[Pacemaker] The question of a function which commits suicide according to STONITH by one node composition

[Andrew Beekhof]

2012/3/14 Yuusuke Iida <iidayuus at intellilink.co.jp>:
> Hi, Andrew
>
> I was going to confirm the suicide function by STONITH implemented by
> the following changes.
> https://github.com/ClusterLabs/pacemaker/commit/1a26605e1b5840f49d44b47a1f083c2cdb6f673b
>
> The environment is as follows.
> Pacemaker-1.1.6(commit 31f6ca36cca57b8c08f7cf89a205331ab8c292ad)
> Heartbeat-3.0.5
>
> I tested it by the following constitution.
> ============
> Last updated: Wed Mar 14 13:46:25 2012
> Last change: Wed Mar 14 13:46:15 2012 via crm_attribute on pm1
> Stack: Heartbeat
> Current DC: pm1 (c610089f-e444-43fb-a0fc-f3d737651925) - partition with
> quorum
> Version: 1.1.6-31f6ca3
> 1 Nodes configured, 2 expected votes
> 1 Resources configured.
> ============
>
> Online: [ pm1 ]
>
> Full list of resources:
>
> stonith1        (stonith:external/libvirt):     Started pm1
>
> Node Attributes:
> * Node pm1:
>
> Migration summary:
> * Node pm1:
>
>
> I executed the following commands to produce stonith.
> # crm node fence pm1
>
> However, pm1 did not commit suicide.
>
> Because after confirming crmd/te_action.c, size of the 134th line
> finalized_nodes becomes 0, a flag does not seem to stand.
> 134     if (finalized_nodes && g_hash_table_size(finalized_nodes) == 1) {
> 135         options |= st_opt_allow_suicide;
> 136     }
> Should not size of confirmed_nodes be checked here?

You mean instead of finalized_nodes?
I think you're right, I appear to have gotten the two hash tables mixed up.

Did you happen to test with such a patch to confirm it works as expected?


> Or is how to handle me wrong?
> Please teach right how to use if wrong.
>
> Best Regards,
> Yuusuke
> --
> ----------------------------------------
> METRO SYSTEMS CO., LTD
>
> Yuusuke Iida
> Mail: iidayuus at intellilink.co.jp
> ----------------------------------------