Thomas Glanzmann schrieb:
Hello Martin,
it is pure luck that I am so bored that I read this list, next time CC
me. :-)
I have read several postings in the mail archive about the
external/ipmi configuration but there are still some questions that
bother me. The last posting from Thomas: did this cib-configuration
worked with your 2-node cluster? I have to configure also 2 nodes and
would like to use the ipmi-plugin but I am unsure if I understand what
the plugin really does.
I have the following configuration on two systems and I verified that
this configuration works as it should be. Someone on this list told me
that I can drop the location constraints, however I decided to keep them
until I verified that.
<configuration>
<crm_config>
<cluster_property_set id="cib-bootstrap-options">
<attributes>
<nvpair name="stonith-enabled" value="true"
id="stonith-enabled"/>
<nvpair name="stonith-action" value="reboot"
id="stonith-action"/>
</attributes>
</cluster_property_set>
</crm_config>
<resources>
<primitive id="apache-01-fencing" class="stonith" type="external/ipmi"
provider="heartbeat">
<operations>
<op id="apache-01-fencing-monitor" name="monitor" interval="60s"
timeout="20s" prereq="nothing"/>
<op id="apache-01-fencing-start" name="start" timeout="20s"
prereq="nothing"/>
</operations>
<instance_attributes id="ia-apache-01-fencing">
<attributes>
<nvpair id="apache-01-fencing-hostname"
name="hostname" value="apache-01"/>
<nvpair id="apache-01-fencing-ipaddr" name="ipaddr"
value="172.18.0.101"/>
<nvpair id="apache-01-fencing-userid" name="userid"
value="Administrator"/>
<nvpair id="apache-01-fencing-passwd" name="passwd"
value="whatever"/>
</attributes>
</instance_attributes>
</primitive>
<primitive id="apache-02-fencing" class="stonith" type="external/ipmi"
provider="heartbeat">
<operations>
<op id="apache-02-fencing-monitor" name="monitor" interval="60s"
timeout="20s" prereq="nothing"/>
<op id="apache-02-fencing-start" name="start" timeout="20s"
prereq="nothing"/>
</operations>
<instance_attributes id="ia-apache-02-fencing">
<attributes>
<nvpair id="apache-02-fencing-hostname"
name="hostname" value="apache-02"/>
<nvpair id="apache-02-fencing-ipaddr" name="ipaddr"
value="172.18.0.102"/>
<nvpair id="apache-02-fencing-userid" name="userid"
value="Administrator"/>
<nvpair id="apache-02-fencing-passwd" name="passwd"
value="whatever"/>
</attributes>
</instance_attributes>
</primitive>
</resources>
<constraints>
<rsc_location id="apache-01-fencing-placement"
rsc="apache-01-fencing">
<rule id="apache-01-fencing-placement-rule-1"
score="-INFINITY">
<expression id="apache-01-fencing-placement-exp-02" value="apache-02"
attribute="#uname" operation="ne"/>
</rule>
</rsc_location>
<rsc_location id="apache-02-fencing-placement"
rsc="apache-02-fencing">
<rule id="apache-02-fencing-placement-rule-1"
score="-INFINITY">
<expression id="apache-02-fencing-placement-exp-02" value="apache-01"
attribute="#uname" operation="ne"/>
</rule>
</rsc_location>
</constraints>
</configuration>
I killed heartbeat with -9 to simulate a node failure.
To configure the plugin, I will create a resource for every node. This
means, two additional resources in my cib.xml because I have two
cluster-nodes.
Correct.
The attributes (nvpair) define variables for the ipmi-script, e.g.
hostname... But what does the constraints tell me? If "#uname" is not
equal to"value" then the score ist -INFINITY, i.e. the resource will
never be started on that node?
you pin ,,apache-01-fencing'' on apache-02 and ,,apache-02-fencing'' on
apache-01. So that the resource that can stonith apache-01 runs on apache-02
and vice versa. Someone stated that heartbeat is able to do a suicide
(stonith itself) but that isn't true at least not via stonith and not in
version 2.1.3. The location constraints seems to be unneccessary because
if the fencing is running on the wrong node and that node misbehaves, it
is restartet on the remainding node and than shoots the misbehaving.
However they don't hurt either. So I leave it exactly in the above
configuration because it proofed to work reliable in real live.
In summary, how does the fencing work? The heartbeat-process gets
killed on one node, so all resources of that node are failed over to
the other node. So far so good, but what role plays the new resource
in the failover? Is this resource also failed over to the other node
an therefore executes the ipmi-script with a "reset"-option to its
"origin"-node?
Nope. The location constraints makes it impossible to fail over the
stonith resource, so it won't be started until the dead node returns to
live. But all other resources migrate over.
I still do not know how this stonith-plugin works exactly, so I have
some problems configuring heartbeat. I hope someone can help me with
these questions, because I want to understand how the STONITH works
instead of just copy and paste some code-snippets.
If a node is misbehaving eg. stop talking to me without saying ,,bye''
it gets shot in the head to make sure it is really dead. After that the
resources that were running before on the other node migrate over.
Thomas
Ok, I just implemented the external/ipmi STONITH-plugin and it worked
fine, thank you very much for the help.
But I encountered the following problem: my STONITH-resources don't
start anymore and as long as they are not running, STONITH won't work.
They exit with returncode 1, which isn't really helpful.
node2-fencing_start_0 (node=node1, call=29, rc=1): Error
Is the returncode coming from the ipmi-script?
How can I initiate a start of the resource via command line, with
crm_resource I can only change the value of the resource or am I wrong?
Greetings,
Martin
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
