[Openais] questions regarding openais-0.80.1

[Hans Feldt]

edrt wrote:

> I'm going to deploy openais in a product, the minimum functionality wanted is:
> 2N model + su failover & autorepair on 2 node configuration. Can previous 
> openais release support it? And which version do you recommend? BTW, can I 
> know the estimated timeframe openais HEAD will support multinode configuration.

I don't think previous versions support what you need. The release of
"Wilson" is not decided but I guess will be delayed until the services
contained in it has reached some critical content and/or stability. This
is open source...

>>> - Although totem mechanism ensures AMF state machines on different nodes get the same
>>>   input event, is it possilbe that AMF sm still get lose syncing? If so, how will 
>>>   openais handle that ?
>>
>>Sorry, don't understand the question.
>>
> 
> 
> Eg. in a two node cluster, node1 multicast an amf event in the cluster (eg. component register).
> Normal execution flow will be: node1 and node2 both receive the event and trigger amf state 
> transition. If, for some reason, node2 experience failure (like malloc failed) during state
> transition, will openais on node2 try to recovery from the failure and resync with cluster
> the amf state?

If aisexec on node2 experience a fatal error (e.g. malloc failed), it
will exit. Other nodes failover the workloads from node2. Is is then
upto the aisexec environment (e.g. an init script) to restart aisexec on
node2. Then aisexec on node2 will sync its state and enter the AMF
cluster again.

The start, supervision and restart of aisexec is not part of openais but
will the responsibility of the openais deployer.

> 
> 
>>> - Below are the AMF B.02.01 features I find openais not supported yet. Could you confirm
>>>   these conclusions? (BTW, do you aware of other AMF-B.02.01 feature openais is missing, 
>>>   I'll build a gaplist)
>>>     - proxied component mgnt
>>>	 - dependency mgnt
>>>	 - administrative api
>>>	 - n-way redundancy model
>>>	 - n-way active redundancy model
>>>	 - component failover & autorepair
>>>	 - node failover & autorepair
>>>	 - node failfast
>>>	 - node switchover & autorepair
>>>	 - application restart
>>>	 - cluster reset
>>>	 - l3 escalation
>>
>>We will soon (this week) publish a gap list in README.amf.
>>
> 
> 
> Great, pls drop notification after it commited.

Will be posted to the mailing list.

Regards,
Hans

> 
> 
> Thanks
> Eddy
> 
> 
>