When switching to the pcs 0.10 providers we moved from the the general
pcmk_resource provider to the pcmk_remote backend. While doing this we
lost the ability to specify the reconnect_interval.
Tested by deploying tripleo with the following:
And correctly observing that the remotes have the right
[root@controller-0 ~]# pcs resource config compute-0 |grep reconnect
Attributes: reconnect_interval=180 server=172.17.1.96
Now that we branched we should try and clean up these old usages.
Tested this by deploying a large train composable HA overcloud (with
stonith and Instance HA) using puppet-pacemaker from master with this
change and everything worked correctly.
Test was done on RHEL 8.3 + composable HA + IHA + TLS-E
New pcs 0.10 needs to implement remotes addition in a completely
different way. It is not sufficient anymore to just create the
ocf::pacemaker::remote resource and make sure that pacemaker_remote is
up and running with the same authkey on all nodes. Now we
need to setup pcsd on all nodes (remotes and controllers), setup the
hacluster user and then do a pcs cluster remote-add command.
We do this only when pcs 0.10 is detected in order to maintain
the old proven behaviour on pcs 0.9x.
Tested this on stein + rhel8 correctly get remotes working on an IHA
[root@controller-0 galera-bundle-1]# pcs status
Cluster name: tripleo_cluster
Current DC: controller-1 (version 2.0.1-4.el8-0eb7991564) - partition with quorum
Last updated: Thu May 23 05:52:33 2019
Last change: Tue May 21 14:20:14 2019 by root via cibadmin on controller-0
17 nodes configured
72 resources configured
Online: [ controller-0 controller-1 controller-2 ]
RemoteOnline: [ compute-0 compute-1 ]
Full list of resources:
compute-0 (ocf::pacemaker:remote): Started controller-0
compute-1 (ocf::pacemaker:remote): Started controller-1
Also tested on queens to make sure no regressions in the old codepaths
are introduced. Additionally tested a redeploy on stein to make sure
idempotency is preserved.