On Wed, Mar 1, 2017 at 11:34 PM, Sage Weil <[email protected]> wrote:

> On Wed, 1 Mar 2017, Stephen Blinick wrote:
> > Hello, I'm chasing down a situation where there's periodic slow requests
> > occurring.   While the specific version in this case is 0.80.7 Firefly, I
> > think this log format is the same in newer versions.  I can verify.
> > There's a host of symptoms going on, but one strange anomaly I found
> that I
> > wasn't able to chase down or find in search has to do with the Op
> parameters
> > for an osd_op being logged as a slow request.
> >
> > Specifically, in some cases the byte range for an op of various types
> (i.e.
> > read, writefull) is sometimes negative.  Here's an example of two slow
> > request log entries:
> >
> > #1
> > 2017-02-28 18:39:09.943169 osd.27 10.0.1.84:6822/2402845 5255 : [WRN]
> slow
> > request 16.440574 seconds old, received at 2017-02-28 18:38:53.502539:
> > osd_op(client.2000529.0:496 ObjectNameOne [writefull 0~4194304]
> 3.3f26fcc9
> > ondisk+write e691) v4 currently commit sent
> >
> > #2
> > 2017-02-28 18:39:05.959253 osd.40 10.0.1.88:6831/2187230 6180 : [WRN]
> slow
> > request 8.470175 seconds old, received at 2017-02-28 18:38:57.489045:
> > osd_op(client.1941470.0:21164213 ObjectNameTwo [read 3670016~524288]
> > 3.3a50c331 ack+read e691) v4 currently started
> >
> > As you can see, some of them show the byte range A~B where B is lower
> than
> > A.  I'm mostly interested to find out if this is an indication of any
>
> This is quirky Ceph convention for printing extents as offset~length (it's
> not start~end).  So these look fine.
>

Ahh I should have guessed.  This makes a lot more sense. I'll update the
parser accordingly.  Thanks!

>
> Firefly 0.80.7?  You should really upgrade.  Twice (to hammer and then to
> jewel).
>
>
Indeed!  We already have for the most part, but this system is in
production, so that always adds friction to any upgrade.  Definitely makes
problem debug more of a 'forensic' exercise :)


> sage
>
> > problem.   This is an EC pool, 3+2.
> >
> > Thanks,
> >
> > Stephen
> >
> >
>
_______________________________________________
ceph-users mailing list
[email protected]
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to