groupd - the group manager for fenced, dlm_controld and gfs_controld
The group daemon, groupd, provides a compatibility layer between the
corosync closed process group (CPG) service and the fenced,
dlm_controld and gfs_controld daemons. New versions of
fenced/dlm_controld/gfs_controld no longer require this compat layer,
and can run without groupd. They use libcpg directly, making them much
more robust. This makes them incompatible with old (cluster2)
versions. Explicitly enable the new incompatible mode with:
This is the default if no groupd_compat setting exists. If
compatibility with old (cluster2) nodes is necessary (mixing cluster2
and cluster3 nodes should be avoided), the daemons can be run in the
old mode, using groupd as before. Enable this mode with:
To perform a rolling upgrade from cluster2 to cluster3, add the
groupd_compat="1" setting to cluster.conf before upgrading any existing
nodes or adding any new cluster3 nodes. The next time the entire
cluster is shut down, after all nodes are upgraded, groupd_compat can
be changed to 0 (or removed entirely). Warning: groupd_compat must not
be changed from 1 to 0 in a running cluster (or from 0 to 1). cluster2
nodes are equivalent to groupd_compat="1", so it is fine to add
groupd_compat="1" to a running cluster2 cluster as described above to
perform a rolling upgrade.
A third option, groupd_compat="2", enables a special phase during
startup where the daemons attempt to discover cluster2 nodes or
cluster3 nodes running in mode 1. If any are found, they run in mode
1, otherwise mode 0. This mode will delay startup and makes cluster
operation less predictable.
The cman init script will not start the groupd daemon unless it is
needed, i.e. groupd_compat=1 (or 2).
By default, log messages at INFO level and above are sent to both
syslog(3) and /var/log/cluster/groupd.log. For complete logging
configuration options see cluster.conf(5).
-D Enable debugging to stderr and don’t fork
-L Enable debugging to log file (see above)
group compatibility mode, 0 off, 1 on, 2 detect
0: use libcpg, no backward compat, best performance
1: use libgroup for compat with cluster2/rhel5
2: detect old, or mode 1, nodes that require compat
seconds to wait for a node’s version message before assuming an
old version requiring compat mode
seconds to delay the mode selection to give time for an old
version to join and force compat mode
-V Print the version information and exit.
-h Print out a help message describing available options, then
The groupd daemon keeps a circular buffer of debug messages that can be
dumped with the ’group_tool dump’ command.