Lustre discuss - Aug 2010 - Enabling async journals while the filesystem is active

Is it safe to enable async journals on the OSS''s while the filesystem
is active?
I''d like to see how it works for us.

Thanks
Erik

Yes, but depending on the Lustre version there are several bugs in the 
async journal code.

Kevin


Erik Froese wrote:
> Is it safe to enable async journals on the OSS''s while the
filesystem is active?
> I''d like to see how it works for us.
>
> Thanks
> Erik
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>

Thanks Kevin. This is 1.8.3.

On Fri, Aug 20, 2010 at 10:02 AM, Kevin Van Maren
<kevin.van.maren at oracle.com> wrote:
> Yes, but depending on the Lustre version there are several bugs in the
async
> journal code.
>
> Kevin
>
>
> Erik Froese wrote:
>>
>> Is it safe to enable async journals on the OSS''s while the
filesystem is
>> active?
>> I''d like to see how it works for us.
>>
>> Thanks
>> Erik
>> _______________________________________________
>> Lustre-discuss mailing list
>> Lustre-discuss at lists.lustre.org
>> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>>
>
>

Really? I had heard of a serious bug before async journal commits were 
included in a release, but I had understood that things were working 
well at the few sites running it.

Kevin Van Maren wrote:
> Yes, but depending on the Lustre version there are several bugs in the 
> async journal code.
>
> Kevin
>
>
> Erik Froese wrote:
>   
>> Is it safe to enable async journals on the OSS''s while the
filesystem is active?
>> I''d like to see how it works for us.
>>
>> Thanks
>> Erik
>> _______________________________________________
>> Lustre-discuss mailing list
>> Lustre-discuss at lists.lustre.org
>> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>>   
>>     
>
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>
>

I remember something about a bug during recovery after a crash in 1.8.1.1.
Erik

On Fri, Aug 20, 2010 at 8:15 PM, Peter Jones <peter.x.jones at oracle.com>
wrote:
> Really? I had heard of a serious bug before async journal commits were
> included in a release, but I had understood that things were working well
at
> the few sites running it.
>
> Kevin Van Maren wrote:
>>
>> Yes, but depending on the Lustre version there are several bugs in the
>> async journal code.
>>
>> Kevin
>>
>>
>> Erik Froese wrote:
>>
>>>
>>> Is it safe to enable async journals on the OSS''s while the
filesystem is
>>> active?
>>> I''d like to see how it works for us.
>>>
>>> Thanks
>>> Erik
>>> _______________________________________________
>>> Lustre-discuss mailing list
>>> Lustre-discuss at lists.lustre.org
>>> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>>>
>>
>> _______________________________________________
>> Lustre-discuss mailing list
>> Lustre-discuss at lists.lustre.org
>> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>>
>>
>

Quite. And async journal commits were officially added in 1.8.2.

Erik Froese wrote:
> I remember something about a bug during recovery after a crash in 1.8.1.1.
> Erik
>
> On Fri, Aug 20, 2010 at 8:15 PM, Peter Jones <peter.x.jones at
oracle.com> wrote:
>   
>> Really? I had heard of a serious bug before async journal commits were
>> included in a release, but I had understood that things were working
well at
>> the few sites running it.
>>
>> Kevin Van Maren wrote:
>>     
>>> Yes, but depending on the Lustre version there are several bugs in
the
>>> async journal code.
>>>
>>> Kevin
>>>
>>>
>>> Erik Froese wrote:
>>>
>>>       
>>>> Is it safe to enable async journals on the OSS''s while
the filesystem is
>>>> active?
>>>> I''d like to see how it works for us.
>>>>
>>>> Thanks
>>>> Erik
>>>> _______________________________________________
>>>> Lustre-discuss mailing list
>>>> Lustre-discuss at lists.lustre.org
>>>> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>>>>
>>>>         
>>> _______________________________________________
>>> Lustre-discuss mailing list
>>> Lustre-discuss at lists.lustre.org
>>> http://lists.lustre.org/mailman/listinfo/lustre-discuss
>>>
>>>
>>>       
>
>

