groups are hardwired to restart children >N when child N restarts
if you want the behavior described, don't use a group... specify the
resources and constraints manually.
On Nov 13, 2007, at 2:26 AM, Aníbal Monsalve Salazar wrote:
We are able to successfully trigger reloads for HA resources (by
changing the value of a non-unique RA parameter), provided they are
not
in a group. Once resources are grouped, changing the same non-unique
parameter always results in a full restart of the resource, and all
dependent resources. Here's a simple demonstration of this problem:
1) Edit /usr/lib/ocf/resource.d/heartbeat/Dummy and add a non-unique
parameter so you can trigger a reload:
--- Dummy.orig 2007-11-12 14:45:21.378089227 +1100
+++ Dummy 2007-11-12 14:15:26.121908269 +1100
@@ -57,6 +57,12 @@
<content type="string"
default="${HA_RSCTMP}/Dummy-{OCF_RESOURCE_INSTANCE}.state" />
</parameter>
+<parameter name="reload-trigger" unique="0">
+<longdesc lang="en">Change this to trigger a reload</longdesc>
+<shortdesc lang="en">Change this to trigger a reload</shortdesc>
+<content type="string"/>
+</parameter>
+
</parameters>
2) Create a standalone resource, a group containing two resources,
and appropriate constraints:
<resources>
<primitive id="test-reload-primitive" class="ocf"
provider="heartbeat" type="Dummy"/>
<group id="test-group">
<primitive id="test-reload-grouped-1" class="ocf"
provider="heartbeat" type="Dummy"/>
<primitive id="test-reload-grouped-2" class="ocf"
provider="heartbeat" type="Dummy"/>
</group>
</resources>
<constraints>
<rsc_location rsc="test-reload-primitive">
<rule score="INFINITY">
<expression attribute="#uname" operation="eq"
value="node-0"/>
</rule>
</rsc_location>
<rsc_location rsc="test-group">
<rule score="INFINITY">
<expression attribute="#uname" operation="eq"
value="node-0"/>
</rule>
</rsc_location>
</constraints>
3) Try to trigger a reload of each of the three resources:
# crm_resource --resource "test-reload-primitive" --set-
parameter \
"reload-trigger" --property-value "$(date --rfc-3339=ns)"
# crm_resource --resource "test-reload-grouped-1" --set-
parameter \
"reload-trigger" --property-value "$(date --rfc-3339=ns)"
# crm_resource --resource "test-reload-grouped-2" --set-
parameter \
"reload-trigger" --property-value "$(date --rfc-3339=ns)"
If you tail /var/log/messages while doing this, you'll see the
standalone resource is reloaded correctly:
Nov 12 14:36:38 node-0 Dummy[20400]: [20407]: ERROR: Reloading...
Nov 12 14:36:38 node-0 Dummy[20400]: [20408]: DEBUG:
test-reload-primitive reload : 0
The other two resources however do a full stop-and-start:
Nov 12 14:37:01 node-0 Dummy[20562]: [20570]: DEBUG:
test-reload-grouped-2 stop : 0
Nov 12 14:37:02 node-0 Dummy[20571]: [20578]: DEBUG:
test-reload-grouped-1 stop : 0
Nov 12 14:37:03 node-0 Dummy[20582]: [20589]: DEBUG:
test-reload-grouped-1 start : 0
Nov 12 14:37:04 node-0 Dummy[20590]: [20597]: DEBUG:
test-reload-grouped-2 start : 0
We need to be able to do a reload of individual resources within a
group. Is there anything extra we should be configuring in order to
allow this?
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems
