[ClusterLabs] fence_scsi won't start

[Marek "marx" Grac]

Hi,


On 04/27/2015 12:18 PM, Duncan Mortimer wrote:
> Hi,
>
> I’m trying to setup a simple two host Apache cluster on Centos 7 using a shared iSCSI backing store (hosted on a third Centos 7 machine).
>
> Following several how tos I have created two iSCSI devices, one for the shared filesystem and one for fencing. I have created /var/target/pr (for persistent reservation support) and confirmed that both cluster hosts can see both iSCSI targets (both read and write).
It looks like the problem is explained there:

Apr 27 10:54:08 [12123] wwwa stonith-ng:     info: update_remaining_timeout: 	Attempted to execute agent fence_scsi (list) the maximum number of times (2) allowed
Apr 27 10:54:08 [12123] wwwa stonith-ng:   notice: dynamic_list_search_cb: 	Disabling port list queries for iscsi-stonith-device (1): Failed: nodename or key is required

Please use '-h' for usage
Apr 27 10:54:08 [12123] wwwa stonith-ng:   notice: remote_op_done: 	Operation on of wwwa by <no-one> forcrmd.46074 at wwwb.4cb65ea2: No such device
Apr 27 10:54:08 [12127] wwwa       crmd:    error: tengine_stonith_notify: 	Unfencing of wwwa by <anyone> failed: No such device (-19)
Apr 27 10:54:08 [12124] wwwa       lrmd:     info: process_lrmd_get_rsc_info: 	Resource 'iscsi-stonith-device' not found (0 active resources)
Apr 27 10:54:08 [12124] wwwa       lrmd:     info: process_lrmd_rsc_register: 	Added 'iscsi-stonith-device' to the rsc list (1 active resources)


Fence_scsi is using unique keys to identify machines that have 
reservations. These keys are either manually selected (key / --key on 
commandline) or automatically derived from cluster name & node name. But 
in order to find out it automatically, we need to know which node name 
should be impacted by fence/unfence action. Adding one of these option 
should help.


m,