wiredtiger_kv_engine.h

Go to the documentation of this file.

// wiredtiger_kv_engine.h

#pragma once

#include <list>
#include <memory>
#include <string>

#include <wiredtiger.h>

#include "mongo/bson/ordering.h"
#include "mongo/bson/timestamp.h"
#include "mongo/db/storage/kv/kv_engine.h"
#include "mongo/db/storage/wiredtiger/wiredtiger_oplog_manager.h"
#include "mongo/db/storage/wiredtiger/wiredtiger_session_cache.h"
#include "mongo/db/storage/wiredtiger/wiredtiger_util.h"
#include "mongo/stdx/functional.h"
#include "mongo/stdx/mutex.h"
#include "mongo/util/elapsed_tracker.h"

namespace mongo {

class ClockSource;
class JournalListener;
class WiredTigerRecordStore;
class WiredTigerSessionCache;
class WiredTigerSizeStorer;

struct WiredTigerFileVersion {
    enum class StartupVersion { IS_34, IS_36, IS_40, IS_42 };

    StartupVersion _startupVersion;
    bool shouldDowngrade(bool readOnly, bool repairMode, bool hasRecoveryTimestamp);
    std::string getDowngradeString();
};

class WiredTigerKVEngine final : public KVEngine {
public:
    static const int kDefaultJournalDelayMillis;
    WiredTigerKVEngine(const std::string& canonicalName,
                       const std::string& path,
                       ClockSource* cs,
                       const std::string& extraOpenOptions,
                       size_t cacheSizeGB,
                       bool durable,
                       bool ephemeral,
                       bool repair,
                       bool readOnly);

    virtual ~WiredTigerKVEngine();

    void setRecordStoreExtraOptions(const std::string& options);
    void setSortedDataInterfaceExtraOptions(const std::string& options);

    virtual bool supportsDocLocking() const;

    virtual bool supportsDirectoryPerDB() const;

    virtual bool isDurable() const {
        return _durable;
    }

    virtual bool isEphemeral() const {
        return _ephemeral;
    }

    virtual RecoveryUnit* newRecoveryUnit();

    virtual Status createRecordStore(OperationContext* opCtx,
                                     StringData ns,
                                     StringData ident,
                                     const CollectionOptions& options) {
        return createGroupedRecordStore(opCtx, ns, ident, options, KVPrefix::kNotPrefixed);
    }

    virtual std::unique_ptr<RecordStore> getRecordStore(OperationContext* opCtx,
                                                        StringData ns,
                                                        StringData ident,
                                                        const CollectionOptions& options) {
        return getGroupedRecordStore(opCtx, ns, ident, options, KVPrefix::kNotPrefixed);
    }

    virtual Status createSortedDataInterface(OperationContext* opCtx,
                                             StringData ident,
                                             const IndexDescriptor* desc) {
        return createGroupedSortedDataInterface(opCtx, ident, desc, KVPrefix::kNotPrefixed);
    }

    virtual SortedDataInterface* getSortedDataInterface(OperationContext* opCtx,
                                                        StringData ident,
                                                        const IndexDescriptor* desc) {
        return getGroupedSortedDataInterface(opCtx, ident, desc, KVPrefix::kNotPrefixed);
    }

    virtual Status createGroupedRecordStore(OperationContext* opCtx,
                                            StringData ns,
                                            StringData ident,
                                            const CollectionOptions& options,
                                            KVPrefix prefix);

    virtual std::unique_ptr<RecordStore> getGroupedRecordStore(OperationContext* opCtx,
                                                               StringData ns,
                                                               StringData ident,
                                                               const CollectionOptions& options,
                                                               KVPrefix prefix);

    virtual Status createGroupedSortedDataInterface(OperationContext* opCtx,
                                                    StringData ident,
                                                    const IndexDescriptor* desc,
                                                    KVPrefix prefix);

    virtual SortedDataInterface* getGroupedSortedDataInterface(OperationContext* opCtx,
                                                               StringData ident,
                                                               const IndexDescriptor* desc,
                                                               KVPrefix prefix);

    virtual Status dropIdent(OperationContext* opCtx, StringData ident);

    virtual void alterIdentMetadata(OperationContext* opCtx,
                                    StringData ident,
                                    const IndexDescriptor* desc);

    virtual Status okToRename(OperationContext* opCtx,
                              StringData fromNS,
                              StringData toNS,
                              StringData ident,
                              const RecordStore* originalRecordStore) const;

    virtual int flushAllFiles(OperationContext* opCtx, bool sync);

    virtual Status beginBackup(OperationContext* opCtx);

    virtual void endBackup(OperationContext* opCtx);

    virtual int64_t getIdentSize(OperationContext* opCtx, StringData ident);

    virtual Status repairIdent(OperationContext* opCtx, StringData ident);

    virtual bool hasIdent(OperationContext* opCtx, StringData ident) const;

    std::vector<std::string> getAllIdents(OperationContext* opCtx) const;

    virtual void cleanShutdown();

