[Gluster-users] self-heal stops some vms (virtual machines)

[João Pagaime]

yes, a real problem, enough to start thinking real hard on architecture 
scenarios

sorry but I can't share any solutions at this time

one (complicated) workaround would be to "medically induce a coma" on a 
VM as the self-heal starts on it, and  resurrect it afterwards.
I mean something like this:
  $ virsh suspend <vm-id>
  (do self-heal on vm's disks)
  $ virsh resume <vm-id>
problems: several, including to the vm users, but better than a kernel 
lock-up. Feasibility problem: how to detect efficiently when the 
self-heal starts on a specific file on the brick

another related problem may be how to mitigate IO starvation on the 
brick, when self-healing kicks in, since that process maybe a IO-hog. 
But I think this is a lesser problem

best regards
Joao


Em 27-02-2014 09:08, Fabio Rosati escreveu:
> Hi All,
>
> I run in exactly the same problem encountered by Joao.
> After rebooting one of the GlusterFS nodes, self-heal starts and some VMs can't access their disk images anymore.
>
> Logs from one of the VMs after one gluster node has rebooted:
>
> Feb 25 23:35:47 fwrt2 kernel: EXT4-fs error (device dm-2): __ext4_get_inode_loc: unable to read inode block - inode=2145, block=417
> Feb 25 23:35:47 fwrt2 kernel: end_request: I/O error, dev vda, sector 15032608
> Feb 25 23:35:47 fwrt2 kernel: end_request: I/O error, dev vda, sector 15307504
> Feb 25 23:35:47 fwrt2 kernel: end_request: I/O error, dev vda, sector 15307552
> Feb 25 23:35:47 fwrt2 kernel: end_request: I/O error, dev vda, sector 15307568
> Feb 25 23:35:47 fwrt2 kernel: end_request: I/O error, dev vda, sector 15307504
> Feb 25 23:35:47 fwrt2 kernel: end_request: I/O error, dev vda, sector 12972672
> Feb 25 23:35:47 fwrt2 kernel: EXT4-fs error (device dm-1): ext4_find_entry: reading directory #123 offset 0
> Feb 25 23:35:47 fwrt2 kernel: Core dump to |/usr/libexec/abrt-hook-ccpp 7 0 2757 0 23 1393367747 e pipe failed
> Feb 25 23:35:47 fwrt2 kernel: end_request: I/O error, dev vda, sector 9250632
> Feb 25 23:35:47 fwrt2 kernel: Read-error on swap-device (253:0:30536)
> Feb 25 23:35:47 fwrt2 kernel: Read-error on swap-device (253:0:30544)
> [...]
>
>
> I few hours later the VM seemed to be freezed and I had to kill and restart it, no more problems after reboot.
>
> This is the volume layout:
>
> Volume Name: gv_pri
> Type: Distributed-Replicate
> Volume ID: 3d91b91e-4d72-484f-8655-e5ed8d38bb28
> Status: Started
> Number of Bricks: 2 x 2 = 4
> Transport-type: tcp
> Bricks:
> Brick1: nw1glus.gem.local:/glustexp/pri1/brick
> Brick2: nw2glus.gem.local:/glustexp/pri1/brick
> Brick3: nw3glus.gem.local:/glustexp/pri2/brick
> Brick4: nw4glus.gem.local:/glustexp/pri2/brick
> Options Reconfigured:
> storage.owner-gid: 107
> storage.owner-uid: 107
> server.allow-insecure: on
> network.remote-dio: on
> performance.write-behind-window-size: 16MB
> performance.cache-size: 128MB
>
> OS: CentOS 6.5
> GlusterFS version: 3.4.2
>
> The qemu-kvm VMs access their qcow2 disk images using the native Gluster support (no fuse mount).
> In the Gluster logs I didn't find anything special logged during self-heal but I can post them if needed.
>
> Anyone have an idea of what can cause these problems?
>
> Thank you
> Fabio
>
>
> ----- Messaggio originale -----
> Da: "João Pagaime" <joao.pagaime at gmail.com>
> A: Gluster-users at gluster.org
> Inviato: Venerdì, 7 febbraio 2014 13:13:59
> Oggetto: [Gluster-users] self-heal stops some vms (virtual machines)
>
> hello all
>
> I have a replicate volume that holds kvm  vms (virtual machines)
>
> I had to stop one gluster-server for maintenance . That part of the
> operation went well: no vms problems after shutdown
>
> the problems started after booting the gluster-server. Self-healing
> started as expected, but some vms  locked up with disk problems
> (time-outs), as self-healing goes by them.
> Some VMs did survive the self-healing . I suppose the ones with low IO
> activity or less sensitive to disk problems
>
> is there some specific gluster configuration to enable a self-healing
> ride-through on running-vms? (cluster.data-self-heal-algorithm is
> already on the diff mode)
>
> is there some tweaks recommended to do on vms running on top of gluster?
>
> current config:
>
> gluster:   3.3.0-1.el6.x86_64
>
> --------------------- volume:
> # gluster volume info VOL
>
> Volume Name: VOL
> Type: Distributed-Replicate
> Volume ID: f44182d9-24eb-4953-9cdd-71464f9517e0
> Status: Started
> Number of Bricks: 2 x 2 = 4
> Transport-type: tcp
> Bricks:
> Brick1: one-gluster01:/san02-v2
> Brick2: one-gluster02:/san02-v2
> Brick3: one-gluster01:/san03
> Brick4: one-gluster02:/san04
> Options Reconfigured:
> diagnostics.count-fop-hits: on
> diagnostics.latency-measurement: on
> nfs.disable: on
> auth.allow:x
> performance.flush-behind: off
> cluster.self-heal-window-size: 1
> performance.cache-size: 67108864
> cluster.data-self-heal-algorithm: diff
> performance.io-thread-count: 32
> cluster.min-free-disk: 250GB
>
> thanks,
> best regards,
> joao
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://supercolony.gluster.org/mailman/listinfo/gluster-users