incubator-couchdb-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Jason Smith <...@iriscouch.com>
Subject Re: couchDB with databases per user: continuous replications stop working, without errors
Date Thu, 05 Jan 2012 16:13:07 GMT
It's nothing that isn't documented in the Wiki. The hard part is
resisting the temptation to just throw every idea at the problem to
minimize downtime. Even if a service is down (in this case, not every
doc in _replicator was triggering a replication), it usually pays to
deliberately identify the real problem, so you can be sure it's fixed.

Couch is somewhat opaque when you need to troubleshoot (especially ASF
CouchDB which doesn't easily let you get an Erlang shell). But it was
just a matter of noticing tracebacks in the logs.

Limits are there for a reason. Hitting them is not necessarily bad.
Some limits are okay to hard-code to something higher; but I usually
like to make limits (files, ports, quota) a function of something,
such as how many databases you have, or what kind of account you have,
or whatever. Also, when you compute that value, you can trigger an
alert if its value is way too high. (You effectively fail-early,
alerting the sysadmin before there is a crisis.) You might have
guessed that this particular issue did not unfold quite so nicely.

On Thu, Jan 5, 2012 at 10:58 PM, Robert Newson <rnewson@apache.org> wrote:
> Hopefully Jason can complete this thread with details of his fix (once
> the sun rolls back around to where he is).
>
> B.
>
> On 5 January 2012 15:37, till <till@php.net> wrote:
>> On Thu, Jan 5, 2012 at 10:18 AM, Gregor Martynus <gregor@martynus.net>wrote:
>>
>>> I experience massive reliability issues with continuous replications in my
>>> app. Each user has an own profile db with a continuous replication to a
>>> central database. Numbers: ~1800 dbs & ~2600 continuous replications (using
>>> _replicator db), ~200 did not get started.
>>>
>>> My hope is that others faced these issues before and found ways to fix or
>>> work around them. So if you are a developer or know developers that have a
>>> similar couchDB setup, would you mind to get me in touch? Thanks for your
>>> help!
>>>
>>> Tech info:
>>> couchDB 1.1.1 (iriscouch)
>>> max_dbs_open is set to 5000
>>> no errors show up in _log
>>> _replicate does not accept POST requests anymore
>>> new replications from _replicator docs do not get picked up
>>>
>>
>> Is your loglevel at debug? It should def. report something – maybe do a
>> simple cURL request and paste that along with a log entry somewhere (
>> friendpaste.com).
>>
>> You might also want to try stock Apache CouchDB – not sure how IrishCouch's
>> version differs.
>>
>> Till



-- 
Iris Couch

Mime
View raw message