBEngine::dropCollection(TRI_vocbase_t& vocbase,
                                               LogicalCollection& collection) {
  auto* coll = toRocksDBCollection(collection);
  bool const prefixSameAsStart = true;
  bool const useRangeDelete = coll->numberDocuments() >= 32 * 1024;

  rocksdb::DB* db = _db->GetRootDB();

  // If we get here the collection is safe to drop.
  //
  // This uses the following workflow:
  // 1. Persist the drop.
  //   * if this fails the collection will remain!
  //   * if this succeeds the collection is gone from user point
  // 2. Drop all Documents
  //   * If this fails we give up => We have data-garbage in RocksDB, Collection
  //   is gone.
  // 3. Drop all Indexes
  //   * If this fails we give up => We have data-garbage in RocksDB, Collection
  //   is gone.
  // 4. If all succeeds we do not have data-garbage, all is gone.
  //
  // (NOTE: The above fails can only occur on full HDD or Machine dying. No
  // write conflicts possible)

  TRI_ASSERT(collection.status() == TRI_VOC_COL_STATUS_DELETED);

  // Prepare collection remove batch
  rocksdb::WriteBatch batch;
  RocksDBLogValue logValue =
      RocksDBLogValue::CollectionDrop(vocbase.id(), collection.id(),
                                      arangodb::velocypack::StringRef(collection.guid()));
  batch.PutLogData(logValue.slice());

  RocksDBKey key;
  key.constructCollection(vocbase.id(), collection.id());
  batch.Delete(RocksDBColumnFamily::definitions(), key.string());

  rocksdb::WriteOptions wo;
  rocksdb::Status s = db->Write(wo, &batch);

  // TODO FAILURE Simulate !res.ok()
  if (!s.ok()) {
    // Persisting the drop failed. Do NOT drop collection.
    return rocksutils::convertStatus(s);
  }

  // Now Collection is gone.
  // Cleanup data-mess

  // Unregister collection metadata
  Result res = RocksDBCollectionMeta::deleteCollectionMeta(db, coll->objectId());
  if (res.fail()) {
    LOG_TOPIC("2c890", ERR, Logger::ENGINES) << "error removing collection meta-data: "
                                    << res.errorMessage();  // continue regardless
  }

  // remove from map
  {
    WRITE_LOCKER(guard, _mapLock);
    _collectionMap.erase(coll->objectId());
  }

  // delete indexes, RocksDBIndex::drop() has its own check
  std::vector<std::shared_ptr<Index>> vecShardIndex = coll->getIndexes();
  TRI_ASSERT(!vecShardIndex.empty());

  for (auto& index : vecShardIndex) {
    RocksDBIndex* ridx = static_cast<RocksDBIndex*>(index.get());
    res = RocksDBCollectionMeta::deleteIndexEstimate(db, ridx->objectId());
    if (res.fail()) {
      LOG_TOPIC("f2d51", WARN, Logger::ENGINES)
          << "could not delete index estimate: " << res.errorMessage();
    }

    auto dropRes = index->drop().errorNumber();

    if (dropRes != TRI_ERROR_NO_ERROR) {
      // We try to remove all indexed values.
      // If it does not work they cannot be accessed any more and leaked.
      // User View remains consistent.
      LOG_TOPIC("97176", ERR, Logger::ENGINES)
          << "unable to drop index: " << TRI_errno_string(dropRes);
      //      return TRI_ERROR_NO_ERROR;
    }
  }

  // delete documents
  RocksDBKeyBounds bounds = RocksDBKeyBounds::CollectionDocuments(coll->objectId());
  auto result = rocksutils::removeLargeRange(db, bounds, prefixSameAsStart, useRangeDelete);

  if (result.fail()) {
    // We try to remove all documents.
    // If it does not work they cannot be accessed any more and leaked.
    // User View remains consistent.
    return TRI_ERROR_NO_ERROR;
  }

#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
  // check if documents have been deleted
  size_t numDocs = rocksutils::countKeyRange(rocksutils::globalRocksDB(), bounds, true);

  if (numDocs > 0) {
    std::string errorMsg(
        "deletion check in collection drop failed - not all documents in the "
        "index have been deleted. remaining: ");
    errorMsg.append(std::to_string(numDocs));
    THROW_ARANGO_EXCEPTION_MESSAGE(TRI_ERROR_INTERNAL, errorMsg);
  }
#endif

  // run compaction for data only if collection contained a considerable
  // amount of documents. otherwise don't run compaction, because it will
  // slow things down a lot, especially during tests that create/drop LOTS
  // of collections
  if (useRangeDelete) {
    coll->compact();
  }

  // if we get here all documents / indexes are gone.
  // We have no data garbage left.
  return Result();
}

void RocksDBEngine::destroyCollection(TRI_vocbase_t& /*vocbase*/, LogicalCollection& /*collection*/
) {
  // not required
}

void RocksDBEngine::changeCollection(TRI_vocbase_t& vocbase,
                                     LogicalCollection const& collection, bool doSync) {
  auto builder = collection.toVelocyPackIgnore(
      {"path", "statusString"},
      LogicalDataSource::makeFlags(LogicalDataSource::Serialize::Detailed,
                                   LogicalDataSource::Serialize::ForPersistence,
                                   LogicalDataSource::Serialize::IncludeInProgress));
  int res =
      writeCreateCollectionMarker(vocbase.id(), collection.id(), builder.slice(),
                                  RocksDBLogValue::CollectionChange(vocbase.id(),
                                                                    collection.id()));

  if (res != TRI_ERROR_NO_ERROR) {
    THROW_ARANGO_EXCEPTION(res);
  }
}

arangodb::Result RocksDBEngine::renameCollection(TRI_vocbase_t& vocbase,
                                                 LogicalCollection const& collection,
                                                 std::string const& oldName) {
  auto builder = collection.toVelocyPackIgnore(
      {"path", "statusString"},
      LogicalDataSource::makeFlags(LogicalDataSource::Serialize::Detailed,
                                   LogicalDataSource::Serialize::ForPersistence,
                                   LogicalDataSource::Serialize::IncludeInProgress));
  int res = writeCreateCollectionMarker(
      vocbase.id(), collection.id(), builder.slice(),
      RocksDBLogValue::CollectionRename(vocbase.id(), collection.id(), arangodb::velocypack::StringRef(oldName)));

  return arangodb::Result(res);
}

void RocksDBEngine::unloadCollection(TRI_vocbase_t& /*vocbase*/, LogicalCollection& collection) {
  collection.setStatus(TRI_VOC_COL_STATUS_UNLOADED);
}

