Support starting corosync-notifyd at cluster startup (draft)

[inouekazu]

This is a draft of a patch to allow corosync-notifyd to also be started with pcs cluster start.

• pcs cluster enable-corosync-notifyd will set ENABLE_COROSYNC_NOTIFYD=yes in /etc/sysconfig/pcs.

  # pcs cluster enable-corosync-notifyd
  Distributing PCS config...
  rhel84-1: PCS config saved
  rhel84-2: PCS config saved
  

  # ssh rhel84-1 cat /etc/sysconfig/pcs
  # This file has been generated by pcs.
  ENABLE_COROSYNC_NOTIFYD=yes
  
  # ssh rhel84-2 cat /etc/sysconfig/pcs
  # This file has been generated by pcs.
  ENABLE_COROSYNC_NOTIFYD=yes
  

• If this setting is present, corosync-notifyd will also be invoked when the cluster is started.

  # pcs cluster start --all
  rhel84-1: Starting Cluster...
  rhel84-2: Starting Cluster...
  
  # ssh rhel84-1 ps -ef | grep -e corosync -e pacemaker
  root        7491       1  5 14:50 ?        00:00:02 /usr/sbin/corosync -f
  root        7504       1  0 14:50 ?        00:00:00 /usr/sbin/corosync-notifyd -f -l -s -m 192.168.122.160
  root        7506       1  0 14:50 ?        00:00:00 /usr/sbin/pacemakerd -f
  haclust+    7507    7506  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-based
  root        7508    7506  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-fenced
  root        7509    7506  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-execd
  haclust+    7510    7506  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-attrd
  haclust+    7511    7506  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-schedulerd
  haclust+    7512    7506  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-controld
  
  # ssh rhel84-2 ps -ef | grep -e corosync -e pacemaker
  root        6778       1  1 14:50 ?        00:00:03 /usr/sbin/corosync -f
  root        6791       1  0 14:50 ?        00:00:00 /usr/sbin/corosync-notifyd -f -l -s -m 192.168.122.160
  root        6793       1  0 14:50 ?        00:00:00 /usr/sbin/pacemakerd -f
  haclust+    6794    6793  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-based
  root        6795    6793  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-fenced
  root        6796    6793  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-execd
  haclust+    6797    6793  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-attrd
  haclust+    6798    6793  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-schedulerd
  haclust+    6799    6793  0 14:50 ?        00:00:00 /usr/libexec/pacemaker/pacemaker-controld
  
  # pcs status
  Cluster name: my_cluster
  
  WARNINGS:
  No stonith devices and stonith-enabled is not false
  
  Cluster Summary:
    * Stack: corosync
    * Current DC: rhel84-2 (version 2.0.5-9.el8_4.1-ba59be7122) - partition with quorum
    * Last updated: Wed May 26 15:03:38 2021
    * Last change:  Wed May 26 14:27:34 2021 by hacluster via crmd on rhel84-2
    * 2 nodes configured
    * 0 resource instances configured
  
  Node List:
    * Online: [ rhel84-1 rhel84-2 ]
  
  Full List of Resources:
    * No resources
  
  Daemon Status:
    corosync: active/disabled
    corosync-notifyd: active/disabled
    pacemaker: active/disabled
    pcsd: active/enabled
  

• pcs cluster disable-corosync-notifyd will set ENABLE_COROSYNC_NOTIFYD=no in /etc/sysconfig/pcs. In this case, corosync-notifyd will not be started.

[knet-ci-bot]

Can one of the admins verify this patch?

[tomjelinek]

Thank you for this pull request and your effort to make it fit pcs architecture. I read through it and found several points to improve.

Corosync-notifyd has not been added to pcs cluster enable and pcs cluster disable commands. That means corosync-notifyd will only get started and stopped by pcs. Even though that may be sufficient for your use case, others may expect corosync-notifyd to be started after boot as well.

I don't think /etc/sysconfig/pcs is a good place to store ENABLE_COROSYNC_NOTIFYD settings. Files in /etc/sysconfig hold environment variables passed to daemons on their start. I perceive them as files modified once or rarely, usually by an admin. Also, their syntax does not make them ideal to modify them by code. On top of that, we already have several config files in pcs and we would like to avoid adding new ones.

If the setting needs to be stored, then /var/lib/pcs_settings.conf is better place to do it. The advantage is the file is automatically synchronized by pcsd. The disadvantage, for now, is that all code managing the file and synchronization is written in ruby and not easily accessible from python.

Alternatively, an option like --with-corosync-notifyd could be added to pcs cluster start | stop | enable | disable commands instead of having the setting stored in a file.

[inouekazu]

Thanks for reviewing it.

Even though that may be sufficient for your use case, others may expect corosync-notifyd to be started after boot as well.

Yes, you're right. I'll deal with it.

If the setting needs to be stored, then /var/lib/pcs_settings.conf is better place to do it.

OK! I'll look into it.

[inouekazu]

I requested a patch (#311) that addresses the issues pointed out.