Hi Olaf,
Dne 05. 01. 26 v 20:20 Faaland, Olaf P. via Users napsal(a):
Hi,
I’m starting to use stonith levels.
My existing (1-level stonith config) looks like this:
#pcs stonith config
Resource: fence_pm (class=stonith type=fence_powerman)
Attributes: fence_pm-instance_attributes
ipaddr=localhost
ipport=10101
pcmk_delay_base=60s
pcmk_delay_max=90s
pcmk_host_check=static-list
pcmk_host_list=gopher1,gopher2,gopher3,gopher4,gopher5,gopher6,gopher7,gopher8,gopher9,gopher10
Operations:
monitor: fence_pm-monitor-interval-60s
interval=60s
And I want to switch to a 2-level stonith config. I’d expected these
commands to work:
lustre_servers=gopher1,gopher2,gopher3,gopher4,gopher5,gopher6,gopher7,gopher8,gopher9,gopher10
# set up new fence agent
pcs stonith create fence_kdump fence_kdump pcmk_host_check=static-list
pcmk_host_list=${lustre_servers}
# set levels (priority) so pacemaker knows which to use first
pcs stonith level add 1${lustre_servers}fence_kdump
pcs stonith level add 2${lustre_servers}fence_pm
# verify config
pcs stonith level verify
The “pcs stonith create” succeeds, but the “pcs stonith level add”
fails. Looking at the man page, it looks like a comma-separated
target list isn’t allowed for “pcs stonith level add”. Is that correct?
It would be helpful to see the error messages pcs printed.
Anyway, the target is supposed to be a single value, e.g. a node name,
as the man page describes. I don't think a list has ever been allowed there.
Since your fence_kdump and fence_pm handle all cluster nodes, you can
use them for all your nodes. Meaning you need to run those two commands
for every node, unless you decide to specify nodes using regular
expressions or node attributes.
In the man page, I see
Target may be a node name <node_name> or %<node_name> or
node%<node_name>, a node name regular expression
regexp%<node_pattern>
or a node attribute value attrib%<name>=<value>.
Is “%” literal, here?
yes
Do
“pcs stonith level add <level> node%<node_name> <stonith_id>” and
“pcs stonith level add <level> %<node_name> <stonith_id>” and
“pcs stonith level add <level> <node_name> <stonith_id>” mean
something different? They seem identical from brief experiments.
They are the same. If type (the part in front of %) is not specified, it
defaults to 'node'.
How is “attrib%<name>=<value>” used?
This makes the level apply to nodes with the specified attribute name
and value. It is documented at
https://clusterlabs.org/projects/pacemaker/doc/3.0/Pacemaker_Explained/html/fencing.html#fencing-topologies
. Node attributes are managed using 'pcs node attribute' command.
Regards,
Tomas
Thanks,
Olaf
_______________________________________________
Manage your subscription:
https://lists.clusterlabs.org/mailman/listinfo/users
ClusterLabs home:https://www.clusterlabs.org/
_______________________________________________
Manage your subscription:
https://lists.clusterlabs.org/mailman/listinfo/users
ClusterLabs home: https://www.clusterlabs.org/
