[ceph-users] Disk Down Emergency

David Turner drakonstein at gmail.com
Thu Nov 16 07:17:24 PST 2017


The first step is to make sure that it is out of the cluster.  Does `ceph
osd stat` show the same number of OSDs as in (it's the same as a line from
`ceph status`)?  It should show 1 less for up, but if it's still
registering the OSD as in then the backfilling won't start.  `ceph osd out
0` should mark it out and let the backfilling start.

If it's already out, then `ceph osd crush remove osd.0; ceph auth del
osd.0; ceph osd rm 0` will finish removing it from the cluster and let you
move forward.  It's a good idea to wait to run these commands once you have
a full copy of your data again, so really try to let the cluster do its
thing if you can.  What generally happens and what everyone (including
myself) is recommending for you, is to let the OSD get marked down (which
it has done), and then marked out.  Once it is marked out, the cluster will
rebalance and make sure that all of the data that is on the out OSD is
replicated to have the full number of copies again.  My guess is that you
just have a setting somewhere preventing the OSD from being marked out.

As far as your customer not understanding that 2 replicas is bad for
production data, write up a contract that they have to sign indemnifying
you of any responsibility if they lose data because you have warned them to
have 3 replicas.  If they don't sign it, then tell them you will no longer
manage Ceph for them.  Hopefully they wake up and make everyone's job
easier by purchasing a third server.

On Thu, Nov 16, 2017 at 9:26 AM Georgios Dimitrakakis <giorgis at acmac.uoc.gr>
wrote:

