[HBASE-10069] Potential duplicate calls to log#appendNoSync() in HRegion#doMiniBatchMutation()

• Type: Bug
• Status: Closed
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 0.98.0, 0.99.0
• Assignee: Sergey Shelukhin
• Reporter: Ted Yu
• Created: 02/Dec/13 23:09
• Updated: 21/Feb/15 23:31
• Resolved: 11/Dec/13 19:25

Description

In HRegion#doMiniBatchMutation():

        if (nonceGroup != currentNonceGroup || nonce != currentNonce) {
          if (walEdit.size() > 0) {
            assert isInReplay;
            txid = this.log.appendNoSync(this.getRegionInfo(), htableDescriptor.getTableName(),
                  walEdit, m.getClusterIds(), now, htableDescriptor, this.sequenceId, true,
                  currentNonceGroup, currentNonce);
            hasWalAppends = true;
          }
          currentNonceGroup = nonceGroup;
          currentNonce = nonce;
        }

        // Add WAL edits by CP
        WALEdit fromCP = batchOp.walEditsFromCoprocessors[i];
        if (fromCP != null) {
          for (KeyValue kv : fromCP.getKeyValues()) {
            walEdit.add(kv);
          }
        }
...
      Mutation mutation = batchOp.getMutation(firstIndex);
      if (walEdit.size() > 0) {
        txid = this.log.appendNoSync(this.getRegionInfo(), this.htableDescriptor.getTableName(),
              walEdit, mutation.getClusterIds(), now, this.htableDescriptor, this.sequenceId,
              true, currentNonceGroup, currentNonce);
        hasWalAppends = true;
      }

If fromCP is null, there may not be new edits added to walEdit.
But log#appendNoSync() would be called one more time at line 2368.

Issue Links

• is broken by: HBASE-3787 Increment is non-idempotent but client retries RPC