[ClusterLabs] Coming in Pacemaker 3.0.1: improvements to daemon IPC

[Chris Lumens]

If you've ever seen the log message "consider setting PCMK_ipc_buffer to
XYZ or higher", this will be of interest to you.  For everyone else, it
might be a little esoteric.

Prior to Pacemaker 3.0.1, large IPC messages would be compressed if they
were too large, and if they were still too large, the above message
would be logged.  The point of this message was to prompt the admin to
change the PCMK_ipc_buffer value in /etc/sysconfig/pacemaker.  But of
course, you had to actually go to the logs to figure this out.

The most common way for you to see that message would be if you had a
very large CIB due to a lot of resources, nodes, or resource history.

Starting with Pacemaker 3.0.1, the pacemaker daemons will instead split
up a too-large IPC message into several smaller ones.  This removes the
need for any compression, and makes the PCMK_ipc_buffer setting
obsolete.

If you've never touched that setting, there's no need to do anything.
If you have, you still don't need to do anything - pacemaker will simply
ignore it now.

I expect this will help a small number of installations, but there are
still problems with very active clusters with very large numbers of
resources (> 200) getting bogged down due to IPC problems.  That's next
on my list of things to address, but it won't be coming in 3.0.1.

- Chris