Result RocksDBEngine::createView(TRI_vocbase_t& vocbase, TRI_voc_cid_t id,
                                 arangodb::LogicalView const& view) {
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
  LOG_TOPIC("0bad8", DEBUG, Logger::ENGINES) << "RocksDBEngine::createView";
#endif
  rocksdb::WriteBatch batch;
  rocksdb::WriteOptions wo;

  RocksDBKey key;
  key.constructView(vocbase.id(), id);
  RocksDBLogValue logValue = RocksDBLogValue::ViewCreate(vocbase.id(), id);

  VPackBuilder props;

  props.openObject();
  view.properties(props,
                  LogicalDataSource::makeFlags(LogicalDataSource::Serialize::Detailed,
                                               LogicalDataSource::Serialize::ForPersistence,
                                               LogicalDataSource::Serialize::IncludeInProgress));
  props.close();

  RocksDBValue const value = RocksDBValue::View(props.slice());

  // Write marker + key into RocksDB inside one batch
  batch.PutLogData(logValue.slice());
  batch.Put(RocksDBColumnFamily::definitions(), key.string(), value.string());

  auto res = _db->Write(wo, &batch);

  LOG_TOPIC_IF("cac6a", TRACE, Logger::VIEWS, !res.ok())
      << "could not create view: " << res.ToString();

  return rocksutils::convertStatus(res);
}

arangodb::Result RocksDBEngine::dropView(TRI_vocbase_t const& vocbase,
                                         LogicalView const& view) {
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
  LOG_TOPIC("fa6e5", DEBUG, Logger::ENGINES) << "RocksDBEngine::dropView";
#endif
  VPackBuilder builder;

  builder.openObject();
  view.properties(builder,
                  LogicalDataSource::makeFlags(LogicalDataSource::Serialize::Detailed,
                                               LogicalDataSource::Serialize::ForPersistence,
                                               LogicalDataSource::Serialize::IncludeInProgress));
  builder.close();

  auto logValue =
      RocksDBLogValue::ViewDrop(vocbase.id(), view.id(), arangodb::velocypack::StringRef(view.guid()));
  RocksDBKey key;
  key.constructView(vocbase.id(), view.id());

  rocksdb::WriteBatch batch;
  batch.PutLogData(logValue.slice());
  batch.Delete(RocksDBColumnFamily::definitions(), key.string());

  rocksdb::WriteOptions wo;
  auto res = _db->GetRootDB()->Write(wo, &batch);
  LOG_TOPIC_IF("fcd22", TRACE, Logger::VIEWS, !res.ok())
      << "could not create view: " << res.ToString();
  return rocksutils::convertStatus(res);
}

void RocksDBEngine::destroyView(TRI_vocbase_t const& /*vocbase*/, LogicalView const& /*view*/
                                ) noexcept {
  // nothing to do here
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
  LOG_TOPIC("5d5f7", DEBUG, Logger::ENGINES) << "RocksDBEngine::destroyView";
#endif
}

Result RocksDBEngine::changeView(TRI_vocbase_t& vocbase,
                                 arangodb::LogicalView const& view, bool /*doSync*/
) {
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
  LOG_TOPIC("405da", DEBUG, Logger::ENGINES) << "RocksDBEngine::changeView";
#endif
  if (inRecovery()) {
    // nothing to do
    return {};
  }

  RocksDBKey key;

  key.constructView(vocbase.id(), view.id());

  VPackBuilder infoBuilder;

  infoBuilder.openObject();
  view.properties(infoBuilder,
                  LogicalDataSource::makeFlags(LogicalDataSource::Serialize::Detailed,
                                               LogicalDataSource::Serialize::ForPersistence,
                                               LogicalDataSource::Serialize::IncludeInProgress));
  infoBuilder.close();

  RocksDBLogValue log = RocksDBLogValue::ViewChange(vocbase.id(), view.id());
  RocksDBValue const value = RocksDBValue::View(infoBuilder.slice());

  rocksdb::WriteBatch batch;
  rocksdb::WriteOptions wo;  // TODO: check which options would make sense
  rocksdb::Status s;

  s = batch.PutLogData(log.slice());

  if (!s.ok()) {
    LOG_TOPIC("6d6a4", TRACE, Logger::VIEWS)
        << "failed to write change view marker " << s.ToString();
    return rocksutils::convertStatus(s);
  }

  s = batch.Put(RocksDBColumnFamily::definitions(), key.string(), value.string());

  if (!s.ok()) {
    LOG_TOPIC("ebb58", TRACE, Logger::VIEWS)
        << "failed to write change view marker " << s.ToString();
    return rocksutils::convertStatus(s);
  }
  auto db = rocksutils::globalRocksDB();
  auto res = db->Write(wo, &batch);
  LOG_TOPIC_IF("6ee8a", TRACE, Logger::VIEWS, !res.ok())
      << "could not change view: " << res.ToString();
  return rocksutils::convertStatus(res);
}

void RocksDBEngine::signalCleanup(TRI_vocbase_t& vocbase) {
  // nothing to do here
}

int RocksDBEngine::shutdownDatabase(TRI_vocbase_t& vocbase) {
  return TRI_ERROR_NO_ERROR;
}