>  Thank you all for your time and support.
>
>  I don't see any backfilling in the logs and the number of
>  "active+degraded" as well as "active+remapped" and "active+clean"
>  objects is the same for some time now. The only thing I see is
>  "scrubbing".
>
>  Wido, I cannot do anything with the data in osd.0 since although the
>  failed disk seems mounted I cannot see anything and I am getting an
>  "Input/output" error.
>
>  So I guess the right action for now is to remove the OSD by issuing
>  "ceph osd crush remove  osd.0" as Sean suggested, correct?
>
>  G.
>
>
> >> Op 16 november 2017 om 14:46 schreef Caspar Smit
> >> <casparsmit at supernas.eu>:
> >>
> >>
> >> 2017-11-16 14:43 GMT+01:00 Wido den Hollander <wido at 42on.com>:
> >>
> >> >
> >> > > Op 16 november 2017 om 14:40 schreef Georgios Dimitrakakis <
> >> > giorgis at acmac.uoc.gr>:
> >> > >
> >> > >
> >> > >  @Sean Redmond: No I don't have any unfound objects. I only have
> >> "stuck
> >> > >  unclean" with "active+degraded" status
> >> > >  @Caspar Smit: The cluster is scrubbing ...
> >> > >
> >> > >  @All: My concern is because of one copy left for the data on
> >> the failed
> >> > >  disk.
> >> > >
> >> >
> >> > Let the Ceph recovery do it's work. Don't do anything manually
> >> now.
> >> >
> >> >
> >> @Wido, i think his cluster might have stopped recovering because of
> >> non-optimal tunables in firefly.
> >>
> >
> > Ah, darn. Yes, that's been a long time ago. Could very well be the
> > case.
> >
> > He could try to remove osd.0 from the CRUSHMap and let recovery
> > progress.
> >
> > I would however advise him not to fiddle with the data on osd.0. Do
> > not try to copy the data somewhere else and try to fix the OSD.
> >
> > Wido
> >
> >>
> >> > >  If I just remove the OSD.0 from crush map does that copy all
> >> its data
> >> > >  from the only one available copy to the rest unaffected disks
> >> which will
> >> > >  consequently end in having again two copies on two different
> >> hosts?
> >> > >
> >> >
> >> > Do NOT copy the data from osd.0 to another OSD. Let the Ceph
> >> recovery
> >> > handle this.
> >> >
> >> > It is already marked as out and within 24 hours or so recovery
> >> will have
> >> > finished.
> >> >
> >> > But a few things:
> >> >
> >> > - Firefly 0.80.9 is old
> >> > - Never, never, never run with size=2
> >> >
> >> > Not trying to scare you, but it's a reality.
> >> >
> >> > Now let Ceph handle the rebalance and wait.
> >> >
> >> > Wido
> >> >
> >> > >  Best,
> >> > >
> >> > >  G.
> >> > >
> >> > >
> >> > > > 2017-11-16 14:05 GMT+01:00 Georgios Dimitrakakis :
> >> > > >
> >> > > >> Dear cephers,
> >> > > >>
> >> > > >> I have an emergency on a rather small ceph cluster.
> >> > > >>
> >> > > >> My cluster consists of 2 OSD nodes with 10 disks x4TB each
> >> and 3
> >> > > >> monitor nodes.
> >> > > >>
> >> > > >> The version of ceph running is Firefly v.0.80.9
> >> > > >> (b5a67f0e1d15385bc0d60a6da6e7fc810bde6047)
> >> > > >>
> >> > > >> The cluster originally was build with "Replicated size=2" and
> >> "Min
> >> > > >> size=1" with the attached crush map,
> >> > > >> which in my understanding this replicates data across hosts.
> >> > > >>
> >> > > >> The emergency comes from the violation of the golden rule:
> >> "Never
> >> > > >> use 2 replicas on a production cluster"
> >> > > >>
> >> > > >> Unfortunately the customers never really understood well the
> >> risk
> >> > > >> and now that one disk is down I am in the middle and I must
> >> do
> >> > > >> everything in my power not to loose any data, thus I am
> >> requesting
> >> > > >> your assistance.
> >> > > >>
> >> > > >> Here is the output of
> >> > > >>
> >> > > >> $ ceph osd tree
> >> > > >> # id    weight  type name       up/down reweight
> >> > > >> -1      72.6    root default
> >> > > >> -2      36.3            host store1
> >> > > >> 0       3.63                    osd.0   down
> >> > > >> 0       ---> DISK DOWN
> >> > > >> 1       3.63                    osd.1   up
> >> > > >> 1
> >> > > >> 2       3.63                    osd.2   up
> >> > > >> 1
> >> > > >> 3       3.63                    osd.3   up
> >> > > >> 1
> >> > > >> 4       3.63                    osd.4   up
> >> > > >> 1
> >> > > >> 5       3.63                    osd.5   up
> >> > > >> 1
> >> > > >> 6       3.63                    osd.6   up
> >> > > >> 1
> >> > > >> 7       3.63                    osd.7   up
> >> > > >> 1
> >> > > >> 8       3.63                    osd.8   up
> >> > > >> 1
> >> > > >> 9       3.63                    osd.9   up
> >> > > >> 1
> >> > > >> -3      36.3            host store2
> >> > > >> 10      3.63                    osd.10  up      1
> >> > > >> 11      3.63                    osd.11  up      1
> >> > > >> 12      3.63                    osd.12  up      1
> >> > > >> 13      3.63                    osd.13  up      1
> >> > > >> 14      3.63                    osd.14  up      1
> >> > > >> 15      3.63                    osd.15  up      1
> >> > > >> 16      3.63                    osd.16  up      1
> >> > > >> 17      3.63                    osd.17  up      1
> >> > > >> 18      3.63                    osd.18  up      1
> >> > > >> 19      3.63                    osd.19  up      1
> >> > > >>
> >> > > >> and here is the status of the cluster
> >> > > >>
> >> > > >> # ceph health
