archiva-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sascha Vogt <sascha.v...@gmail.com>
Subject Re: Degrading upload performance during day
Date Tue, 20 May 2014 14:59:41 GMT
Hi,

please find 10 stacks from this morning (upload from Jenkins to Archiva
took 1 min) and 10 stacks from just now (upload from Jenkins to Archiva
took 5:42)

https://dl.dropboxusercontent.com/u/44225754/stacks.zip

Hope you find something valuable

Greetings
-Sascha-

Am 20.05.2014 09:15, schrieb Sascha Vogt:
> Hi,
> 
> sure, I'll get a few this evening. And to note: Not the whole system is
> slow, just uploads via Maven from our Jenkins.
> 
> Searches in the UI are "normal", also builds / downloads don't seem to
> be particularly slow.
> 
> Maybe Jackrabbit does some cleanup jobs during the night / once a day? A
> colleague which works on a project also using Jackrabbit as datastore
> said something that there are very small index files which eventually
> get merged to some bigger index. Could that be it? I think our SAN
> behaves particularly bad when a lot of small files are involved.
> 
> Anyway, I just counted the files in the index directory and took the
> first 10 thread dumps. Will do the same later
> 
> Greetings
> -Sascha-
> 
> Am 20.05.2014 01:52, schrieb Olivier Lamy:
>> Hi,
>> Really hard to know exactly what happen.
>> Do you have time to get a jstack while the system is slow?
>>
>> Olivier
>>
>> On 15 May 2014 18:24, Sascha Vogt <sascha.vogt@gmail.com> wrote:
>>> Hi all,
>>>
>>> we're experiencing a strange behavior in our Archiva instance (latest
>>> 2.1.0-SNAPSHOT):
>>>
>>> During the day, the upload performance from Jenkins degrades heavily. We
>>> have a job which uploads a few hundred artifacts and in the morning this
>>> takes between 50 seconds and 1:10. During the day this gets slower and
>>> slower (around 1 PM we're at 3-5 minutes, at 4 PM we're usually between
>>> 8-11 minutes). This is with no other uploads running in parallel and I
>>> don't see any load on the Archiva server.
>>>
>>> Then after a night has passed, we're back to 50 seconds and the same
>>> starts over.
>>>
>>> Is there anything which could cause this? I checked, the repo scanning
>>> queue and index queue are 0 when I start my tests. I also don't see any
>>> big network or IO load which could explain it. And as this is now for
>>> over a week, ever day the same, I suspect it is something inside Archvia.
>>>
>>> Are there any jobs running at night / once a day, which could explain
>>> the behavior? Something within Jackrabbit? The index? Is it that we run
>>> the repo crons at night (1 am) and they do some sort of cleanup?
>>>
>>> Greetings
>>> -Sascha-


Mime
View raw message