couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Alex Markham (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (COUCHDB-1505) Error on cancelling replication - possbily related to hanging replications
Date Thu, 18 Oct 2012 09:20:03 GMT

     [ https://issues.apache.org/jira/browse/COUCHDB-1505?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Alex Markham updated COUCHDB-1505:
----------------------------------

    Attachment: couchcrash171012redact.log

Attaching a new log - couchcrash171012redact.log
We had a couch die for 5 seconds yesterday, where it seems heart restarted it.
The stack trace before the crash looks almost identical, except this time it took out the
whole server rather than just print errors.
                
> Error on cancelling replication - possbily related to hanging replications
> --------------------------------------------------------------------------
>
>                 Key: COUCHDB-1505
>                 URL: https://issues.apache.org/jira/browse/COUCHDB-1505
>             Project: CouchDB
>          Issue Type: Bug
>          Components: Replication
>    Affects Versions: 1.2
>         Environment: CentOS 5.6 x64. WAN replication (between datacentres). Cronjob controlled
replication curls every 5 mins. Using pull replication with a filter.
>            Reporter: Alex Markham
>              Labels: cancel, hang, replication
>         Attachments: couchcrash171012redact.log, couchjs.txt, replicationcancelerror1.log
>
>
> We run a cronjob to cancel replication, and then start it again every 5 minutes. Occasionally
when cancelling replication jobs, a stack trace appears in the couchdb log (attached)
> Other observations : perhaps unrelated, but over time we slowly start to gather "zombie"
couchjs processes. After a month or so (different for each server) we start to get up to near
our os_process_limit of 200 and we restart couchdb. "zombie" is speculation here, but there
seems to be no need for the hundred+ couchjs processes when just replicating 10 databases
and occasional indexing, after restart it drops right back down. The started time of those
processes are also weeks old. This may be normal, not sure.
> Why do we cancel replication and restart it? We found that if we don't do this then WAN
replications can hang, where curling /_replicate would say that the continuous replication
is already running, but that the replications were not updating, and the document counts in
the databases would diverge. Immediately after re-enabling the "cancel":true /_replicate beforehand,
these stack traces re-appeared and the replication caught up.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message