[sheepdog] [sheepdog-users] [PATCH 1/3] collie: add delay_recovery {start|stop} command

[Liu Yuan]

On 07/30/2012 04:53 PM, Dietmar Maurer wrote:
>> Do you mean: shutdown whole cluster before manual recovery finished?
>>
>> If user start manual recovery, and add some new nodes info cluster, and
>> then shutdown the cluster directly, I think the cluster should kept the original
>> status, just like not adding any new nodes into cluster.
> 
> I simply do not want 'automatic' discovery. Instead, I think the admin should trigger that.
> 
> So far, you implemented a way to turn it off, but that flag does not survive a cluster restart?
> 
> 
> 

This is farm away from the implementation that let's safely stop
automatic recovery and switch to a completely manual recovery.

Every one is warned that, if the failed node event is blocked for some
time window (for e.g, 120s for Linux), most likely the VMs will put its
filesystem to read-only.

So this is kind of command need a caution.

Thanks,
Yuan