On 06/25/2013 01:43 PM, praveen malviya wrote:
> Thanks for the comments and discussion. I will respond for other
> comments soon, as of now starting with this mail.
> Please see the response below.
>
> One more point can be brought into discussion here. The case is when
> admin restart operation is invoked on the component and
> saAmfDisableRestart is true for it.
>   Now if saAmfSUFailover is also true for the SU of this component, then:
> case1) AMF should honor saAmfSUFailover and perform the failover of
> whole SU.
> case2) Amf should reject the comp restart admin operation because whole
> su should fail-over as a single entity. In my opinion AMF should reject
> the operation.

Agree simplest and safest for now

>
> Any comments?
>
> On 25-Jun-13 3:20 PM, Mathivanan Naickan Palanivelu wrote:
>>> -----Original Message-----
>>> From: Hans Feldt [mailto:[email protected]]
>>> Sent: Tuesday, June 25, 2013 3:01 PM
>>> To: Mathivanan Naickan Palanivelu
>>> Cc: Praveen Malviya; Nagendra Kumar; [email protected]
>>> Subject: Re: [PATCH 6 of 6] amf: handle sufailover in SU FSM and Comp
>>> FSM
>>> at amfnd [#98]
>>>
>>>
>>> On 06/24/2013 04:39 PM, Mathivanan Naickan Palanivelu wrote:
>>>> Hi Praveen,
>>>>
>>>> Thanks for the clarification.
>>>> Please find some other comments below:
>>>>
>>>> - I'm sure you would have had a reason to do it this way, but I thought
>>> alternatively performing the repair at the AMFND itself(instead of
>>> notifying
>>> to AMFD) is one viable option!
>>>
>>> Guess not it interferes with a potential auto adjust feature right?
>> I meant that we could avoid the 'extra step' of informing AMFD.
>> May be, we could reduce the latency (in repairing) if avoid going
>> through AMFD.
>> Well, AMFND could be made aware of autoadjust attributes, isn't it!
> saAmfSGAutoRepair attribute is maintained at amfd only. Current
> implementation support repair of SU through admin operation.
> When repair admin operation is invoked, amfd informs amfnd to perform
> repair. So functionally it is amfnd that performs the repair.
> In these patches, this same mechanism is used to perform repair. After
> performing failover of assignments of faulted SU (this only amfd can
> do), amfd checks  saAmfSGAutoRepair and if it is true informs amfnd to
> perform repair. Rest of the flow is same as admin repair operation.
> If saAmfSGAutoRepair  is maintained at amfnd, then in ccb modify
> operations on this attribute  amfd will have to update to each amfnd
> hosting SUs of this SG. These again will be "extra steps".

Yes I think your chosen design is elegant and lean! Reuse the same logic 
for auto and manual repair...

>
>>>> Some other comments I had in mind are:
>>>> - is si-si dependencies automatically taken care by these changes?
>>>> - are we honouring 'order' of termination of components based on the
>>> 'instantiationlevel'(in reverse)?
>>>
>>> 3.11.1.3.2 SU failover:
>>>
>>> "If the service unit is configured to fail over as a single entity
>>> (saAmfSUFailover set to SA_TRUE), all other components of the service
>>> unit
>>> are abruptly terminated"
>>>
>>> Does not say anything about ordering. Should we add some defined
>>> ordering
>>> semantics on top of that you mean?
>>>
>>>
>> I meant the below. Also, SU failover 'effectively' involves restart of
>> the failed SU!
> -In case of SU failover no quiesced assignment will be given. So SI dep
> will be honored only while giving active assignment to the standby or
> spare su.
> - As a part of su failover components will be abruptly terminated
> without honoring instantiation-level. I think instantiation-level is to
> be considered during graceful termination
> of component like Lock-in operation.

We could of course honour inst level for abrupt termination also if we 
want to. It is our choice. Could start unordered though as in the patch 
series.

> In case of SU restart one ticket exists  #315
>>> FYI just realized ordering and semantics of "SU restart" is wrong. Will
>>> write a defect.
> As pointed out ticket already exists #315.

My memory is rather short, seems like I wrote it myself...

But it is just not the level, it is the way SU restart is done. See 
3.11.1.2:

Restarting a service unit is achieved by the following actions:
• First, all components in the service unit are terminated in the order 
dictated by their instantiation-levels.
• In a second step, all components in the service unit are instantiated 
in the order dictated by their instantiation-levels.

That is obviously not the case since SU restart is just an iteration of 
comp restart. Should be comp termination and then comp instantiation

Thanks,
Hans






------------------------------------------------------------------------------
This SF.net email is sponsored by Windows:

Build for Windows Store.

http://p.sf.net/sfu/windows-dev2dev
_______________________________________________
Opensaf-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/opensaf-devel

Reply via email to