Re: Turning on SCRUB back on - any suggestion ?

[Henrik Korkuc <lists@xxxxxxxxx>]

I think settings apply to both kinds of scrubs

On 3/13/15 13:31, Andrija Panic wrote:

Interesting....thx for that Henrik.

BTW, my placements groups are arround 1800 objects (ceph pg dump) - meainng max of 7GB od data at the moment,

regular scrub just took 5-10sec to finish. Deep scrub would I guess take some minutes for sure

What about deepscrub - timestamp is still some months ago, but regular scrub is fine now with fresh timestamp...?

I don't see max deep scrub setings - or are these settings applied in general for both kind on scrubs ?

Thanks

On 13 March 2015 at 12:22, Henrik Korkuc <lists@xxxxxxxxx> wrote:

I think that there will be no big scrub, as there are limits of maximum scrubs at a time.
http://ceph.com/docs/master/rados/configuration/osd-config-ref/#scrubbing

If we take "osd max scrubs" which is 1 by default, then you will not get more than 1 scrub per OSD.

I couldn't quickly find if there are cluster wide limits.

On 3/13/15 10:46, Wido den Hollander wrote:

On 13-03-15 09:42, Andrija Panic wrote:

Hi all,

I have set nodeep-scrub and noscrub while I had small/slow hardware for
the cluster.
It has been off for a while now.

Now we are upgraded with hardware/networking/SSDs and I would like to
activate - or unset these flags.

Since I now have 3 servers with 12 OSDs each (SSD based Journals) - I
was wondering what is the best way to unset flags - meaning if I just
unset the flags, should I expect that the SCRUB will start all of the
sudden on all disks - or is there way to let the SCRUB do drives one by
one...

So, I *think* that unsetting these flags will trigger a big scrub, since
all PGs have a very old last_scrub_stamp and last_deepscrub_stamp

You can verify this with:

$ ceph pg <pgid> query

A solution would be to scrub each PG manually first in a timely fashion.

$ ceph pg scrub <pgid>

That way you set the timestamps and slowly scrub each PG.

When that's done, unset the flags.

Wido

In other words - should I expect BIG performance impact or....not ?

Any experience is very appreciated...

Thanks,

-- 

Andrija Panić


_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

--

Andrija Panić

_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com