    SnapshotManager* getSnapshotManager() const final {
        return &_sessionCache->snapshotManager();
    }

    void setJournalListener(JournalListener* jl) final;

    virtual void setStableTimestamp(Timestamp stableTimestamp) override;

    virtual void setInitialDataTimestamp(Timestamp initialDataTimestamp) override;

    virtual void setOldestTimestampFromStable() override;

    virtual void setOldestTimestamp(Timestamp newOldestTimestamp) override;

    virtual bool supportsRecoverToStableTimestamp() const override;

    virtual StatusWith<Timestamp> recoverToStableTimestamp(OperationContext* opCtx) override;

    virtual boost::optional<Timestamp> getRecoveryTimestamp() const override;

    virtual boost::optional<Timestamp> getLastStableCheckpointTimestamp() const override;

    virtual Timestamp getAllCommittedTimestamp() const override;

    bool supportsReadConcernSnapshot() const final;

    // wiredtiger specific
    // Calls WT_CONNECTION::reconfigure on the underlying WT_CONNECTION
    // held by this class
    int reconfigure(const char* str);

    WT_CONNECTION* getConnection() {
        return _conn;
    }
    void dropSomeQueuedIdents();
    std::list<WiredTigerCachedCursor> filterCursorsWithQueuedDrops(
        std::list<WiredTigerCachedCursor>* cache);
    bool haveDropsQueued() const;

    void syncSizeInfo(bool sync) const;

    /*
     * An oplog manager is always accessible, but this method will start the background thread to
     * control oplog entry visibility for reads.
     *
     * On mongod, the background thread will be started when the first oplog record store is
     * created, and stopped when the last oplog record store is destroyed, at shutdown time. For
     * unit tests, the background thread may be started and stopped multiple times as tests create
     * and destroy oplog record stores.
     */
    void startOplogManager(OperationContext* opCtx,
                           const std::string& uri,
                           WiredTigerRecordStore* oplogRecordStore);
    void haltOplogManager();

    /*
     * Always returns a non-nil pointer. However, the WiredTigerOplogManager may not have been
     * initialized and its background refreshing thread may not be running.
     *
     * A caller that wants to get the oplog read timestamp, or call
     * `waitForAllEarlierOplogWritesToBeVisible`, is advised to first see if the oplog manager is
     * running with a call to `isRunning`.
     *
     * A caller that simply wants to call `triggerJournalFlush` may do so without concern.
     */
    WiredTigerOplogManager* getOplogManager() const {
        return _oplogManager.get();
    }

    /*
     * This function is called when replication has completed a batch.  In this function, we
     * refresh our oplog visiblity read-at-timestamp value.
     */
    void replicationBatchIsComplete() const override;

    static void setInitRsOplogBackgroundThreadCallback(stdx::function<bool(StringData)> cb);

    static bool initRsOplogBackgroundThread(StringData ns);

    static void appendGlobalStats(BSONObjBuilder& b);

    bool isCacheUnderPressure(OperationContext* opCtx) const override;

    Timestamp getStableTimestamp() const;
    Timestamp getOldestTimestamp() const;

private:
    class WiredTigerJournalFlusher;
    class WiredTigerCheckpointThread;

    Status _salvageIfNeeded(const char* uri);
    void _checkIdentPath(StringData ident);

    bool _hasUri(WT_SESSION* session, const std::string& uri) const;

    std::string _uri(StringData ident) const;

    Timestamp _calculateHistoryLagFromStableTimestamp(Timestamp stableTimestamp);

    void _setOldestTimestamp(Timestamp newOldestTimestamp, bool force);

    WT_CONNECTION* _conn;
    WiredTigerEventHandler _eventHandler;
    std::unique_ptr<WiredTigerSessionCache> _sessionCache;
    ClockSource* const _clockSource;

    // Mutex to protect use of _oplogManagerCount by this instance of KV engine.
    mutable stdx::mutex _oplogManagerMutex;
    std::size_t _oplogManagerCount = 0;
    std::unique_ptr<WiredTigerOplogManager> _oplogManager;

    std::string _canonicalName;
    std::string _path;
    std::string _wtOpenConfig;

    std::unique_ptr<WiredTigerSizeStorer> _sizeStorer;
    std::string _sizeStorerUri;
    mutable ElapsedTracker _sizeStorerSyncTracker;

    bool _durable;
    bool _ephemeral;
    const bool _inRepairMode;
    bool _readOnly;
    std::unique_ptr<WiredTigerJournalFlusher> _journalFlusher;  // Depends on _sizeStorer
    std::unique_ptr<WiredTigerCheckpointThread> _checkpointThread;

    std::string _rsOptions;
    std::string _indexOptions;

    mutable stdx::mutex _dropAllQueuesMutex;
    mutable stdx::mutex _identToDropMutex;
    std::list<std::string> _identToDrop;

    mutable Date_t _previousCheckedDropsQueued;

    std::unique_ptr<WiredTigerSession> _backupSession;
    Timestamp _recoveryTimestamp;
    WiredTigerFileVersion _fileVersion;

    // Ensures accesses to _oldestTimestamp and _stableTimestamp, respectively, are multi-core safe.
    mutable stdx::mutex _oldestTimestampMutex;
    mutable stdx::mutex _stableTimestampMutex;

    // Tracks the stable and oldest timestamps we've set on the storage engine.
    Timestamp _oldestTimestamp;
    Timestamp _stableTimestamp;
};
}