Re: Couchdb trunk purge_docs timeout

[Jan Lehnardt <ja...@apache.org>]

Hi Mike,

we did a fix in this area recently that affected purging of docs in conflict:

  http://svn.apache.org/viewvc?rev=1086241&view=rev

A couple of reviewers deemed the patch safe, but this is a seldom exercised
part of the code, so we may have introduced your issue.

Can you provide us with a reproducing script that maybe doesn't depend on
56m docs? :)

Also, can you paste the full error stack trace?

A few more questions:

 - Is replication involved here?
 - Do you have more I/O than before on the system?

CCing dev@.

Cheers
Jan
-- 

On 14 Apr 2011, at 17:14, Mike Leddy wrote:

> Hi,
> 
> I have a couch node current using version 1.2.0abaa0e30-git. I decided
> to try a database maintenance task that I formerly used to use on
> couchdb 1.0.2 to purge documents in batches of 500 on a database that
> contains some 56 million documents.
> 
> From what I can gather from the logs the call to purge_docs is timing
> out after 5 seconds and terminating.
> 
> [Wed, 13 Apr 2011 20:25:57 GMT] [info] [<0.5192.19>] 172.17.17.3 - - GET /iris/_design/tidy/_view/conflicts?limit=0 200
> [Wed, 13 Apr 2011 20:26:02 GMT] [error] [<0.5194.19>] Uncaught error in HTTP request: {exit,
>                                                       {timeout,
>                                                        {gen_server,call,
>                                                         [<0.150.0>,
>                                                          {purge_docs,
>                                                           [{<<"1294099271F6261">>,
>                                                             [{1,
>                                                               <<181,64,95,
>                                                                 54,247,104,
>                                                                 56,34,109,
>                                                                 228,7,108,
>                                                                 250,72,57,
>                                                                 190>>}]},
>                                                            {<<"1294099281F7327">>,
>                                                             [{1,
>                                                               <<80,246,15,
>                                                                 155,182,61,
>                                                                 43,238,207,
>                                                                 43,159,136,
>                                                                 178,134,
>                                                                 137,214>>}]},
> ... removed for brevity ....
> [Wed, 13 Apr 2011 20:26:02 GMT] [info] [<0.5194.19>] Stacktrace: [{io_lib_pretty,cind_tag_tuple,7},
>                                  {io_lib_pretty,while_fail,3},
>                                  {io_lib_pretty,print,6},
>                                  {io_lib_format,build,3},
>                                  {io_lib_format,build,3},
>                                  {io_lib_format,build,3},
>                                  {io_lib_format,build,3},
>                                  {io_lib_format,build,3}]
> [Wed, 13 Apr 2011 20:26:02 GMT] [info] [<0.5194.19>] 172.17.17.3 - - POST /iris/_purge 500
> 
> I am pretty sure that this was not the case with 1.0.2. Does anyone 
> have any insight regarding what is the root cause of the problem ?
> 
> Meanwhile I'm digging through the code looking for clues....
> 
> Thanks,
> 
> Mike
>