[ceph-users] Options/scenarios for a dead monitor

Gregory Farnum greg at inktank.com
Thu Feb 21 13:35:12 PST 2013


On Thu, Feb 21, 2013 at 1:31 PM, Gregory Farnum <greg at inktank.com> wrote:
> On Thu, Feb 21, 2013 at 1:22 PM, Martin B Nielsen <martin at unity3d.com> wrote:
>> Hi,
>>
>> Just curious here - if a setup would have only 1 monitor - and that
>> monitor dies; what would happen?
>>
>> If say the cluster is only used as rbd with blockdevices running VM -
>> would they keep running or would everything just stop?
>>
>> Also, what options would there be if the disk with the monitor died
>> and there was no backup? Can one create a new monitor and just dump
>> into the cluster or would that require a complete backup of whatever
>> the original monitor had (as in, if the disks of the one monitor dies
>> and one has no backup = game over?). Or would a backup not even
>> suffice?
>
> Game over, essentially. You shouldn't run anything but the most
> trivial test clusters with a single monitor. I could reconstruct a
> usable monitor from data available through the rest of the system if
> necessary, and for a sufficient fee to Inktank — someday we will have
> tools for this, but they don't exist right now — but it's a lot
> cheaper to just run three monitors from the get-go. ;)
> -Greg

Oh, and to be more specific about the clients and OSDs — they would
keep running for probably a couple minutes, but eventually they would
do something that required a monitor acknowledgement, or time out
their maps and stop doing any work until they could re-establish a
connection.


More information about the ceph-users mailing list