ceph/src/rocksdb/utilities/transactions/transaction_util.cc

   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 #ifndef ROCKSDB_LITE
   7
   8 #ifndef __STDC_FORMAT_MACROS
   9 #define __STDC_FORMAT_MACROS
  10 #endif
  11
  12 #include "utilities/transactions/transaction_util.h"
  13
  14 #include <inttypes.h>
  15 #include <string>
  16 #include <vector>
  17
  18 #include "db/db_impl.h"
  19 #include "rocksdb/status.h"
  20 #include "rocksdb/utilities/write_batch_with_index.h"
  21 #include "util/string_util.h"
  22
  23 namespace rocksdb {
  24
  25 Status TransactionUtil::CheckKeyForConflicts(
  26     DBImpl* db_impl, ColumnFamilyHandle* column_family, const std::string& key,
  27     SequenceNumber snap_seq, bool cache_only, ReadCallback* snap_checker) {
  28   Status result;
  29
  30   auto cfh = reinterpret_cast<ColumnFamilyHandleImpl*>(column_family);
  31   auto cfd = cfh->cfd();
  32   SuperVersion* sv = db_impl->GetAndRefSuperVersion(cfd);
  33
  34   if (sv == nullptr) {
  35     result = Status::InvalidArgument("Could not access column family " +
  36                                      cfh->GetName());
  37   }
  38
  39   if (result.ok()) {
  40     SequenceNumber earliest_seq =
  41         db_impl->GetEarliestMemTableSequenceNumber(sv, true);
  42
  43     result = CheckKey(db_impl, sv, earliest_seq, snap_seq, key, cache_only,
  44                       snap_checker);
  45
  46     db_impl->ReturnAndCleanupSuperVersion(cfd, sv);
  47   }
  48
  49   return result;
  50 }
  51
  52 Status TransactionUtil::CheckKey(DBImpl* db_impl, SuperVersion* sv,
  53                                  SequenceNumber earliest_seq,
  54                                  SequenceNumber snap_seq,
  55                                  const std::string& key, bool cache_only,
  56                                  ReadCallback* snap_checker) {
  57   Status result;
  58   bool need_to_read_sst = false;
  59
  60   // Since it would be too slow to check the SST files, we will only use
  61   // the memtables to check whether there have been any recent writes
  62   // to this key after it was accessed in this transaction.  But if the
  63   // Memtables do not contain a long enough history, we must fail the
  64   // transaction.
  65   if (earliest_seq == kMaxSequenceNumber) {
  66     // The age of this memtable is unknown.  Cannot rely on it to check
  67     // for recent writes.  This error shouldn't happen often in practice as
  68     // the Memtable should have a valid earliest sequence number except in some
  69     // corner cases (such as error cases during recovery).
  70     need_to_read_sst = true;
  71
  72     if (cache_only) {
  73       result = Status::TryAgain(
  74           "Transaction ould not check for conflicts as the MemTable does not "
  75           "countain a long enough history to check write at SequenceNumber: ",
  76           ToString(snap_seq));
  77     }
  78   } else if (snap_seq < earliest_seq) {
  79     need_to_read_sst = true;
  80
  81     if (cache_only) {
  82       // The age of this memtable is too new to use to check for recent
  83       // writes.
  84       char msg[300];
  85       snprintf(msg, sizeof(msg),
  86                "Transaction could not check for conflicts for operation at "
  87                "SequenceNumber %" PRIu64
  88                " as the MemTable only contains changes newer than "
  89                "SequenceNumber %" PRIu64
  90                ".  Increasing the value of the "
  91                "max_write_buffer_number_to_maintain option could reduce the "
  92                "frequency "
  93                "of this error.",
  94                snap_seq, earliest_seq);
  95       result = Status::TryAgain(msg);
  96     }
  97   }
  98
  99   if (result.ok()) {
 100     SequenceNumber seq = kMaxSequenceNumber;
 101     bool found_record_for_key = false;
 102
 103     Status s = db_impl->GetLatestSequenceForKey(sv, key, !need_to_read_sst,
 104                                                 &seq, &found_record_for_key);
 105
 106     if (!(s.ok() || s.IsNotFound() || s.IsMergeInProgress())) {
 107       result = s;
 108     } else if (found_record_for_key) {
 109       bool write_conflict = snap_checker == nullptr
 110                                 ? snap_seq < seq
 111                                 : !snap_checker->IsVisible(seq);
 112       if (write_conflict) {
 113         result = Status::Busy();
 114       }
 115     }
 116   }
 117
 118   return result;
 119 }
 120
 121 Status TransactionUtil::CheckKeysForConflicts(DBImpl* db_impl,
 122                                               const TransactionKeyMap& key_map,
 123                                               bool cache_only) {
 124   Status result;
 125
 126   for (auto& key_map_iter : key_map) {
 127     uint32_t cf_id = key_map_iter.first;
 128     const auto& keys = key_map_iter.second;
 129
 130     SuperVersion* sv = db_impl->GetAndRefSuperVersion(cf_id);
 131     if (sv == nullptr) {
 132       result = Status::InvalidArgument("Could not access column family " +
 133                                        ToString(cf_id));
 134       break;
 135     }
 136
 137     SequenceNumber earliest_seq =
 138         db_impl->GetEarliestMemTableSequenceNumber(sv, true);
 139
 140     // For each of the keys in this transaction, check to see if someone has
 141     // written to this key since the start of the transaction.
 142     for (const auto& key_iter : keys) {
 143       const auto& key = key_iter.first;
 144       const SequenceNumber key_seq = key_iter.second.seq;
 145
 146       result = CheckKey(db_impl, sv, earliest_seq, key_seq, key, cache_only);
 147
 148       if (!result.ok()) {
 149         break;
 150       }
 151     }
 152
 153     db_impl->ReturnAndCleanupSuperVersion(cf_id, sv);
 154
 155     if (!result.ok()) {
 156       break;
 157     }
 158   }
 159
 160   return result;
 161 }
 162
 163
 164 }  // namespace rocksdb
 165
 166 #endif  // ROCKSDB_LITE