Description of problem:
While the latest pacemaker in rhel-7.4 (1.1.16-12) has a pretty much working
bundle implementation, we found a number of issues around bundles that need
fixing for OSP12. Namely:
- Unfencing bundles
- Nested container connections
- Notify not working inside bundles
- container-attribute-target support (to let RAs decide where to store attributes)
- Colocation bundle issues
The fixes for the above have already been developed by Andrew and are all
present in the package we have tested for OSP12 in the last weeks:
This BZ is mainly to make sure we get all these fixes released for OSP12
This bug is not raised correctly. In order to have a zstream bug we need to clone from Y stream bug, we cannot raise directly a zstream bug.
Please check the zstream process documentation, https://mojo.redhat.com/docs/DOC-1021938#jive_content_id_The_workflow
yeah did this by mistake, undone. Thanks
FYI the unfencing part has its own Bug 1394418
Additional patches in http://people.redhat.com/mbaldess/rpms/container-repo/srpms/pacemaker-1.1.16-12.13.el7.src.rpm
+ Andrew Beekhof (31 hours ago) 2709e5c: Fix: PE: Do not send notifications to unclean bundles
+ Andrew Beekhof (31 hours ago) e3485d6: Fix: PE: Prevent graph loops when fencing the host out from underneath a bundle
These are all fixed upstream
(Release note work)
It turns out bundles remain in tech. preview for 7.5, but I don't think that affects this release note writeup.
I downgraded the all-caps in your original note, since that's a little bit out of style. We can move this note to the top of the new feature list for clustering and it now has a title which should cause anybody who uses guest nodes to at least notice it and it begins with the isolated sentence about this being important (without being in all-caps).
Do you think that's sufficient?
(For release note: We don't capitalize "Remote Node" in the docs so I uncapitalized it here.)
The new text looks good, except for the Pacemaker Remote capitalization. I find the Pacemaker Remote terminology unfortunate, but it's what we have to work with ...
* cluster node = node running the full cluster stack including corosync
* remote node = node running pacemaker_remote and integrated into the cluster via an ocf:pacemaker:remote resource
* guest node = node running pacemaker_remote and integrated into the cluster via a resource such as VirtualDomain with the remote-node parameter set
There's no accepted term for "any node running pacemaker_remote", aka "remote node or guest node", so I've been using "Pacemaker Remote node". Feel free to use something else, but the intention is not the same as "remote node".
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.