mogadm host delete not so good?

[Brad Fitzpatrick]

The latest svn now has much-expanded tests which reproduce this.

Next up: a fix.


On Thu, 21 Dec 2006, Brett G. Durrett wrote:

>
> Using 2_02
>
> We have some old mogile storage nodes that I wanted to take out of the
> cluster and out of the mogile to reduce clutter in the output of
> mogadm.  About two weeks ago I marked all devices on the host as 'dead'
> and based on the database activity, it looks like their files got
> replicated elsewhere.  The machine has been idle for pretty much a week
> so I decided it would be okay to purge it from mogile.
>
> I did this:
>
> # mogadm host delete OLDHOST
>
> And pretty quickly nothing worked.  Looking at the daemon log, it
> appears the tracker does not like having a host go MIA so it dies.
>
> Dec 21 14:40:41 AF001079 mogilefsd[5728]: crash log: No host for dev 6
> (host 4) at /usr/bin/mogilefsd line 329.
> Dec 21 14:40:41 AF001079 mogilefsd[5728]: ending run
> Dec 21 14:40:43 AF001079 mogilefsd[9415]: Child 5728 (monitor) died: 0
> (UNEXPECTED)
>
> I recovered by going into the database and manually re-adding the row
> for the host (setting its status to 'dead').  This appears to have
> addressed the big failure.
>
> My question: is there any way to remove a device from mogile and if so,
> what is the correct procedure?
>
> Thanks,
>
> B-
>
>