[HBASE-7756] Strange code in ServerCallable#shouldRetry

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.94.4, 0.95.2
• Fix Version/s: 0.95.0
• Component/s: Client
• Hadoop Flags: Reviewed
• Assignee: Nicolas Liochon
• Reporter: Nicolas Liochon
• Created: 04/Feb/13 17:22
• Updated: 23/Sep/13 18:30
• Resolved: 05/Feb/13 12:49

Description

This code is strange.
If you configure a different value for HBASE_CLIENT_OPERATION_TIMEOUT, you can have only one socket timeout (whatever your setting for the socket timeout). If not, the first time we will retry (again, whatever your setting and the real timeout).

  public void shouldRetry(Throwable throwable) throws IOException {
    if (this.callTimeout != HConstants.DEFAULT_HBASE_CLIENT_OPERATION_TIMEOUT)
      if (throwable instanceof SocketTimeoutException
          || (this.endTime - this.startTime > this.callTimeout)) {
        throw (SocketTimeoutException) new SocketTimeoutException(
            "Call to access row '" + Bytes.toString(row) + "' on table '"
                + Bytes.toString(tableName)
                + "' failed on socket timeout exception: " + throwable + " this.callTimeout="+this.callTimeout + " time= "+(this.endTime - this.startTime))
            .initCause(throwable);
      } else {
        this.callTimeout = ((int) (this.endTime - this.startTime));
      }
  }

I don't get all the implications yet, but for sure it's too smart to be good, and, at least, breaks my tests for HBASE-7590

Issue Links

• is required by: HBASE-7590 Add a costless notifications mechanism from master to regionservers & clients
• is required by: HBASE-5843 Improve HBase MTTR - Mean Time To Recover