Re: Call trace in ext4_es_lru_add on 3.10 stable

[Stefan Priebe <s.priebe@profihost.ag>]

Am 23.09.2014 11:42, schrieb Jan Kara:
> On Tue 23-09-14 09:50:25, Stefan Priebe - Profihost AG wrote:
>>
>> Am 22.09.2014 um 22:20 schrieb Theodore Ts'o:
>>> On Mon, Sep 22, 2014 at 08:29:54PM +0200, Stefan Priebe wrote:
>>>> Hi,
>>>> Am 22.09.2014 18:47, schrieb Theodore Ts'o:
>>>>> On Mon, Sep 22, 2014 at 08:56:23AM +0200, Stefan Priebe wrote:
>>>>>>> That's not the whole message; you just weren't able to capture it all.
>>>>>>> How are you capturing these messages, by the way?  Serial console?
>>>>>>
>>>>>> Sorry this was an incomplete copy and paste by me.
>>>>>>
>>>>>> Here is the complete output:
>>>>>> [1578544.839610] BUG: soft lockup - CPU#7 stuck for 22s! [mysqld:29281]
>>>>>> [1578544.893450] Modules linked in: nf_conntrack_ipv4 nf_defrag_ipv4
>>>>>
>>>>> OK, thanks, this is a known bug, where when ext4 is under heavy memory
>>>>> pressure, we can end up stalling in reclaim.  This message indicates
>>>>> that the system got stalled for 22 seconds, which is not good, since
>>>>> it impacts the interactivity of your system, and increases the
>>>>> long-tail latency of requests to servers running on your system, but
>>>>> it doesn't cause any data loss or will cause any of your processes to
>>>>> crash or otherwise stop functioning (except for temporarily).
>>>>>
>>>>> It's something that we are working on, and there are patches which
>>>>> Zheng Liu submitted that still need a bit of polishing, but I hope to
>>>>> have it addressed soon.
>>>>
>>>> Thanks for your feedback. Will those patches go to stable? Any link to
>>>> those patches?
>>>
>>> I'm not sure they will go to Stable when they are ready, because the
>>> patches are somewhat complex and so they may not apply cleanly to much
>>> older kernels.
>>>
>>> The patches under discussion (some have been applied, others hae been
>>> waiting for some requested changes) can be found here:
>>>
>>> http://patchwork.ozlabs.org/patch/377720
>>> http://patchwork.ozlabs.org/patch/377721
>>> http://patchwork.ozlabs.org/patch/377722
>>> http://patchwork.ozlabs.org/patch/377723
>>> http://patchwork.ozlabs.org/patch/377724
>>> http://patchwork.ozlabs.org/patch/377725
>>> http://patchwork.ozlabs.org/patch/377727
>>
>> hui that's a lot. Are they ALL needed to fix this?
>    Yes, all of them are needed.

How can i get notified when they're ready / polished?

Stefan

>> No workaround possible?
>    I don't know about any.
>
>> What will Redhat do with their 3.10 RHEL 7 kernel?
>    Well, I cannot speak for RH guys but for SLES if there's a customer
> request, we'll just go and backport the patches...
> 								Honza
>