/// @brief Add engine-specific optimizer rules
void RocksDBEngine::addOptimizerRules() {
  RocksDBOptimizerRules::registerResources();
}

/// @brief Add engine-specific V8 functions
void RocksDBEngine::addV8Functions() {
  // there are no specific V8 functions here
  RocksDBV8Functions::registerResources();
}

/// @brief Add engine-specific REST handlers
void RocksDBEngine::addRestHandlers(rest::RestHandlerFactory& handlerFactory) {
  RocksDBRestHandlers::registerResources(&handlerFactory);
}

void RocksDBEngine::addCollectionMapping(uint64_t objectId, TRI_voc_tick_t did,
                                         TRI_voc_cid_t cid) {
  if (objectId != 0) {
    WRITE_LOCKER(guard, _mapLock);
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
    auto it = _collectionMap.find(objectId);
    if (it != _collectionMap.end()) {
      TRI_ASSERT(it->second.first == did);
      TRI_ASSERT(it->second.second == cid);
    }
#endif
    _collectionMap[objectId] = std::make_pair(did, cid);
  }
}

std::vector<std::pair<TRI_voc_tick_t, TRI_voc_cid_t>> RocksDBEngine::collectionMappings() const {
  std::vector<std::pair<TRI_voc_tick_t, TRI_voc_cid_t>> res;
  READ_LOCKER(guard, _mapLock);
  for (auto const& it : _collectionMap) {
    res.emplace_back(it.second.first, it.second.second);
  }
  return res;
}

void RocksDBEngine::addIndexMapping(uint64_t objectId, TRI_voc_tick_t did,
                                    TRI_voc_cid_t cid, TRI_idx_iid_t iid) {
  if (objectId != 0) {
    WRITE_LOCKER(guard, _mapLock);
#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
    auto it = _indexMap.find(objectId);
    if (it != _indexMap.end()) {
      TRI_ASSERT(std::get<0>(it->second) == did);
      TRI_ASSERT(std::get<1>(it->second) == cid);
      TRI_ASSERT(std::get<2>(it->second) == iid);
    }
#endif
    _indexMap[objectId] = std::make_tuple(did, cid, iid);
  }
}

void RocksDBEngine::removeIndexMapping(uint64_t objectId) {
  if (objectId != 0) {
    WRITE_LOCKER(guard, _mapLock);
    _indexMap.erase(objectId);
  }
}

RocksDBEngine::CollectionPair RocksDBEngine::mapObjectToCollection(uint64_t objectId) const {
  READ_LOCKER(guard, _mapLock);
  auto it = _collectionMap.find(objectId);
  if (it == _collectionMap.end()) {
    return {0, 0};
  }
  return it->second;
}

RocksDBEngine::IndexTriple RocksDBEngine::mapObjectToIndex(uint64_t objectId) const {
  READ_LOCKER(guard, _mapLock);
  auto it = _indexMap.find(objectId);
  if (it == _indexMap.end()) {
    return RocksDBEngine::IndexTriple(0, 0, 0);
  }
  return it->second;
}

/// @brief return a list of the currently open WAL files
std::vector<std::string> RocksDBEngine::currentWalFiles() const {
  rocksdb::VectorLogPtr files;
  std::vector<std::string> names;

  auto status = _db->GetSortedWalFiles(files);
  if (!status.ok()) {
    return names;  // TODO: error here?
  }

  for (size_t current = 0; current < files.size(); current++) {
    auto f = files[current].get();
    try {
      names.push_back(f->PathName());
    } catch (...) {
      return names;
    }
  }

  return names;
}

Result RocksDBEngine::flushWal(bool waitForSync, bool waitForCollector,
                               bool /*writeShutdownFile*/) {
  if (_syncThread) {
    // _syncThread may be a nullptr, in case automatic syncing is turned off
    _syncThread->syncWal();
  }

  if (waitForCollector) {
    rocksdb::FlushOptions flushOptions;
    flushOptions.wait = waitForSync;

    for (auto cf : RocksDBColumnFamily::_allHandles) {
      rocksdb::Status status = _db->GetBaseDB()->Flush(flushOptions, cf);
      if (!status.ok()) {
        return rocksutils::convertStatus(status);
      }
    }
  }
  return Result();
}

void RocksDBEngine::waitForEstimatorSync(std::chrono::milliseconds maxWaitTime) {
  auto start = std::chrono::high_resolution_clock::now();
  auto beginSeq = _db->GetLatestSequenceNumber();

  while (std::chrono::high_resolution_clock::now() - start < maxWaitTime) {
    if (_settingsManager->earliestSeqNeeded() >= beginSeq) {
      // all synced up!
      break;
    }
    std::this_thread::sleep_for(std::chrono::milliseconds(50));
  }
}

Result RocksDBEngine::registerRecoveryHelper(std::shared_ptr<RocksDBRecoveryHelper> helper) {
  try {
    _recoveryHelpers.emplace_back(helper);
  } catch (std::bad_alloc const&) {
    return {TRI_ERROR_OUT_OF_MEMORY};
  }

  return {TRI_ERROR_NO_ERROR};
}

std::vector<std::shared_ptr<RocksDBRecoveryHelper>> const& RocksDBEngine::recoveryHelpers() {
  return _recoveryHelpers;
}

void RocksDBEngine::determinePrunableWalFiles(TRI_voc_tick_t minTickExternal) {
  WRITE_LOCKER(lock, _walFileLock);
  TRI_voc_tick_t minTickToKeep = std::min(_useReleasedTick ? _releasedTick : std::numeric_limits<TRI_voc_tick_t>::max(), minTickExternal);

