Re: regression since 2.1.3 (solaris/zfs)

[Jens Axboe <axboe@kernel.dk>]

On 03/05/2014 02:12 PM, Robin P. Blanchard wrote:
>
> On Mar 5, 2014, at 16:05 PM, Jens Axboe <axboe@kernel.dk> wrote:
>
>> On 03/05/2014 01:39 PM, Robin P. Blanchard wrote:
>>>
>>> On Mar 5, 2014, at 12:06 PM, Jens Axboe <axboe@kernel.dk> wrote:
>>>
>>>> On Mar 5, 2014, at 9:27 AM, Robin P. Blanchard <robin@coraid.com> wrote:
>>>>
>>>>>
>>>>> On Mar 5, 2014, at 11:18 AM, Jens Axboe <axboe@kernel.dk> wrote:
>>>>>
>>>>>> On 2014-03-05 09:11, Robin P. Blanchard wrote:
>>>>>>>
>>>>>>> On Mar 3, 2014, at 10:25 AM, Jens Axboe <axboe@kernel.dk> wrote:
>>>>>>>
>>>>>>>> On 03/02/2014 07:59 AM, Robin P. Blanchard wrote:
>>>>>>>>> My config file has direct=0, which until 2.1.4 worked as expected.
>>>>>>>>> Things seem to regress since.
>>>>>>>>>
>>>>>>>>> I apologize in advance if this has already been reported. Please
>>>>>>>>> let me know what I can do to further help (truss/debug).
>>>>>>>>
>>>>>>>> This isn't a known issue, so thanks for reporting it. The easiest way to debug this is to git bisect it. Looks like you are running from the tar balls, but I assume you have git installed? I'm assuming fio-2.1.3 worked for you - if not, just replace fio-2.1.3 in the below with whatever latest version did work. If you do, the cheat sheet is something ala:
>>>>>>>>
>>>>>>>> $ git clone git://git.kernel.dk/fio
>>>>>>>> $ cd fio; make
>>>>>>>> $ git bisect start
>>>>>>>> $ git bisect good fio-2.1.3
>>>>>>>> $ git bisect bad fio-2.1.4
>>>>>>>>
>>>>>>>> This starts the bisect series, now do:
>>>>>>>>
>>>>>>>> $ make clean; make
>>>>>>>>
>>>>>>>> and re-run your direct=0 job file. If it worked, then you do
>>>>>>>>
>>>>>>>> $ git bisect good
>>>>>>>>
>>>>>>>> and if not, you do git bisect bad instead. This gets you a new point in the tree to test, so repeat the make clean; make and re-run the test.
>>>>>>>> Keep doing this good/bad iteration until fio tells you what commit broke the test for you. Then send those results here!
>>>>>>>>
>>>>>>>> --
>>>>>>>> Jens Axboe
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Here�s where it started working again:
>>>>>>>
>>>>>>> # git bisect good
>>>>>>> Bisecting: 4 revisions left to test after this (roughly 2 steps)
>>>>>>> [3bb0a7b0fda9945973f799ab253c70d3cb0e5c8b] howto: Fix redundant entries
>>>>>>>
>>>>>>> Let me know how else I can help.
>>>>>>
>>>>>> Please keep going until it tells you what the definitively bad commit is. It'll end up spitting out that info, if you keep doing git bisect good/bad on each test point. You need just ~2 more tests after this one.
>>>>>>
>>>>>> --
>>>>>> Jens Axboe
>>>>>>
>>>>>
>>>>> Here you go:
>>>>>
>>>>> # git bisect good
>>>>> ddc0cc31a2b75b1c7dde870c8867af11fa44db92 is the first bad commit
>>>>> commit ddc0cc31a2b75b1c7dde870c8867af11fa44db92
>>>>> Author: Jens Axboe <axboe@kernel.dk>
>>>>> Date:   Fri Oct 11 10:27:28 2013 -0600
>>>>>
>>>>>    ppc: disable CPU clock until we can detect whether we have it or not
>>>>>
>>>>>    The child segfault test should catch it, however it does not on
>>>>>    AIX at least.
>>>>>
>>>>>    Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>>>
>>>> Hmm, that can�t possibly be correct. Would you mind redoing the bisection,
>>>> just to double check?
>>>>
>>>> Thanks!
>>>>
>>>> �
>>>> Jens Axboe
>>>
>>> Ok. Let�s try this again. Silly user, me.
>>>
>>> # git bisect good
>>> 7cb024f89dbbc314e740885afccd9a05da056cf1 is the first bad commit
>>> commit 7cb024f89dbbc314e740885afccd9a05da056cf1
>>> Author: Jens Axboe <axboe@kernel.dk>
>>> Date:   Wed Nov 6 15:37:35 2013 -0700
>>>
>>>      solaris: ensure that -D_REENTRANT gets set
>>>
>>>      Apparently some Solaris' require this for threadsafe
>>>      errno.
>>>
>>>      Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>>
>>> :100755 100755 b6bfe19aa743fc4104eb587b3ff6068fb5dc67ef ef7be0180258abecd4703ebfcf4ed63625d6392f M      configure
>>>
>>>
>>> I have the complete git/bisect session in a screen log if you�d like it.
>>
>> That makes a lot more sense! Can you do:
>>
>> $ git checkout -f master
>> $ git revert 7cb024f89dbbc314e740885afccd9a05da056cf1
>> $ make clean; make
>>
>> and retest just to be on the safe side? Also, please attach the job file you are using.
>>
>> --
>> Jens Axboe
>
>
> Actually, based on the git bisect work, I took out -D_REENTRANT from configure,
 > and fio-2.1.6 now works as expected. Still want the job file?

I do, since the direct part makes little sense to me, if you have 
direct=0 set.

-- 
Jens Axboe