Re: [devel] [PATCH 1 of 1] amfd: mark NG locked if controller failovers during shutdown op on NG [#1513]

[Hans N. <han...@er...>]

ack, code review only/Thanks HansN

On 10/05/2015 12:47 PM, pra...@or... wrote:
>   osaf/services/saf/amf/amfd/sgproc.cc |  21 +++++++++++++++++++++
>   1 files changed, 21 insertions(+), 0 deletions(-)
>
>
> NG gets stuck in SHUTTING_DOWN state during shutdown op and controller failover.
>
> During SHUTDOWN admin operation on NG, initial admin state is set to SHUTTING_DOWN and
> it is checkpointed to standby AMFD. On decoding it, standby AMFD sets node->admin_ng
> and it clears it when active AMFD checkpoints the LOCKED state. Now after fail-over when
> AMFD gets quiescing success response from AMFND it clears this pointer in
> process_su_si_response_for_ng() assuming there is only one SU hosted on that node.
> After this when response for second SU comes, this response is not processed from NG
> perspective as AMFD has already cleared node->admin_ng. Issue does not occur when node hosts
> only one application SU.
>
> Patch fixes the problem by avoiding clearing of node->admin_ng when NG is in SHUTTING_DOWN state.
>
> diff --git a/osaf/services/saf/amf/amfd/sgproc.cc b/osaf/services/saf/amf/amfd/sgproc.cc
> --- a/osaf/services/saf/amf/amfd/sgproc.cc
> +++ b/osaf/services/saf/amf/amfd/sgproc.cc
> @@ -400,6 +400,27 @@ void process_su_si_response_for_ng(AVD_S
>   		ng->node_oper_list.erase(Amf::to_string(&node->name));
>   		TRACE("node_oper_list size:%u",ng->oper_list_size());
>   	}
> +
> +	/*Handling for the case: There are pending assignments on more than one SUs
> +	  on same node of nodegroup with atleast one quiescing assignment and controller
> +	  failover occured.
> +	  Below if block will be hit only when assignments for quiescing state are still pending
> +	  on atleast one SU and on atleast one node of NG.
> +	*/
> +	if ((ng->saAmfNGAdminState == SA_AMF_ADMIN_SHUTTING_DOWN) &&
> +			(ng->admin_ng_pend_cbk.admin_oper == 0) &&
> +			(ng->admin_ng_pend_cbk.invocation == 0)) {
> +		/*During SHUTDOWN admin operation on NG, initial admin state is set to SHUTTING_DOWN
> +		  and it is checkpointed to standby AMFD. On decoding it, standby AMFD sets
> +		  node->admin_ng and it clears it when active AMFD checkpoints the LOCKED state.
> +		  In case active AMFD sends quiescing state and reboots after checkpointing only
> +		  SHUTTING_DOWN state, standby AMFD will be able to mark NG LOCKED by processing
> +		  response of assignments as it has set node->admin_ng. So this pointer should be
> +		  cleared only when NG is marked LOCKED. And in that case we will not be in this if block.
> +		 */
> +		TRACE_1("'%s' in shutting_down state after failover.",ng->name.value);
> +		goto done;
> +	}
>   	/*If assignment changes are done on all the SUs on each node of nodegroup
>   	  then reply to IMM for status of admin operation.*/
>   	if (ng->node_oper_list.empty())