  // Retrieve the sorted list of all wal files with earliest file first
  rocksdb::VectorLogPtr files;
  auto status = _db->GetSortedWalFiles(files);
  if (!status.ok()) {
    LOG_TOPIC("078ef", INFO, Logger::ENGINES) << "could not get WAL files " << status.ToString();
    return;
  }

  uint64_t totalArchiveSize = 0;
  for (size_t current = 0; current < files.size(); current++) {
    auto const& f = files[current].get();

    if (f->Type() != rocksdb::WalFileType::kArchivedLogFile) {
      // we are only interested in files of the archive
      continue;
    }

    // determine the size of the archive only if it there is a cap on the archive size
    // otherwise we can save the underlying file access
    if (_maxWalArchiveSizeLimit > 0) {
      totalArchiveSize += f->SizeFileBytes();
    }

    // check if there is another WAL file coming after the currently-looked-at
    // There should be at least one live WAL file after it, however, let's be
    // paranoid and do a proper check. If there is at least one WAL file following,
    // we need to take its start tick into account as well, because the following
    // file's start tick can be assumed to be the end tick of the current file!
    if (f->StartSequence() < minTickToKeep &&
        current < files.size() - 1) {
      auto const& n = files[current + 1].get();
      if (n->StartSequence() < minTickToKeep) {
        // this file will be removed because it does not contain any data we
        // still need
        if (_prunableWalFiles.find(f->PathName()) == _prunableWalFiles.end()) {
          LOG_TOPIC("9f7a4", DEBUG, Logger::ENGINES)
              << "RocksDB WAL file '" << f->PathName() << "' with start sequence "
              << f->StartSequence() << " added to prunable list because it is not needed anymore";
          _prunableWalFiles.emplace(f->PathName(), TRI_microtime() + _pruneWaitTime);
        }
      }
    }
  }

  if (_maxWalArchiveSizeLimit == 0) {
    // size of the archive is not restricted. done!
    return;
  }

  // print current archive size
  LOG_TOPIC("8d71b", TRACE, Logger::ENGINES) << "total size of the RocksDB WAL file archive: " << totalArchiveSize;

  if (totalArchiveSize <= _maxWalArchiveSizeLimit) {
    // archive is smaller than allowed. all good
    return;
  }

  // we got more archived files than configured. time for purging some files!
  for (size_t current = 0; current < files.size(); current++) {
    auto const& f = files[current].get();

    if (f->Type() != rocksdb::WalFileType::kArchivedLogFile) {
      continue;
    }

    // force pruning
    bool doPrint = false;
    auto it = _prunableWalFiles.find(f->PathName());

    if (it == _prunableWalFiles.end()) {
      doPrint = true;
      // using an expiration time of -1.0 indicates the file is subject to deletion
      // because the archive outgrew the maximum allowed size
      _prunableWalFiles.emplace(f->PathName(), -1.0);
    } else {
      // file already in list. now set its expiration time to the past
      // so we are sure it will get deleted

      // using an expiration time of -1.0 indicates the file is subject to deletion
      // because the archive outgrew the maximum allowed size
      if ((*it).second > 0.0) {
        doPrint = true;
      }
      (*it).second = -1.0;
    }

    if (doPrint) {
      LOG_TOPIC("d9793", WARN, Logger::ENGINES)
          << "forcing removal of RocksDB WAL file '" << f->PathName() << "' with start sequence "
          << f->StartSequence() << " because of overflowing archive. configured maximum archive size is " << _maxWalArchiveSizeLimit << ", actual archive size is: " << totalArchiveSize;
    }

    TRI_ASSERT(totalArchiveSize >= f->SizeFileBytes());
    totalArchiveSize -= f->SizeFileBytes();

    if (totalArchiveSize <= _maxWalArchiveSizeLimit) {
      // got enough files to remove
      break;
    }
  }
}

RocksDBFilePurgePreventer RocksDBEngine::disallowPurging() noexcept {
  return RocksDBFilePurgePreventer(this);
}

RocksDBFilePurgeEnabler RocksDBEngine::startPurging() noexcept {
  return RocksDBFilePurgeEnabler(this);
}

void RocksDBEngine::pruneWalFiles() {
  // this struct makes sure that no other threads enter WAL tailing while we
  // are in here. If there are already other threads in WAL tailing while we
  // get here, we go on and only remove the WAL files that are really safe
  // to remove
  RocksDBFilePurgeEnabler purgeEnabler(rocksutils::globalRocksEngine()->startPurging());

  WRITE_LOCKER(lock, _walFileLock);

  // go through the map of WAL files that we have already and check if they are
  // "expired"
  for (auto it = _prunableWalFiles.begin(); it != _prunableWalFiles.end();
       /* no hoisting */) {
    // check if WAL file is expired
    bool deleteFile = false;

    if ((*it).second <= 0.0) {
      // file can be deleted because we outgrew the configured max archive size,
      // but only if there are no other threads currently inside the WAL tailing
      // section
      deleteFile = purgeEnabler.canPurge();
    } else if ((*it).second < TRI_microtime()) {
      // file has expired, and it is always safe to delete it
      deleteFile = true;
    }

    if (deleteFile) {
      LOG_TOPIC("68e4a", DEBUG, Logger::ENGINES)
          << "deleting RocksDB WAL file '" << (*it).first << "'";
      rocksdb::Status s;
      if (basics::FileUtils::exists(basics::FileUtils::buildFilename(_options.wal_dir, (*it).first))) {
        // only attempt file deletion if the file actually exists.
        // otherwise RocksDB may complain about non-existing files and log a big error message
        s = _db->DeleteFile((*it).first);
      } else {
        LOG_TOPIC("c2cc9", DEBUG, Logger::ROCKSDB)
            << "to-be-deleted RocksDB WAL file '" << (*it).first << "' does not exist. skipping deletion";
      }
      // apparently there is a case where a file was already deleted
      // but is still in _prunableWalFiles. In this case we get an invalid
      // argument response.
      if (s.ok() || s.IsInvalidArgument()) {
        it = _prunableWalFiles.erase(it);
        continue;
      }
    }

