[
https://issues.apache.org/jira/browse/COUCHDB-1817?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13677314#comment-13677314
]
Eli Stevens commented on COUCHDB-1817:
--------------------------------------
45MB documents were running couchjs out of stack space. Removing the large
documents caused the DB to be able to serve views again. See updated
description.
> Exceeding the couchjs stack size does not have a clear error message
> --------------------------------------------------------------------
>
> Key: COUCHDB-1817
> URL: https://issues.apache.org/jira/browse/COUCHDB-1817
> Project: CouchDB
> Issue Type: Bug
> Components: JavaScript View Server
> Reporter: Eli Stevens
> Priority: Minor
> Attachments: couchdb__couchdb_files.png,
> couchdb__httpd_status_codes.png, couchdb_mem.png, loadavg.png, memory2.png
>
>
> Updated description:
> When a large document is uploaded and the couchjs process runs out of stack,
> the error message that gets produced does not make it easy to debug what is
> going wrong.
> In our case we had some data that should have been an attachment but was
> ending up in the document proper, which caused us to have 45MB documents,
> which triggered the issue. Having an error message that says which document
> ran the view server out of space, etc. would be nice.
> ---
> Original description:
> We have started seeing errors crop up in our application that we have not
> seen before, and we're at a loss for how to start debugging it.
> [~dch] Said that we might look into system resource limits, so we started
> collecting all of the output from _stats into RRD (along with memory, load,
> etc. that we were already collecting), but nothing is jumping out at us as
> obviously problematic.
> We can semi-reliably reproduce the problem, but it's far from a minimal test
> case (basically, we load up several large chunks of data, and then halfway
> through the processing run, we get the error). The error doesn't seem to
> happen if we load up each chunk by itself.
> The DB in question has about 100 docs in it, none particularly large (nothing
> over a couple KB would be my guess), with a couple hundred MB in attachments.
> 10ish design docs, coffeescript. In general, there isn't anything that
> seems obviously resource intensive.
> We have seen this issue on 1.2.0, 1.2.1, and we're working on getting a
> machine with 1.3.0 set up (the PPA we'd been using hasn't been updated yet).
> Ubuntu 12.04, spinning disk, etc. The system is under load when it happens,
> but the load isn't more than 1.5x the number of cores. I don't have disk IO
> numbers at hand, but I'd be surprised if that was being strained.
> Error as it appears in couch.log:
> https://gist.github.com/wickedgrey/e7fd3fc14b6d43e95564
> The design doc in question:
> https://gist.github.com/wickedgrey/db41b0c3c75a590e2109
> An example document: https://gist.github.com/wickedgrey/a8422aab261ddd2ce4fe
> We have some preliminary evidence that the problem persists after the system
> goes quiet, but we're not certain.
> Either CouchDB isn't handling things correctly, in which case this bug is
> "prz fix" or we're doing something wrong (hitting a resource limit, or
> something), in which case this bug is "prz make the error message more
> informative".
> Thanks!
--
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
