[COUCHDB-1429] Server stops responding after heavy load against _list function

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Won't Fix
• Affects Version/s: 1.1.1
• Component/s: View Server Support

Environment

Ubuntu 11.10 on AWS EC2 t1.micro instance, ~512MB memory and no swap

• Assignee: Unassigned
• Reporter: Nathan Vander Wilt
• Created: 05/Mar/12 20:34
• Updated: 02/Sep/18 06:36
• Resolved: 02/Sep/18 06:36

Description

Under heavy load, CouchDB crashes and stops responding to all incoming requests.

To reproduce this, basically:
1. build-couchdb (1.1.1) on an EC2 t1.micro running Ubuntu
2. Add the '42' file at the path Blitz.io is looking for (not sure the easiest way to do this natively, I inserted a rule for it at the nginx level)
3. Run their default rush on a _list function (mine happened to do a fair amount of work,
doing a Markdown conversion and Mustache templating)
4. Around about the 40 concurrent user mark in my case, CouchDB dies a terrible horrible death with a
bunch of zombie couchjs process.

In this log https://gist.github.com/a059c4db5bce19f1df7f (warning: large!) you'll see the heavy requests being handled but suddenly crashing. The "restart" seen at the end was due to manual intervention from the shell, CouchDB did not gracefully handle the issue.

I later tried turning on some swap space in case memory was an issue, but didn't seem to have an appreciable effect.

May be related to the discussion here: http://mail-archives.apache.org/mod_mbox/couchdb-dev/201203.mbox/[e-mail]%3e