[HBASE-17033] LogRoller makes a lot of allocations unnecessarily

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 1.4.0, 2.0.0
• Hadoop Flags: Reviewed
• Assignee: Enis Soztutar
• Reporter: Enis Soztutar
• Created: 05/Nov/16 01:38
• Updated: 08/Nov/16 04:57
• Resolved: 07/Nov/16 22:40

Description

I was looking at the other allocations for HBASE-17017. Seems that log roller thread allocates 200MB for ~7% of the TLAB space. This is a lot of allocations.

I think the reason is this:

while (true) {
        if (this.safePointAttainedLatch.await(1, TimeUnit.NANOSECONDS)) {
          break;
        }
        if (syncFuture.isThrowable()) {
          throw new FailedSyncBeforeLogCloseException(syncFuture.getThrowable());
        }
      }

This busy wait is causing a lot allocations because the thread is added to the waiting list.