On Fri, 2010-08-20 at 17:36 -0700, Peter Jones wrote:
> Quite. And async journal commits were officially added in 1.8.2.
> 
> Erik Froese wrote:
> > I remember something about a bug during recovery after a crash in
1.8.1.1.
> > Erik
> >
> > On Fri, Aug 20, 2010 at 8:15 PM, Peter Jones <peter.x.jones at
oracle.com> wrote:
> >   
> >> Really? I had heard of a serious bug before async journal commits
were
> >> included in a release, but I had understood that things were
working well at
> >> the few sites running it.
We''ve been quite happy with it. I recall a bug being reported against
them, but IIRC that was actually a longstanding bug that predated async
journals. My recollection is foggy, though so perhaps Oleg can set the
record straight.
-- 
Dave Dillow
National Center for Computational Science
Oak Ridge National Laboratory
(865) 241-6602 office

I think we''re going to benefit from it.

Last question: My clients and routers are still running Lustre
1.8.1.1. The MDS/OSSes are on 1.8.3.
Would you recommend a client upgrade before enabling async journals?

Erik

On Fri, Aug 20, 2010 at 9:35 PM, David Dillow <dillowda at ornl.gov>
wrote:
> On Fri, 2010-08-20 at 17:36 -0700, Peter Jones wrote:
>> Quite. And async journal commits were officially added in 1.8.2.
>>
>> Erik Froese wrote:
>> > I remember something about a bug during recovery after a crash in
1.8.1.1.
>> > Erik
>> >
>> > On Fri, Aug 20, 2010 at 8:15 PM, Peter Jones <peter.x.jones at
oracle.com> wrote:
>> >
>> >> Really? I had heard of a serious bug before async journal
commits were
>> >> included in a release, but I had understood that things were
working well at
>> >> the few sites running it.
>
> We''ve been quite happy with it. I recall a bug being reported
against
> them, but IIRC that was actually a longstanding bug that predated async
> journals. My recollection is foggy, though so perhaps Oleg can set the
> record straight.
> --
> Dave Dillow
> National Center for Computational Science
> Oak Ridge National Laboratory
> (865) 241-6602 office
>
>

Hello!

  You would need to upgrade your clients to at least 1.8.2, otherwise you might
hit bug 19128
  during replay that would lead to losing some of the data being replayed.
  Version on the routers is not important for async journals feature.

Bye,
    Oleg
On Aug 21, 2010, at 10:06 AM, Erik Froese wrote:
> I think we''re going to benefit from it.
> 
> Last question: My clients and routers are still running Lustre
> 1.8.1.1. The MDS/OSSes are on 1.8.3.
> Would you recommend a client upgrade before enabling async journals?
> 
> Erik
> 
> On Fri, Aug 20, 2010 at 9:35 PM, David Dillow <dillowda at ornl.gov>
wrote:
>> On Fri, 2010-08-20 at 17:36 -0700, Peter Jones wrote:
>>> Quite. And async journal commits were officially added in 1.8.2.
>>> 
>>> Erik Froese wrote:
>>>> I remember something about a bug during recovery after a crash
in 1.8.1.1.
>>>> Erik
>>>> 
>>>> On Fri, Aug 20, 2010 at 8:15 PM, Peter Jones <peter.x.jones
at oracle.com> wrote:
>>>> 
>>>>> Really? I had heard of a serious bug before async journal
commits were
>>>>> included in a release, but I had understood that things
were working well at
>>>>> the few sites running it.
>> 
>> We''ve been quite happy with it. I recall a bug being reported
against
>> them, but IIRC that was actually a longstanding bug that predated async
>> journals. My recollection is foggy, though so perhaps Oleg can set the
>> record straight.
>> --
>> Dave Dillow
>> National Center for Computational Science
>> Oak Ridge National Laboratory
>> (865) 241-6602 office
>> 
>> 
> _______________________________________________
> Lustre-discuss mailing list
> Lustre-discuss at lists.lustre.org
> http://lists.lustre.org/mailman/listinfo/lustre-discuss