[
https://issues.apache.org/jira/browse/ACCUMULO-4191?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15243240#comment-15243240
]
ASF GitHub Bot commented on ACCUMULO-4191:
------------------------------------------
Github user ShawnWalker commented on the pull request:
https://github.com/apache/accumulo/pull/94#issuecomment-210550520
@joshelser
> I was looking for some understanding that you ... had a plausible idea of
why this was sometimes happening. Sounds fine to me.
It seemed obvious to me that not propagating the tracing was an error.
Before you asked, I had just decided it was "some random race condition".
Coming up with a plausible explanation made me read what was going on much more
closely.
If this really is broken for everyone on master, there's a much more
plausible explanation why it might have occasionally failed before: the
`TimerTask` created at `TabletServerBatchWriter` line 210 can cause mutations
to be sent without propagating the tracing information. Particularly the fact
that it has an initial delay of 0 would make for a timing sensitive race
condition: either {{close()}} arranges to send the mutations (traceable) or
the timer arranges to send the mutations (untraceable).
Fixing the timer task to propagate tracing would not be quite so trivial a
task as the patch proposed in this pull request.
> Tracing on client can sometimes lose "sendMutations" events.
> ------------------------------------------------------------
>
> Key: ACCUMULO-4191
> URL: https://issues.apache.org/jira/browse/ACCUMULO-4191
> Project: Accumulo
> Issue Type: Bug
> Reporter: Shawn Walker
> Assignee: Shawn Walker
> Priority: Minor
>
> In {{TabletServerBatchWriter.MutationWriter.queueMutations(...)}} (line 736),
> HTrace {{Span}}'s are not properly propagated across thread boundaries.
> Consequently, tracing doesn't propagate into
> {{TabletServerBatchWriter.SendTask.send(...)}}, and so the {{sendMutation}}
> event can fail to be logged.
> This can cause {{ShellServerIT.trace(...)}} to fail.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