    // cannot delete this file yet... must forward iterator to prevent an
    // endless loop
    ++it;
  }
}

Result RocksDBEngine::dropDatabase(TRI_voc_tick_t id) {
  using namespace rocksutils;
  arangodb::Result res;
  rocksdb::WriteOptions wo;
  rocksdb::DB* db = _db->GetRootDB();

  // remove view definitions
  res = rocksutils::removeLargeRange(db, RocksDBKeyBounds::DatabaseViews(id),
                                     true, /*rangeDel*/ false);
  if (res.fail()) {
    return res;
  }

#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
  size_t numDocsLeft = 0;
#endif

  // remove collections
  auto dbBounds = RocksDBKeyBounds::DatabaseCollections(id);
  iterateBounds(dbBounds, [&](rocksdb::Iterator* it) {
    RocksDBKey key(it->key());
    RocksDBValue value(RocksDBEntryType::Collection, it->value());

    uint64_t const objectId =
        basics::VelocyPackHelper::stringUInt64(value.slice(), "objectId");

    auto const cnt = RocksDBCollectionMeta::loadCollectionCount(_db, objectId);
    uint64_t const numberDocuments = cnt._added - cnt._removed;
    bool const useRangeDelete = numberDocuments >= 32 * 1024;

    // remove indexes
    VPackSlice indexes = value.slice().get("indexes");
    if (indexes.isArray()) {
      for (auto const& it : VPackArrayIterator(indexes)) {
        // delete index documents
        uint64_t objectId =
            basics::VelocyPackHelper::stringUInt64(it, "objectId");
        res = RocksDBCollectionMeta::deleteIndexEstimate(db, objectId);
        if (res.fail()) {
          return;
        }

        TRI_ASSERT(it.get(StaticStrings::IndexType).isString());
        auto type = Index::type(it.get(StaticStrings::IndexType).copyString());
        bool unique =
            basics::VelocyPackHelper::getBooleanValue(it, StaticStrings::IndexUnique, false);

        RocksDBKeyBounds bounds = RocksDBIndex::getBounds(type, objectId, unique);
        // edge index drop fails otherwise
        bool const prefixSameAsStart = type != Index::TRI_IDX_TYPE_EDGE_INDEX;
        res = rocksutils::removeLargeRange(db, bounds, prefixSameAsStart, useRangeDelete);
        if (res.fail()) {
          return;
        }

#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
        // check if documents have been deleted
        numDocsLeft += rocksutils::countKeyRange(db, bounds, prefixSameAsStart);
#endif
      }
    }

    // delete documents
    RocksDBKeyBounds bounds = RocksDBKeyBounds::CollectionDocuments(objectId);
    res = rocksutils::removeLargeRange(db, bounds, true, useRangeDelete);
    if (res.fail()) {
      LOG_TOPIC("6dbc6", WARN, Logger::ENGINES)
          << "error deleting collection documents: '" << res.errorMessage() << "'";
      return;
    }
    // delete collection meta-data
    res = RocksDBCollectionMeta::deleteCollectionMeta(db, objectId);
    if (res.fail()) {
      LOG_TOPIC("484d0", WARN, Logger::ENGINES)
          << "error deleting collection metadata: '" << res.errorMessage() << "'";
      return;
    }
    // remove collection entry
    rocksdb::Status s =
        db->Delete(wo, RocksDBColumnFamily::definitions(), value.string());
    if (!s.ok()) {
      LOG_TOPIC("64b4e", WARN, Logger::ENGINES)
          << "error deleting collection definition: " << s.ToString();
      return;
    }

#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
    // check if documents have been deleted
    numDocsLeft += rocksutils::countKeyRange(db, bounds, true);
#endif
  });

  if (res.fail()) {
    return res;
  }

  // remove database meta-data
  RocksDBKey key;
  key.constructDatabase(id);
  rocksdb::Status s = db->Delete(wo, RocksDBColumnFamily::definitions(), key.string());
  if (!s.ok()) {
    LOG_TOPIC("9948c", WARN, Logger::ENGINES)
        << "error deleting database definition: " << s.ToString();
  }

  // remove VERSION file for database. it's not a problem when this fails
  // because it will simply remain there and be ignored on subsequent starts
  TRI_UnlinkFile(versionFilename(id).c_str());

#ifdef ARANGODB_ENABLE_MAINTAINER_MODE
  if (numDocsLeft > 0) {
    std::string errorMsg(
        "deletion check in drop database failed - not all documents have been "
        "deleted. remaining: ");
    errorMsg.append(std::to_string(numDocsLeft));
    THROW_ARANGO_EXCEPTION_MESSAGE(TRI_ERROR_INTERNAL, errorMsg);
  }
#endif

  return res;
}

bool RocksDBEngine::systemDatabaseExists() {
  velocypack::Builder builder;
  getDatabases(builder);

  for (auto const& item : velocypack::ArrayIterator(builder.slice())) {
    if (item.get("name").copyString() == StaticStrings::SystemDatabase) {
      return true;
    }
  }
  return false;
}

void RocksDBEngine::addSystemDatabase() {
  // create system database entry
  TRI_voc_tick_t id = TRI_NewTickServer();
  VPackBuilder builder;
  builder.openObject();
  builder.add("id", VPackValue(std::to_string(id)));
  builder.add("name", VPackValue(StaticStrings::SystemDatabase));
  builder.add("deleted", VPackValue(false));
  builder.close();