> >> > > >> HEALTH_WARN 497 pgs degraded; 549 pgs stuck unclean; recovery
> >> > > >> 51916/2552684 objects degraded (2.034%)
> >> > > >>
> >> > > >> Althoug OSD.0 is shown as mounted it cannot be started
> >> (probably
> >> > > >> failed disk controller problem)
> >> > > >>
> >> > > >> # df -h
> >> > > >> Filesystem      Size  Used Avail Use% Mounted on
> >> > > >> /dev/sda3       251G  4.1G  235G   2% /
> >> > > >> tmpfs            24G     0   24G   0% /dev/shm
> >> > > >> /dev/sda1       239M  100M  127M  44% /boot
> >> > > >> /dev/sdj1       3.7T  223G  3.5T   6%
> >> > > >> /var/lib/ceph/osd/ceph-8
> >> > > >> /dev/sdh1       3.7T  205G  3.5T   6%
> >> > > >> /var/lib/ceph/osd/ceph-6
> >> > > >> /dev/sdg1       3.7T  199G  3.5T   6%
> >> > > >> /var/lib/ceph/osd/ceph-5
> >> > > >> /dev/sde1       3.7T  180G  3.5T   5%
> >> > > >> /var/lib/ceph/osd/ceph-3
> >> > > >> /dev/sdi1       3.7T  187G  3.5T   6%
> >> > > >> /var/lib/ceph/osd/ceph-7
> >> > > >> /dev/sdf1       3.7T  193G  3.5T   6%
> >> > > >> /var/lib/ceph/osd/ceph-4
> >> > > >> /dev/sdd1       3.7T  212G  3.5T   6%
> >> > > >> /var/lib/ceph/osd/ceph-2
> >> > > >> /dev/sdk1       3.7T  210G  3.5T   6%
> >> > > >> /var/lib/ceph/osd/ceph-9
> >> > > >> /dev/sdb1       3.7T  164G  3.5T   5%
> >> > > >> /var/lib/ceph/osd/ceph-0    ---> This is the problematic OSD
> >> > > >> /dev/sdc1       3.7T  183G  3.5T   5%
> >> > > >> /var/lib/ceph/osd/ceph-1
> >> > > >>
> >> > > >> # service ceph start osd.0
> >> > > >> find: `/var/lib/ceph/osd/ceph-0: Input/output error
> >> > > >> /etc/init.d/ceph: osd.0 not found (/etc/ceph/ceph.conf
> >> defines
> >> > > >> mon.store1 osd.6 osd.9 osd.1 osd.4 osd.3 osd.2 osd.8 osd.5
> >> osd.7
> >> > > >> mds.store1 mon.store3, /var/lib/ceph defines mon.store1 osd.6
> >> osd.9
> >> > > >> osd.1 osd.4 osd.3 osd.2 osd.8 osd.5 osd.7 mds.store1)
> >> > > >>
> >> > > >> I have found this:
> >> > > >>
> >> > > >
> >> > > > http://ceph.com/geen-categorie/admin-guide-
> >> > replacing-a-failed-disk-in-a-ceph-cluster/
> >> > > >> [1]
> >> > > >>
> >> > > >> and I am looking for your guidance in order to properly
> >> perform all
> >> > > >> actions in order not to loose any data and keep the ones of
> >> the
> >> > > >> second copy.
> >> > > >
> >> > > > What guidance are you looking for besides the steps to replace
> >> a
> >> > > > failed disk (which you already found) ?
> >> > > > If i look at your situation, there is nothing down in terms of
> >> > > > availability of pgs, just a failed drive which needs to be
> >> replaced.
> >> > > >
> >> > > > Is the cluster still recovering? It should reach HEALTH_OK
> >> again
> >> > > > after
> >> > > > rebalancing the cluster when an OSD goes down.
> >> > > >
> >> > > > If it stopped recovering it might have to do with the ceph
> >> tunables
> >> > > > which are not set to optimal by default on firefly and that
> >> prevents
> >> > > > further rebalancing.
> >> > > > WARNING: Dont just set tunables to optimal because it will
> >> trigger a
> >> > > > massive rebalance!
> >> > > >
> >> > > > Perhaps the second golden rule is to never run a CEPH
> >> production
> >> > > > cluster without knowing (and testing) how to replace a failed
> >> drive.
> >> > > > (Im not trying to be harsh here).
> >> > > >
> >> > > > Kind regards,
> >> > > > Caspar
> >> > > >
> >> > > >
> >> > > >> Best regards,
> >> > > >>
> >> > > >> G.
> >> > > >> _______________________________________________
> >> > > >> ceph-users mailing list
> >> > > >> ceph-users at lists.ceph.com [2]
> >> > > >> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com [3]
> >> > > >
> >> > > >
> >> > > >
> >> > > > Links:
> >> > > > ------
> >> > > > [1]
> >> > > >
> >> > > > http://ceph.com/geen-categorie/admin-guide-
> >> > replacing-a-failed-disk-in-a-ceph-cluster/
> >> > > > [2] mailto:ceph-users at lists.ceph.com
> >> > > > [3] http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
> >> > > > [4] mailto:giorgis at acmac.uoc.gr
> >> > > _______________________________________________
> >> > > ceph-users mailing list
> >> > > ceph-users at lists.ceph.com
> >> > > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
> >> > _______________________________________________
> >> > ceph-users mailing list
> >> > ceph-users at lists.ceph.com
> >> > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
> >> >
> >> _______________________________________________
> >> ceph-users mailing list
> >> ceph-users at lists.ceph.com
> >> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
> > _______________________________________________
> > ceph-users mailing list
> > ceph-users at lists.ceph.com
> > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>
> --
>  Dr. Dimitrakakis Georgios
>
>  Networks and Systems Administrator
>
>  Archimedes Center for Modeling, Analysis & Computation (ACMAC)
>  School of Sciences and Engineering
>  University of Crete
>  P.O. Box 2208
>  710 - 03 Heraklion
>  Crete, Greece
>
>  Tel: +30 2810 393717
>  Fax: +30 2810 393660
>
>  E-mail: giorgis at acmac.uoc.gr
> _______________________________________________
> ceph-users mailing list
> ceph-users at lists.ceph.com
> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ceph.com/pipermail/ceph-users-ceph.com/attachments/20171116/10e06682/attachment.html>


More information about the ceph-users mailing list