Hello,
On Thu, 30 Jul 2015 10:55:30 +0200 Khalid Ahsein wrote:
> Hello everybody,
>
> I’m running since 4 months a ceph cluster configured with two monitors :
>
> 1 host : 16GB RAM - 12x 4TB disks - 12 OSD - 1 monitor - RAID-1 for
> system 1 host : 16GB RAM - 12x 4TB disks - 12 OSD - 1 monitor - RAID-1
> for system
>
Too little RAM, just 2 monitors, just 2 nodes...
> This night I’ve encountered an issue with the crash of the first host.
>
> My first question is why with 1 host down, all my cluster was down
> (unable to do ceph status — hang command) and all my rbd was stuck
> without possibility to R/W.
Re-read the documentation, you need at least 3 monitors to survive the
loss of one (monitor) node.
Your osd_pool_default_min_size would have left in a usable situation, 2
nodes is really a minimal case.
> I rebooted the first host, and 2 hours later
> the second go down with the same issue (all rbd down and ceph hang).
>
> After reboot, here is ceph status :
>
> # ceph status
> cluster 9c29f469-7bad-4b64-97bf-3fbb1bbc0c5f
> health HEALTH_ERR
> 3 pgs inconsistent
> 1 pgs peering
> 1 pgs stuck inactive
> 1 pgs stuck unclean
> 36 requests are blocked > 32 sec
> 928 scrub errors
> clock skew detected on mon.drt-becks
> monmap e1: 2 mons at
> {drt-becks=172.16.21.6:6789/0,drt-marco=172.16.21.4:6789/0} election
> epoch 26, quorum 0,1 drt-marco,drt-becks osdmap e961: 24 osds: 24 up, 24
> in pgmap v2532968: 400 pgs, 1 pools, 512 GB data, 130 kobjects
> 1039 GB used, 88092 GB / 89177 GB avail
> 393 active+clean
> 3 active+clean+scrubbing+deep
> 3 active+clean+inconsistent
> 1 peering
> client io 57290 B/s wr, 7 op/s
>
You will want to:
a) fix your NTP, clock skew.
b) check your logs about the scrub errors
c) same for the stuck requests
> Also I found this error on DMESG about the crash :
>
> Message from syslogd@drt-marco at Jul 30 04:03:57 ...
> kernel:[4876519.657178] BUG: soft lockup - CPU#7 stuck for 22s!
> [btrfs-cleaner:32713]
>
> All my volumes are on BTRFS, maybe it was not a good idea ?
>
Depending on your OS, kernel version, most definitely.
Plenty of BTRFS problems in the ML archives to be found.
Christian
--
Christian Balzer Network/Systems Engineer
chibi@xxxxxxx Global OnLine Japan/Fusion Communications
http://www.gol.com/
_______________________________________________
ceph-users mailing list
ceph-users@xxxxxxxxxxxxxx
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