  RocksDBLogValue log = RocksDBLogValue::DatabaseCreate(id);
  Result res = writeDatabaseMarker(id, builder.slice(), std::move(log));
  if (res.fail()) {
    LOG_TOPIC("8c26a", FATAL, arangodb::Logger::STARTUP)
        << "unable to write database marker: " << res.errorMessage();
    FATAL_ERROR_EXIT();
  }
}

/// @brief open an existing database. internal function
std::unique_ptr<TRI_vocbase_t> RocksDBEngine::openExistingDatabase(
    TRI_voc_tick_t id, std::string const& name, bool wasCleanShutdown, bool isUpgrade, bool isVersionCheck) {
  auto vocbase = std::make_unique<TRI_vocbase_t>(TRI_VOCBASE_TYPE_NORMAL, id, name);

  // scan the database path for views
  if (!isVersionCheck) {
    try {
      VPackBuilder builder;
      int res = getViews(*vocbase, builder);

      if (res != TRI_ERROR_NO_ERROR) {
        THROW_ARANGO_EXCEPTION(res);
      }

      VPackSlice const slice = builder.slice();
      TRI_ASSERT(slice.isArray());

      for (auto const& it : VPackArrayIterator(slice)) {
        TRI_ASSERT(!it.get("id").isNone());

        LogicalView::ptr view;
        auto res = LogicalView::instantiate(view, *vocbase, it);

        if (!res.ok()) {
          THROW_ARANGO_EXCEPTION(res);
        }

        if (!view) {
          THROW_ARANGO_EXCEPTION_MESSAGE( // exception
            TRI_ERROR_INTERNAL, // code
            std::string("failed to instantiate view in vocbase'") + vocbase->name() + "' from definition: " + it.toString()
          );
        }

        StorageEngine::registerView(*vocbase, view);

        view->open();
      }
    } catch (std::exception const& ex) {
      LOG_TOPIC("554b1", ERR, arangodb::Logger::ENGINES)
          << "error while opening database: " << ex.what();
      throw;
    } catch (...) {
      LOG_TOPIC("5933d", ERR, arangodb::Logger::ENGINES)
          << "error while opening database: unknown exception";
      throw;
    }
  }

  // scan the database path for collections
  try {
    VPackBuilder builder;
    int res = getCollectionsAndIndexes(*vocbase, builder, wasCleanShutdown, isUpgrade);

    if (res != TRI_ERROR_NO_ERROR) {
      THROW_ARANGO_EXCEPTION(res);
    }

    VPackSlice slice = builder.slice();
    TRI_ASSERT(slice.isArray());

    for (VPackSlice it : VPackArrayIterator(slice)) {
      // we found a collection that is still active
      TRI_ASSERT(!it.get("id").isNone() || !it.get("cid").isNone());
      auto uniqCol = std::make_shared<arangodb::LogicalCollection>(*vocbase, it, false);
      auto collection = uniqCol.get();
      TRI_ASSERT(collection != nullptr);

      auto phy = static_cast<RocksDBCollection*>(collection->getPhysical());
      TRI_ASSERT(phy != nullptr);
      phy->meta().deserializeMeta(_db, *collection);
      phy->loadInitialNumberDocuments();

      StorageEngine::registerCollection(*vocbase, uniqCol);
      LOG_TOPIC("39404", DEBUG, arangodb::Logger::ENGINES)
          << "added document collection '" << collection->name() << "'";
    }

    return vocbase;
  } catch (std::exception const& ex) {
    LOG_TOPIC("8d427", ERR, arangodb::Logger::ENGINES)
        << "error while opening database: " << ex.what();
    throw;
  } catch (...) {
    LOG_TOPIC("0268e", ERR, arangodb::Logger::ENGINES)
        << "error while opening database: unknown exception";
    throw;
  }
}

void RocksDBEngine::getStatistics(VPackBuilder& builder) const {
  // add int properties
  auto addInt = [&](std::string const& s) {
    std::string v;
    if (_db->GetProperty(s, &v)) {
      int64_t i = basics::StringUtils::int64(v);
      builder.add(s, VPackValue(i));
    }
  };

  // add string properties
  auto addStr = [&](std::string const& s) {
    std::string v;
    if (_db->GetProperty(s, &v)) {
      builder.add(s, VPackValue(v));
    }
  };

  // get string property from each column family and return sum;
  auto addIntAllCf = [&](std::string const& s) {
    int64_t sum = 0;
    for (auto cfh : RocksDBColumnFamily::_allHandles) {
      std::string v;
      if (_db->GetProperty(cfh, s, &v)) {
        int64_t temp=basics::StringUtils::int64(v);

        // -1 returned for somethings that are valid property but no value
        if (0 < temp) {
          sum += temp;
        } // if
      } // if
    } // for
    builder.add(s, VPackValue(sum));
  };

  // add column family properties
  auto addCf = [&](std::string const& name, rocksdb::ColumnFamilyHandle* c) {
    std::string v;
    builder.add(name, VPackValue(VPackValueType::Object));
    if (_db->GetProperty(c, rocksdb::DB::Properties::kCFStats, &v)) {
      builder.add("dbstats", VPackValue(v));
    }

    // re-add this line to count all keys in the column family (slow!!!)
    // builder.add("keys", VPackValue(rocksutils::countKeys(_db, c)));

    // estimate size on disk and in memtables
    uint64_t out = 0;
    rocksdb::Range r(rocksdb::Slice("\x00\x00\x00\x00\x00\x00\x00\x00", 8), rocksdb::Slice("\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff\xff",
                                                                                           16));

