[ClusterLabs Developers] FYI possibly incompatible change in Corosync CPG service in master
jfriesse at redhat.com
Thu Jun 13 09:49:17 EDT 2019
I would like to inform all developers who is using Corosync CPG service
about possibly "incompatible" change in Corosync master branch.
Previously CPG sent joinlist always contained only 1 member. This
behavior was not entirely correct so we fixed that so now only 1 CPG
join list confchg event with all joined members is sent by node/group.
Such change doesn't break API, because confchg callback parameter is
array of joined members, and it was never guaranteed that callback will
always contain only 1 member. So there is no .so version bump planned.
I'm pretty sure applications using CPG are well written and are able to
handle described change without any problems, but better safe than sorry.
Plan is to have this change included with next stable Camelback release
(3.0.3 - in about half of a year). There are no plans to backport this
change into Needle (2.x) branch.
So if you are developer (or passionate user) of software which is using
Corosync CPG then I would like to ask you to give a test to current
master (3c7f19a02fbff19414638952043691d92be86d41) and fix/report
problems you've found. Currently Pacemaker (thanks Ken for testing) and
Proxmox (Fabian is author of the patch) are known to work well.
More information about the Developers