[Cluster-devel] Re: fence-agents: master - fencing: New option '--missing-as-off' to return OFF is machine is missing

[Marek 'marx' Grác]

David Teigland wrote:
>> fencing: New option '--missing-as-off' to return OFF is machine is missing
>>
>> If a blade is not present (i.e. removed for maintenance), the fence_bladecenter
>> cannot check the state as it is reported empty.
>>
>> Resolves: bz#248006
>>     
>
>   
>> --- a/fence/agents/bladecenter/fence_bladecenter.py
>> +++ b/fence/agents/bladecenter/fence_bladecenter.py
>> @@ -30,7 +30,10 @@ def get_power_status(conn, options):
>>  		i = conn.log_expect(options, [ node_cmd, "system>" ] , int(options["-Y"]))
>>  		if i == 1:
>>  			## Given blade number does not exist
>> -			fail(EC_STATUS)
>> +			if options.has_key("-M"):
>> +				return "off"
>> +			else:
>> +				fail(EC_STATUS)
>>     
>
> I've never used bladecenter, so I don't know when a blade number doesn't
> exist.  Does it reliably indicate that the blade is off?  If so, then
> should we default to that without a new option?  If not, then this option
> sounds bad, because it's effectively an automation of manual override, no?
>   
Yes, it is reliable. We can turn it on as default but imho it will cause 
inconsistency on how fence agents works as all other refuse to work with 
port that does not exist. In general we can divide fence agents in 
several categories:

 * iLO-like -> 1 machine we have to connect to --> if we can't connect 
we don't know if problem is in iLO or connection to machine
* APC-like -> N machines connected to central device --> we can switch 
outlet ON/OFF and we don't have to care if there is something connected
* virtual machines -> we usually don't need to remove them for maintaince
* Blade-like -> unlike APC we can work only with machines that are 
connected - so maintainance can be problem

> The typical issue with nodes removed for maintenance is that startup
> fencing tries to fence them and can't, for any agent.  The solution to
> that has always been manual override or removing the node from
> cluster.conf.
>   

m,