    _db->GetApproximateSizes(c, &r, 1, &out,
                             static_cast<uint8_t>(
                                 rocksdb::DB::SizeApproximationFlags::INCLUDE_MEMTABLES |
                                 rocksdb::DB::SizeApproximationFlags::INCLUDE_FILES));

    builder.add("memory", VPackValue(out));
    builder.close();
  };

  builder.openObject();
  for (int i = 0; i < _options.num_levels; ++i) {
    addIntAllCf(rocksdb::DB::Properties::kNumFilesAtLevelPrefix + std::to_string(i));
    // ratio needs new calculation with all cf, not a simple add operation
    addIntAllCf(rocksdb::DB::Properties::kCompressionRatioAtLevelPrefix + std::to_string(i));
  }
  // caution:  you must read rocksdb/db/interal_stats.cc carefully to
  //           determine if a property is for whole database or one column family
  addIntAllCf(rocksdb::DB::Properties::kNumImmutableMemTable);
  addIntAllCf(rocksdb::DB::Properties::kNumImmutableMemTableFlushed);
  addIntAllCf(rocksdb::DB::Properties::kMemTableFlushPending);
  addIntAllCf(rocksdb::DB::Properties::kCompactionPending);
  addInt(rocksdb::DB::Properties::kBackgroundErrors);
  addIntAllCf(rocksdb::DB::Properties::kCurSizeActiveMemTable);
  addIntAllCf(rocksdb::DB::Properties::kCurSizeAllMemTables);
  addIntAllCf(rocksdb::DB::Properties::kSizeAllMemTables);
  addIntAllCf(rocksdb::DB::Properties::kNumEntriesActiveMemTable);
  addIntAllCf(rocksdb::DB::Properties::kNumEntriesImmMemTables);
  addIntAllCf(rocksdb::DB::Properties::kNumDeletesActiveMemTable);
  addIntAllCf(rocksdb::DB::Properties::kNumDeletesImmMemTables);
  addIntAllCf(rocksdb::DB::Properties::kEstimateNumKeys);
  addIntAllCf(rocksdb::DB::Properties::kEstimateTableReadersMem);
  addInt(rocksdb::DB::Properties::kNumSnapshots);
  addInt(rocksdb::DB::Properties::kOldestSnapshotTime);
  addIntAllCf(rocksdb::DB::Properties::kNumLiveVersions);
  addInt(rocksdb::DB::Properties::kMinLogNumberToKeep);
  addIntAllCf(rocksdb::DB::Properties::kEstimateLiveDataSize);
  addIntAllCf(rocksdb::DB::Properties::kLiveSstFilesSize);
  addStr(rocksdb::DB::Properties::kDBStats);
  addStr(rocksdb::DB::Properties::kSSTables);
  addInt(rocksdb::DB::Properties::kNumRunningCompactions);
  addInt(rocksdb::DB::Properties::kNumRunningFlushes);
  addInt(rocksdb::DB::Properties::kIsFileDeletionsEnabled);
  addIntAllCf(rocksdb::DB::Properties::kEstimatePendingCompactionBytes);
  addInt(rocksdb::DB::Properties::kBaseLevel);
  addInt(rocksdb::DB::Properties::kBlockCacheCapacity);
  addInt(rocksdb::DB::Properties::kBlockCacheUsage);
  addInt(rocksdb::DB::Properties::kBlockCachePinnedUsage);
  addIntAllCf(rocksdb::DB::Properties::kTotalSstFilesSize);
  addInt(rocksdb::DB::Properties::kActualDelayedWriteRate);
  addInt(rocksdb::DB::Properties::kIsWriteStopped);

  if (_options.statistics) {
    for (auto const& stat : rocksdb::TickersNameMap) {
      builder.add(stat.second, VPackValue(_options.statistics->getTickerCount(stat.first)));
    }

    uint64_t walWrite, flushWrite, compactionWrite, userWrite;
    walWrite = _options.statistics->getTickerCount(rocksdb::WAL_FILE_BYTES);
    flushWrite = _options.statistics->getTickerCount(rocksdb::FLUSH_WRITE_BYTES);
    compactionWrite = _options.statistics->getTickerCount(rocksdb::COMPACT_WRITE_BYTES);
    userWrite = _options.statistics->getTickerCount(rocksdb::BYTES_WRITTEN);
    builder.add("rocksdbengine.write.amplification.x100",
                VPackValue( (0 != userWrite) ? ((walWrite+flushWrite+compactionWrite)*100)/userWrite : 100));
  }

  cache::Manager* manager = CacheManagerFeature::MANAGER;
  if (manager != nullptr) {
    // cache turned on
    auto rates = manager->globalHitRates();
    builder.add("cache.limit", VPackValue(manager->globalLimit()));
    builder.add("cache.allocated", VPackValue(manager->globalAllocation()));
    // handle NaN
    builder.add("cache.hit-rate-lifetime",
                VPackValue(rates.first >= 0.0 ? rates.first : 0.0));
    builder.add("cache.hit-rate-recent",
                VPackValue(rates.second >= 0.0 ? rates.second : 0.0));
  } else {
    // cache turned off
    builder.add("cache.limit", VPackValue(0));
    builder.add("cache.allocated", VPackValue(0));
    // handle NaN
    builder.add("cache.hit-rate-lifetime", VPackValue(0));
    builder.add("cache.hit-rate-recent", VPackValue(0));
  }

  // print column family statistics
  //  warning: output format limits numbers to 3 digits of precision or less.
  builder.add("columnFamilies", VPackValue(VPackValueType::Object));
  addCf("definitions", RocksDBColumnFamily::definitions());
  addCf("documents", RocksDBColumnFamily::documents());
  addCf("primary", RocksDBColumnFamily::primary());
  addCf("edge", RocksDBColumnFamily::edge());
  addCf("vpack", RocksDBColumnFamily::vpack());
  addCf("geo", RocksDBColumnFamily::geo());
  addCf("fulltext", RocksDBColumnFamily::fulltext());
  builder.close();

  if (_listener) {
    builder.add("rocksdbengine.throttle.bps", VPackValue(_listener->GetThrottle()));
  } // if

  builder.close();
}

Result RocksDBEngine::handleSyncKeys(DatabaseInitialSyncer& syncer,
                                     LogicalCollection& col, std::string const& keysId) {
  return handleSyncKeysRocksDB(syncer, &col, keysId);
}

