There is a chance this will actually work without any changes in pcs as we do not care if the specified resource is a bundle. We will see once pacemaker part is ready.

I tried restarting bundles with pacemaker-1.1.16-12.el7_4.4 installed both in the cluster machines and in containers. "crm_resource --restart <bundle>" seems to be working fine. However, "crm_resource --restart <bundle> --node <node>" seems to be restarting the bundle on all nodes.

# crm_resource --restart --resource httpd-bundle

Set 'httpd-bundle' option: id=httpd-bundle-meta_attributes-target-role set=httpd-bundle-meta_attributes name=target-role=stopped
Waiting for 1 resources to stop:
 * httpd-bundle
Deleted 'httpd-bundle' option: id=httpd-bundle-meta_attributes-target-role name=target-role
Waiting for 1 resources to start again:
 * httpd-bundle

# crm_resource --restart --resource httpd-bundle --node rh74-node1

Set 'httpd-bundle' option: id=httpd-bundle-meta_attributes-target-role set=httpd-bundle-meta_attributes name=target-role=stopped
Waiting for 1 resources to stop:
 * httpd-bundle
Deleted 'httpd-bundle' option: id=httpd-bundle-meta_attributes-target-role name=target-role
Waiting for 1 resources to start again:
 * httpd-bundle

While this is happening, I can see the bundle resource stopping and starting on all nodes in both cases.


Ken, is the bundle restart supposed to work with the --node option? If the option is not going to be supported, nothing has to be done in pcs, as 'pcs resource restart httpd-bundle' already works. In the other case, a small fix in pcs is needed to allow specifying a node when restarting a bundle.

Grabbing this as a pacemaker bug because nothing is needed from pcs.

I think crm_resource --restart should treat bundles the same way as clones, i.e. ban the resource from the specified node rather than set target-role to Stopped. This should be a fairly simple change we can get into 7.5.

(In reply to Ken Gaillot from comment #4)
> Grabbing this as a pacemaker bug because nothing is needed from pcs.

That depends. Pcs currently does not allow specifying a node when restarting bundles:

# pcs resource restart httpd-bundle rh74-node1
Error: can only restart on a specific node for a clone or master/slave resource

As I said, a small fix in pcs is needed to allow specifying a node when restarting a bundle.


If a node is not specified, everything works with current pcs:

# pcs resource restart httpd-bundle
httpd-bundle successfully restarted

Ah, I missed that. We can clone this bz then, but let me make sure my idea works first.

Fixed upstream as of commit d6eb1cb4

When restarting a bundle with multiple replicas on a single node, crm_resource will now do so by banning the bundle from that node (same as it would with a clone). Note that if there are any free nodes available to run the instance, the instance may begin to start there, but once crm_resource notices it stopped on the original node, it will remove the ban and the instance will likely move back to the original node (same as would happen with a clone with clone-max less than the number of available nodes).

I noticed "crm_resource --wait" does not return after restarting a bundle, but that appears to be a separate issue, not affecting this bz.

Also, for completeness, it is likely there are other places where tools should treat bundles like clones. That also will be investigated separately from this bz.

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHEA-2018:0860