[ACCUMULO-2208] Provide better error handling when unable to cleanly close log

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.5.0
• Fix Version/s: 1.5.1, 1.6.0
• Component/s: tserver
• Assignee: John Vines
• Reporter: John Vines
• Created: 16/Jan/14 17:14
• Updated: 23/Apr/14 15:22
• Resolved: 22/Apr/14 21:40

Description

I saw this once over 36 tserver/days, caused by an InvalidArgumentException, so odd occurrence. It was doing CI but before agitation started so it was a non issue, but it could be. The code around this is

for (DfsLogger logger : loggers) {
        try {
          logger.close();
        } catch (DfsLogger.LogClosedException ex) {
          // ignore
        } catch (Throwable ex) {
          log.error("Unable to cleanly close log " + logger.getFileName() + ": " + ex);
        }
      }
      loggers.clear();
      logSizeEstimate.set(0);

so it seems like it blindly powered forward. However, because of the frequency of syncs and such, even if there was agitation the odds of losing data are very minute, but I do believe their still there.

But because the log message just relies on error.toString, I didn't get the violating stack trace. At the bare minimum, the message should be updated to log the full stack trace.