[
https://issues.apache.org/jira/browse/HBASE-5706?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13245770#comment-13245770
]
Andrew Purtell commented on HBASE-5706:
---------------------------------------
Interesting, this came up today here too. I've been playing around with a
slightly modified version of this patch. Under high load the FS latency stat
buffers can fill up before the metrics thread drains them, doesn't seem an
exceptional condition. Removing logDroppedLatencyStat() and logging around it
gets rid of a number of branches and a CAS from a hot code path, so that's what
we did.
> "Dropping fs latency stats since buffer is full" spam
> -----------------------------------------------------
>
> Key: HBASE-5706
> URL: https://issues.apache.org/jira/browse/HBASE-5706
> Project: HBase
> Issue Type: Improvement
> Reporter: Jean-Daniel Cryans
> Assignee: Shaneal Manek
> Priority: Minor
> Fix For: 0.94.0, 0.96.0
>
>
> I see tons of this while running tests (note that it's a WARN):
> {noformat}
> 2012-04-03 18:54:47,172 WARN org.apache.hadoop.hbase.io.hfile.HFile: Dropping
> fs latency stats since buffer is full
> {noformat}
> While the code says this:
> {noformat}
> // we don't want to fill up the logs with this message, so only log it
> // once every 30 seconds at most
> // I also want to avoid locks on the 'critical path' (the common case will
> be
> // uncontended) - hence the CAS
> private static void logDroppedLatencyStat() {
> {noformat}
> It doesn't seem like this message is actionnable and even though it's printed
> "only" every 30 seconds it's still very spammy.
> We should get rid of it or make it more useful (I don't know which).
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
