[Pacemaker] Does "stonith_admin --confirm" work?

[Andrew Beekhof]

On 17/05/2013, at 6:22 PM, Староверов Никита Александрович <nastaroverov at kola.so-ups.ru> wrote:

> Hello, pacemaker users and developers.
> 
> First, many thanks to clusterlabs.org for their software, Pacemaker helps us very much!
> 
> I am testing cluster configuration based on Pacemaker+CMAN. I configured fencing as described in  Pacemker documentation about CMAN based clusters and it works.
> May be I misunderstood something, but I can't  acknowledge nodes fencing manually.
> I use fence_ipmilan as device and when I plug out power cable from server stonith  fails. I expected this,  of course, but I don't know how to acknowledge manual fencing.
> When I try stonith_admin -C node_name, it does nothing. 
> I see this in logs:
> 
> May 17 11:46:52 NODE1 stonith-ng[5434]:   notice: stonith_manual_ack: Injecting manual confirmation that NODE2 is safely off/down
> May 17 11:46:52 NODE1 stonith-ng[5434]:   notice: log_operation: Operation 'off' [0] (call 2 from stonith_admin.10959) for host 'NODE2' with device 'manual_ack' returned: 0 (OK)
> May 17 11:46:52 NODE1 stonith-ng[5434]:    error: crm_abort: do_local_reply: Triggered assert at main.c:241 : client_obj->request_id                                                          
> May 17 11:46:52 NODE1 stonith-ng[5434]:    error: crm_abort: crm_ipcs_sendv: Triggered assert at ipc.c:575 : header->qb.id != 0                                                               
> May 17 11:47:35 NODE1 stonith_admin[11162]:   notice: crm_log_args: Invoked: stonith_admin -C NODE2                                                                              
> May 17 11:47:35 NODE1 stonith-ng[5434]:   notice: merge_duplicates: Merging stonith action off for node NODE2 originating from client stonith_admin.11162.b42172b1 with identical request from stonith_admin.10959 at NODE1.f2048550 (0s)                                                                                                                                                                                                             
> May 17 11:47:35 NODE1 stonith-ng[5434]:   notice: stonith_manual_ack: Injecting manual confirmation that NODE2 is safely off/down                                                                                  
> May 17 11:47:35 NODE1 stonith-ng[5434]:   notice: log_operation: Operation 'off' [0] (call 2 from stonith_admin.11162) for host 'NODE2' with device 'manual_ack' returned: 0 (OK)                                  
> May 17 11:47:35 NODE1 stonith-ng[5434]:    error: crm_abort: do_local_reply: Triggered assert at main.c:241 : client_obj->request_id                                                                                            
> May 17 11:47:35 NODE1 stonith-ng[5434]:    error: crm_abort: crm_ipcs_sendv: Triggered assert at ipc.c:575 : header->qb.id != 0    

Well, thats not nothing, but it certainly doesn't look right either.
I will investigate.  Which version is this?

> 
> Nothing happened after stonith_admin -C.
> Fenced still trying to fence_pcmk, and I see lots of "Timer expired" from stonith-ng, and failed fence_ipmilan operations.
> 
> Yes,  I can do fence_ack_manual on cman-master node, and then cleanup node state with cibadmin, but it is very sloooow way. 
> If I lost many servers in cluster, for example, lost power in one rack with two or more servers, I need a way to running again services on remaining nodes as quickly as possible.
> 
> I think fencing manual acknowledgement must be fast and simple and I suppose that stonith_admin --confirm have to do that.
> 
> 
> 
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
> 
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://bugs.clusterlabs.org