couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Adam Kocoloski <kocol...@apache.org>
Subject Re: stats test failure (was Re: [VOTE] Apache CouchDB 0.9.1 release)
Date Tue, 30 Jun 2009 16:05:37 GMT
Hi Paul, I was thinking we wouldn't even patch this particular bug in  
0.9.1.  Do you think we should?  I suppose now's the time, since I  
just checked in the code to fix COUCHDB-398 and we're otherwise ready  
for Noah to cut another release.  Cheers,

Adam

On Jun 30, 2009, at 11:47 AM, Paul Davis wrote:

> Also my new version would probably only exacerbate this issue.
>
> To do this proper we should add a stats end point to query the raw
> values. Adding the counter could be done incrementally, but applying
> my patch would surely not make it into the 0.9.x branch as its a hefty
> change.
>
> As Adam just posted, I'd change this with the minimal code needed for
> 0.9.x and then make sure the updates fix it proper.
>
> On Tue, Jun 30, 2009 at 11:24 AM, Robert
> Dionne<dionne@dionne-associates.com> wrote:
>> Adam,
>>
>>  there's also a nit in init_timers, 3 of them are off by a zero  
>> according to
>> the comments, though these will be made moot by Paul's new version
>>
>> Cheers,
>>
>> Bob
>>
>>
>> On Jun 30, 2009, at 11:20 AM, Adam Kocoloski wrote:
>>
>>> Well, hmmm .... fixing the counters so that /_restart actually  
>>> restarts
>>> the counters was easy enough, but now I've discovered another  
>>> bug.  The part
>>> of the stats test that fails is
>>>
>>> couchdb.open_databases.max < 5
>>>
>>> Let's review.  There are two types of counters in the stats module:
>>> absolute and incremental.  For incremental counters, the mean/min/ 
>>> max/stddev
>>> calculations measure the *rate of change* of the counter.  The
>>> open_databases counter is an incremental one, so  
>>> couchdb.open_databases.max
>>> measures the rate of DB creation, not the largest number of open  
>>> DBs!  I
>>> think the test should simply be
>>>
>>> couchdb.open_databases.current < 5
>>>
>>> I also think this was a reason why the test suite was only failing
>>> intermittently -- it really depended on how quickly Couch could  
>>> create new
>>> files.  We should review the tests for other cases like this at  
>>> some point.
>>>
>>> I'm cleaning my edits up now and will check stuff in shortly.   
>>> Cheers,
>>>
>>> Adam
>>>
>>> On Jun 29, 2009, at 7:36 PM, Chris Anderson wrote:
>>>
>>>> On Tue, Jun 30, 2009 at 1:32 AM, Adam Kocoloski<adam.kocoloski@gmail.com

>>>> >
>>>> wrote:
>>>>>
>>>>> Yes, it was the 0.9.1 tarball that failed for me.
>>>>>
>>>>> The fix is as easy as disabling the test, so if we're going to do
>>>>> anything I
>>>>> think we might as well fix it. Cheers,
>>>>>
>>>>
>>>> +1 on backporting a fix, since it's so simple.
>>>>
>>>>> Adam
>>>>>
>>>>> Sent from my iPhone
>>>>>
>>>>> On Jun 29, 2009, at 7:29 PM, Chris Anderson <jchris@apache.org>
 
>>>>> wrote:
>>>>>
>>>>>> On Mon, Jun 29, 2009 at 5:21 PM, Adam Kocoloski<kocolosk@apache.org

>>>>>> >
>>>>>> wrote:
>>>>>>>
>>>>>>> On Jun 25, 2009, at 2:01 PM, Wojciech Kaczmarek wrote:
>>>>>>>
>>>>>>>> Hi!
>>>>>>>>
>>>>>>>> On Thu, Jun 25, 2009 at 14:51, <mickael.bailly@free.fr>
wrote:
>>>>>>>>>
>>>>>>>>> got an error on the testsuite on Fedora 11 :
>>>>>>>>>
>>>>>>>>> on the test "stats" : # Assertion 'open_databases >
0 && max  
>>>>>>>>> >=
>>>>>>>>> open_databases, name' failed: should keep the same number
of  
>>>>>>>>> open
>>>>>>>>> databases
>>>>>>>>> when reaching the max_dbs_open limit
>>>>>>>>
>>>>>>>> I've had the same error for 0.9.0 when I ran couchdb as root
>>>>>>>> (different architecture though, it was OSX 10.4 PPC). I haven't
>>>>>>>> investigated why is it so, just noticed that test failed
 w/ 
>>>>>>>> root and
>>>>>>>> passed otherwise.
>>>>>>>
>>>>>>> I also see this failure running CouchDB on OS X 10.5.7 as a 

>>>>>>> non-admin
>>>>>>> user.
>>>>>>> I think we might want to dig into this a bit more and either
 
>>>>>>> fix it or
>>>>>>> disable the test before releasing 0.9.1.
>>>>>>>
>>>>>>> Adam
>>>>>>>
>>>>>>
>>>>>> I haven't seen this failure in a while, but if it's the 0.9.1  
>>>>>> release
>>>>>> candidate that's failing, I wouldn't be against disabling the  
>>>>>> test.
>>>>>>
>>>>>> Can we confirm that it's the 0.9.1 tarball that fails? Works  
>>>>>> for me
>>>>>> here.
>>>>>>
>>>>>> Chris
>>>>>>
>>>>>> --
>>>>>> Chris Anderson
>>>>>> http://jchrisa.net
>>>>>> http://couch.io
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Chris Anderson
>>>> http://jchrisa.net
>>>> http://couch.io
>>>
>>
>>


Mime
View raw message