incubator-couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Nicolas Maupu (JIRA)" <j...@apache.org>
Subject [jira] Commented: (COUCHDB-1052) CouchDB crashes by Erlang Heartbeat Runtime System
Date Wed, 02 Feb 2011 09:27:29 GMT

    [ https://issues.apache.org/jira/browse/COUCHDB-1052?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12989570#comment-12989570
] 

Nicolas Maupu commented on COUCHDB-1052:
----------------------------------------

What I mean by graceful restart is just to be sure CouchDB restart after a erlang "heart attack"
instead of stop
I suspect system clock to cause this erlang event because I don't have any NTP server (and
can't use it at the moment).

I added patch I use in attached files.
I changed HEART_COMMAND and start / stop functions. I also changed signal sent to PID to stop
(1 to 15) because on my system, this signal did not stop process as expected.

> CouchDB crashes by Erlang Heartbeat Runtime System
> --------------------------------------------------
>
>                 Key: COUCHDB-1052
>                 URL: https://issues.apache.org/jira/browse/COUCHDB-1052
>             Project: CouchDB
>          Issue Type: Improvement
>          Components: Infrastructure
>    Affects Versions: 0.10.1, 1.0.1
>         Environment: Solaris 10
>            Reporter: Nicolas Maupu
>         Attachments: couchdb-bin-1.0.1.diff
>
>   Original Estimate: 2h
>  Remaining Estimate: 2h
>
> Hello,
> We have several couchdb in our production environment (starting from 0.10.1 and upgraded
to 1.0.1). Some instances began to crash several times a day.
> After research (I am not aware of all erlang tips and tricks), I found that the problem
was caused by erlang heartbeat runtime system (http://www.erlang.org/doc/man/heart.html).
> For some reasons (it seems to be a problem with system clock), this heartbeat process
sends a signal to main erlang process to trigger HEARTBEAT_COMMAND. 
> The command configured by couchdb startup script is to call 'bin/couchdb -k' which kill
server ...
> I don't know if this is wanted or not but is Erlang heartbeat system provided to restart
a service if this one crash ?
> I patched mine to restart process gracefully but I would like to know if this is a feature
or a bug : next time I upgrade, I will have to patch again.
> My patch is a little bit specific to my architecture so I did not provide it. Ask me
if you are interested.
> nm.

-- 
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Mime
View raw message