[sheepdog-users] Failing disk tests: umount, device NOT busy

[Liu Yuan]

On Wed, Oct 16, 2013 at 05:03:38PM +0200, Valerio Pachera wrote:
> Hi, I'm going to simulate some of the worst but very possible situation.
> 
> On my cluster test I tried to umount a device used by sheepdog.
> I was sure it was going to print the typical error message
> 
> umount: /mnt/sheep/dsk03: device is busy.
>         (In some cases useful info about processes that use
>          the device is found by lsof(8) or fuser(1))
> 
> But it doesn't.
> 
> Later some error messages start to appear in sheep.log.
> 
> Oct 16 16:22:41  ERROR [gway 3381] gateway_replication_read(268) local
> read 8036657100000000 failed, No object found
> Oct 16 16:22:41  ERROR [gway 3300] gateway_replication_read(268) local
> read 8036657100000000 failed, No object found
> Oct 16 16:22:41  ERROR [gway 3381] gateway_replication_read(268) local
> read 807ab46f00000000 failed, No object found
> Oct 16 16:22:41  ERROR [gway 3300] gateway_replication_read(268) local
> read 807ab46f00000000 failed, No object found
> Oct 16 16:22:41  ERROR [gway 3381] gateway_replication_read(268) local
> read 80a34c6700000000 failed, No object found
> Oct 16 16:22:41  ERROR [gway 3300] gateway_replication_read(268) local
> read 80a34c6700000000 failed, No object found
> 
> Sheep start writing chunks in the non mounted folder.
> That means it will fast fill up root ( / ).
> 
> Obviously the right procedure is to remove the disk by 'dog node md
> unplug', but this is the simulation of a mistake.
> 
> Your considerations are well come.

I am not sure if sheepdog should add a lock file in the data directory or it is
better accomplished by other tools/software ? I am inclined to the idea that
sheepdog should do what it is considered as must or best do it in sheepdog other
than elsewhere.

Thanks
Yuan