ceph/src/rocksdb/utilities/transactions/write_prepared_txn.h

   1 // Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
   2 //  This source code is licensed under both the GPLv2 (found in the
   3 //  COPYING file in the root directory) and Apache 2.0 License
   4 //  (found in the LICENSE.Apache file in the root directory).
   5
   6 #pragma once
   7
   8 #ifndef ROCKSDB_LITE
   9
  10 #include <algorithm>
  11 #include <atomic>
  12 #include <mutex>
  13 #include <stack>
  14 #include <string>
  15 #include <unordered_map>
  16 #include <vector>
  17
  18 #include "db/write_callback.h"
  19 #include "rocksdb/db.h"
  20 #include "rocksdb/slice.h"
  21 #include "rocksdb/snapshot.h"
  22 #include "rocksdb/status.h"
  23 #include "rocksdb/types.h"
  24 #include "rocksdb/utilities/transaction.h"
  25 #include "rocksdb/utilities/transaction_db.h"
  26 #include "rocksdb/utilities/write_batch_with_index.h"
  27 #include "util/autovector.h"
  28 #include "utilities/transactions/pessimistic_transaction.h"
  29 #include "utilities/transactions/pessimistic_transaction_db.h"
  30 #include "utilities/transactions/transaction_base.h"
  31 #include "utilities/transactions/transaction_util.h"
  32
  33 namespace rocksdb {
  34
  35 class WritePreparedTxnDB;
  36
  37 // This impl could write to DB also uncommitted data and then later tell apart
  38 // committed data from uncommitted data. Uncommitted data could be after the
  39 // Prepare phase in 2PC (WritePreparedTxn) or before that
  40 // (WriteUnpreparedTxnImpl).
  41 class WritePreparedTxn : public PessimisticTransaction {
  42  public:
  43   WritePreparedTxn(WritePreparedTxnDB* db, const WriteOptions& write_options,
  44                    const TransactionOptions& txn_options);
  45
  46   virtual ~WritePreparedTxn() {}
  47
  48   // To make WAL commit markers visible, the snapshot will be based on the last
  49   // seq in the WAL that is also published, LastPublishedSequence, as opposed to
  50   // the last seq in the memtable.
  51   using Transaction::Get;
  52   virtual Status Get(const ReadOptions& options,
  53                      ColumnFamilyHandle* column_family, const Slice& key,
  54                      PinnableSlice* value) override;
  55
  56   // Note: The behavior is undefined in presence of interleaved writes to the
  57   // same transaction.
  58   // To make WAL commit markers visible, the snapshot will be
  59   // based on the last seq in the WAL that is also published,
  60   // LastPublishedSequence, as opposed to the last seq in the memtable.
  61   using Transaction::GetIterator;
  62   virtual Iterator* GetIterator(const ReadOptions& options) override;
  63   virtual Iterator* GetIterator(const ReadOptions& options,
  64                                 ColumnFamilyHandle* column_family) override;
  65
  66   virtual void SetSnapshot() override;
  67
  68  protected:
  69   void Initialize(const TransactionOptions& txn_options) override;
  70   // Override the protected SetId to make it visible to the friend class
  71   // WritePreparedTxnDB
  72   inline void SetId(uint64_t id) override { Transaction::SetId(id); }
  73
  74  private:
  75   friend class WritePreparedTransactionTest_BasicRecoveryTest_Test;
  76   friend class WritePreparedTxnDB;
  77   friend class WriteUnpreparedTxnDB;
  78   friend class WriteUnpreparedTxn;
  79
  80   Status PrepareInternal() override;
  81
  82   Status CommitWithoutPrepareInternal() override;
  83
  84   Status CommitBatchInternal(WriteBatch* batch, size_t batch_cnt) override;
  85
  86   // Since the data is already written to memtables at the Prepare phase, the
  87   // commit entails writing only a commit marker in the WAL. The sequence number
  88   // of the commit marker is then the commit timestamp of the transaction. To
  89   // make WAL commit markers visible, the snapshot will be based on the last seq
  90   // in the WAL that is also published, LastPublishedSequence, as opposed to the
  91   // last seq in the memtable.
  92   Status CommitInternal() override;
  93
  94   Status RollbackInternal() override;
  95
  96   virtual Status ValidateSnapshot(ColumnFamilyHandle* column_family,
  97                                   const Slice& key,
  98                                   SequenceNumber* tracked_at_seq) override;
  99
 100   virtual Status RebuildFromWriteBatch(WriteBatch* src_batch) override;
 101
 102   // No copying allowed
 103   WritePreparedTxn(const WritePreparedTxn&);
 104   void operator=(const WritePreparedTxn&);
 105
 106   WritePreparedTxnDB* wpt_db_;
 107   // Number of sub-batches in prepare
 108   size_t prepare_batch_cnt_ = 0;
 109 };
 110
 111 }  // namespace rocksdb
 112
 113 #endif  // ROCKSDB_LITE