Description of problem: Instance attributes validation of fence_scsi agent via stonith_admin always fails. The problem is with validation of the ‘plug’ option which is required by the agent. Even the explicitly specified ‘plug’ or ‘pcmk_host_list’ options doesn’t help. From pcs point of view, stonith_admin should handle all of these cases as they are a valid configuration. Version-Release number of selected component (if applicable): [root@rhel91-devel1 pcs]# rpm -q pacemaker fence-agents-scsi pacemaker-2.1.4-5.el9.x86_64 fence-agents-scsi-4.10.0-30.el9.noarch How reproducible: always Steps to Reproduce: Note: all commands are executed on a single node cluster with node name ’localhost’. [root@rhel91-devel1 pcs]# /usr/sbin/stonith_admin --validate --agent fence_scsi Validation of fence_scsi failed 2022-09-15 11:25:16,694 ERROR: Failed: unable to parse output of corosync-cmapctl or node does not exist 2022-09-15 11:25:16,695 ERROR: Please use '-h' for usage [root@rhel91-devel1 pcs]# echo $? 1 [root@rhel91-devel1 pcs]# /usr/sbin/stonith_admin --validate --agent fence_scsi --option pcmk_host_list=localhost Validation of fence_scsi failed 2022-09-15 11:26:02,022 ERROR: Failed: unable to parse output of corosync-cmapctl or node does not exist 2022-09-15 11:26:02,022 ERROR: Please use '-h' for usage [root@rhel91-devel1 pcs]# echo $? 1 [root@rhel91-devel1 pcs]# /usr/sbin/stonith_admin --validate --agent fence_scsi --option plug=localhost Validation of fence_scsi failed 2022-09-15 11:26:16,965 ERROR: Failed: unable to parse output of corosync-cmapctl or node does not exist 2022-09-15 11:26:16,966 ERROR: Please use '-h' for usage [root@rhel91-devel1 pcs]# echo $? 1 [root@rhel91-devel1 pcs]# corosync-cmapctl | grep localhost nodelist.node.0.name (str) = localhost nodelist.node.0.ring0_addr (str) = localhost totem.interface.0.bindnetaddr (str) = localhost Actual results: Validation fails Expected results: Successful validation Additional info: Running just agent itself works ok: [root@rhel91-devel1 pcs]# fence_scsi -o validate-all --plug non-existing-node; echo $? 2022-09-15 12:44:54,360 ERROR: Failed: unable to parse output of corosync-cmapctl or node does not exist 2022-09-15 12:44:54,360 ERROR: Please use '-h' for usage 1 [root@rhel91-devel1 pcs]# fence_scsi -o validate-all --plug localhost; echo $? 0 From this snippet of strace output I assume that this is written to stdin of agent for validation: [root@rhel91-devel1 pcs]# strace /usr/sbin/stonith_admin --validate --agent fence_scsi --option plug=localhost <snip> write(8, "nodename=node1\n", 15) = 15 write(8, "action=validate-all\n", 20) = 20 write(8, "port=node1\n", 11) = 11 write(8, "plug=localhost\n", 15) = 15 <snip> So I tried to execute agent with the same stdin and indeed it fails with the same error: [root@rhel91-devel1 pcs]# echo -e "nodename=node1\naction=validate-all\nport=node1\nplug=localhost\n" | fence_scsi; echo $? 2022-09-15 13:01:36,710 ERROR: Failed: unable to parse output of corosync-cmapctl or node does not exist 2022-09-15 13:01:36,710 ERROR: Please use '-h' for usage 1 However, validation passes after removing “nodename=node1” from stdin, even though the “nodename” should not be accepted by the agent according to its man page. [root@rhel91-devel1 pcs]# echo -e "action=validate-all\nport=node1\nplug=localhost\n" | fence_scsi; echo $? 0 I didn’t try any other agent but I would guess that other agents may be affected as well.
This sounds like something the agent's doing. At the least its documentation doesn't seem to match what options are actually accepted If the main problem is that the target should be passed via "plug", then pcmk_host_argument=plug might help. Note that pacemaker will check the metadata and use port if advertised, otherwise plug if advertised, so this could be a metadata issue.
Well, pcmk_host_argument=plug doesn't solve anything because stonith_admin then sends "plug=node1" to the agent instead of real node name. I assume that the agent is checking plug value against node list from corosync, which is not an unreasonable check. But this check always fails as it gets a dummy node name from stonith_admin. If stonith_admin would send a real node name in 'nodename' option, fence_scsi validation will be successful. So I can see 2 options to fix this: * stonith_admin should send real node names to agent when running 'validate-all' action * agents should not check if plug (or port or however it's called in other agents) is a real cluster node
nodename is supported for backwards compatibility: https://github.com/ClusterLabs/fence-agents/blob/8b0ea39d49453c284d6ae708ec5f14c2bf867780/agents/scsi/fence_scsi.py#L560-L563 It's help text have been removed, so it's not advertised in manpage/metadata.