Result RocksDBEngine::createLoggerState(TRI_vocbase_t* vocbase, VPackBuilder& builder) {
  builder.openObject();  // Base
  rocksdb::SequenceNumber lastTick = _db->GetLatestSequenceNumber();

  // "state" part
  builder.add("state", VPackValue(VPackValueType::Object));  // open
  builder.add("running", VPackValue(true));
  builder.add("lastLogTick", VPackValue(std::to_string(lastTick)));
  builder.add("lastUncommittedLogTick", VPackValue(std::to_string(lastTick)));
  builder.add("totalEvents",
              VPackValue(lastTick));  // s.numEvents + s.numEventsSync
  builder.add("time", VPackValue(utilities::timeString()));
  builder.close();

  // "server" part
  builder.add("server", VPackValue(VPackValueType::Object));  // open
  builder.add("version", VPackValue(ARANGODB_VERSION));
  builder.add("serverId", VPackValue(std::to_string(ServerIdFeature::getId())));
  builder.add("engine", VPackValue(EngineName));  // "rocksdb"
  builder.close();

  // "clients" part
  builder.add("clients", VPackValue(VPackValueType::Array));  // open
  if (vocbase != nullptr) {
    vocbase->replicationClients().toVelocyPack(builder);
  }
  builder.close();  // clients

  builder.close();  // base

  return Result{};
}

Result RocksDBEngine::createTickRanges(VPackBuilder& builder) {
  rocksdb::TransactionDB* tdb = rocksutils::globalRocksDB();
  rocksdb::VectorLogPtr walFiles;
  rocksdb::Status s = tdb->GetSortedWalFiles(walFiles);
  Result res = rocksutils::convertStatus(s);
  if (res.fail()) {
    return res;
  }

  builder.openArray();
  for (auto lfile = walFiles.begin(); lfile != walFiles.end(); ++lfile) {
    auto& logfile = *lfile;
    builder.openObject();
    // filename and state are already of type string
    builder.add("datafile", VPackValue(logfile->PathName()));
    if (logfile->Type() == rocksdb::WalFileType::kAliveLogFile) {
      builder.add("status", VPackValue("open"));
    } else if (logfile->Type() == rocksdb::WalFileType::kArchivedLogFile) {
      builder.add("status", VPackValue("collected"));
    }
    rocksdb::SequenceNumber min = logfile->StartSequence();
    builder.add("tickMin", VPackValue(std::to_string(min)));
    rocksdb::SequenceNumber max;
    if (std::next(lfile) != walFiles.end()) {
      max = (*std::next(lfile))->StartSequence();
    } else {
      max = tdb->GetLatestSequenceNumber();
    }
    builder.add("tickMax", VPackValue(std::to_string(max)));
    builder.close();
  }
  builder.close();
  return Result{};
}

Result RocksDBEngine::firstTick(uint64_t& tick) {
  Result res{};
  rocksdb::TransactionDB* tdb = rocksutils::globalRocksDB();
  rocksdb::VectorLogPtr walFiles;
  rocksdb::Status s = tdb->GetSortedWalFiles(walFiles);

  if (!s.ok()) {
    res = rocksutils::convertStatus(s);
    return res;
  }
  // read minium possible tick
  if (!walFiles.empty()) {
    tick = walFiles[0]->StartSequence();
  }
  return res;
}

Result RocksDBEngine::lastLogger(TRI_vocbase_t& vocbase,
                                 std::shared_ptr<transaction::Context> transactionContext,
                                 uint64_t tickStart, uint64_t tickEnd,
                                 std::shared_ptr<VPackBuilder>& builderSPtr) {
  bool includeSystem = true;
  size_t chunkSize = 32 * 1024 * 1024;  // TODO: determine good default value?

  // construct vocbase with proper handler
  auto builder = std::make_unique<VPackBuilder>(transactionContext->getVPackOptions());

  builder->openArray();

  RocksDBReplicationResult rep = rocksutils::tailWal(&vocbase, tickStart, tickEnd, chunkSize,
                                                     includeSystem, 0, *builder);

  builder->close();
  builderSPtr = std::move(builder);

  return std::move(rep).result();
}

WalAccess const* RocksDBEngine::walAccess() const {
  TRI_ASSERT(_walAccess);
  return _walAccess.get();
}

/// @brief get compression supported by RocksDB
std::string RocksDBEngine::getCompressionSupport() const {
  std::string result;

  for (auto const& type : rocksdb::GetSupportedCompressions()) {
    std::string out;
    rocksdb::GetStringFromCompressionType(&out, type);

    if (out.empty()) {
      continue;
    }
    if (!result.empty()) {
      result.append(", ");
    }
    result.append(out);
  }
  return result;
}

// management methods for synchronizing with external persistent stores
TRI_voc_tick_t RocksDBEngine::currentTick() const {
  return static_cast<TRI_voc_tick_t>(_db->GetLatestSequenceNumber());
}

TRI_voc_tick_t RocksDBEngine::releasedTick() const {
  READ_LOCKER(lock, _walFileLock);
  return _releasedTick;
}

void RocksDBEngine::releaseTick(TRI_voc_tick_t tick) {
  WRITE_LOCKER(lock, _walFileLock);
  if (tick > _releasedTick) {
    _releasedTick = tick;
  }
}

}  // namespace arangodb

// -----------------------------------------------------------------------------
// --SECTION--                                                       END-OF-FILE
// -----------------------------------------------------------------------------