[Pacemaker] Service failback issue with SLES11 and HAE 11

ben180 benchiu180 at gmail.com
Tue Jun 8 02:00:37 UTC 2010


Dear all,

There are two nodes in my customer's environment. We installed SuSE
Linux Enterprise Server 11 and HAE on the two node. The cluster is for
oracle database service HA purpose.
We have set clone resource for pingd, and constraints for detecting if
network interface is down. And we added a resource group which
includes Filesystem, IPAddr, oracle, oralsnr primitives, and use sbd
for fence device.

There's our settings :

- <resources>
- <clone id="Connected">
- <meta_attributes id="Connected-meta_attributes">
  <nvpair id="nvpair-c25cd67c-f681-4652-8007-64c0e50fabe6"
name="clone-max" value="2" />
  <nvpair id="nvpair-28b5d065-7984-424d-b5a8-fb8fc7b2f6dc"
name="target-role" value="Started" />
  </meta_attributes>
- <primitive class="ocf" id="ping" provider="pacemaker" type="pingd">
- <meta_attributes id="ping-meta_attributes">
  <nvpair id="nvpair-e7f1cec6-f5a7-4db2-b20b-0002ad31d9fa"
name="target-role" value="Started" />
  </meta_attributes>
- <operations id="ping-operations">
  <op id="ping-op-monitor-10" interval="10" name="monitor"
start-delay="1m" timeout="20" />
  </operations>
- <instance_attributes id="ping-instance_attributes">
  <nvpair id="nvpair-cab0ee56-cd6c-47b7-b206-a19bce16d445"
name="dampen" value="5" />
  <nvpair id="nvpair-c660e572-40ba-4166-9293-6e99f5d024e8"
name="host_list" value="10.224.1.254" />
  </instance_attributes>
  </primitive>
  </clone>
- <group id="DB">
- <meta_attributes id="DB-meta_attributes">
  <nvpair id="nvpair-a0ce4033-555a-40c3-8e92-191552596a97"
name="target-role" value="Started" />
  </meta_attributes>
- <primitive class="ocf" id="FileSystem" provider="heartbeat" type="Filesystem">
- <meta_attributes id="FileSystem-meta_attributes">
  <nvpair id="nvpair-6e46d65a-86d4-41d4-9c7f-7ea502ca9f36"
name="target-role" value="started" />
  </meta_attributes>
- <operations id="FileSystem-operations">
  <op id="FileSystem-op-monitor-20" interval="20" name="monitor"
start-delay="10" timeout="40" />
  </operations>
- <instance_attributes id="FileSystem-instance_attributes">
  <nvpair id="nvpair-99da66a3-ebdf-4c3b-9647-05a065ff8309"
name="device" value="/dev/dm-0" />
  <nvpair id="nvpair-c882d532-3fc5-41a4-b1a3-6b03b2b3d54d"
name="directory" value="/oracle" />
  <nvpair id="nvpair-643ad766-eb95-4667-8b33-452f8266ba10"
name="fstype" value="ext3" />
  </instance_attributes>
  </primitive>
- <primitive class="ocf" id="ServiceIP" provider="heartbeat" type="IPaddr">
- <meta_attributes id="ServiceIP-meta_attributes">
  <nvpair id="nvpair-03afe5cc-226f-43db-b1e5-ee2f5f1cb66e"
name="target-role" value="Started" />
  </meta_attributes>
- <operations id="ServiceIP-operations">
  <op id="ServiceIP-op-monitor-5s" interval="5s" name="monitor"
start-delay="1s" timeout="20s" />
  </operations>
- <instance_attributes id="ServiceIP-instance_attributes">
  <nvpair id="nvpair-10b45737-aa05-4a7f-9469-b1f75e138834" name="ip"
value="10.224.1.138" />
  </instance_attributes>
  </primitive>
- <primitive class="ocf" id="Instance" provider="heartbeat" type="oracle">
- <meta_attributes id="Instance-meta_attributes">
  <nvpair id="nvpair-2bbbe865-1339-4cbf-8add-8aa107736260"
name="target-role" value="Started" />
  <nvpair id="nvpair-6cfd1675-ce77-4c05-8031-242ed176b890"
name="failure-timeout" value="1" />
  <nvpair id="nvpair-e957ff0a-c40e-494d-b691-02d3ea67440b"
name="migration-threshold" value="1" />
  <nvpair id="nvpair-4283547a-4c34-4b26-b82f-50730dc4c4fa"
name="resource-stickiness" value="INFINITY" />
  <nvpair id="nvpair-9478fb7c-e1aa-405e-b4a9-c031971bc612"
name="is-managed" value="true" />
  </meta_attributes>
- <operations id="Instance-operations">
  <op enabled="true" id="Instance-op-monitor-120" interval="30"
name="monitor" role="Started" start-delay="1m" timeout="240" />
  </operations>
- <instance_attributes id="Instance-instance_attributes">
  <nvpair id="nvpair-30288e1c-e9e9-4360-b658-045f2f353704" name="sid"
value="BpmDBp" />
  </instance_attributes>
  </primitive>
- <primitive class="ocf" id="Listener" provider="heartbeat" type="oralsnr">
- <meta_attributes id="Listener-meta_attributes">
  <nvpair id="nvpair-f6219b53-5d6a-42cb-8dec-d8a17b0c240c"
name="target-role" value="Started" />
  <nvpair id="nvpair-ae38b2bd-b3ee-4a5a-baec-0a998ca7742d"
name="failure-timeout" value="1" />
  </meta_attributes>
- <operations id="Listener-operations">
  <op id="Listener-op-monitor-10" interval="10" name="monitor"
start-delay="10" timeout="30" />
  </operations>
- <instance_attributes id="Listener-instance_attributes">
  <nvpair id="nvpair-96615bed-b8a1-4385-a61c-0f399225e63e" name="sid"
value="BpmDBp" />
  </instance_attributes>
  </primitive>
  </group>
- <clone id="Fence">
- <meta_attributes id="Fence-meta_attributes">
  <nvpair id="nvpair-59471c10-ec9d-4eb8-becc-ef6d91115614"
name="clone-max" value="2" />
  <nvpair id="nvpair-6af8cf4a-d96f-449b-9625-09a10c206a5f"
name="target-role" value="Started" />
  </meta_attributes>
- <primitive class="stonith" id="sbd-stonith" type="external/sbd">
- <meta_attributes id="sbd-stonith-meta_attributes">
  <nvpair id="nvpair-8c7cfdb7-ec19-4fcc-a7f8-39a9957326d2"
name="target-role" value="Started" />
  </meta_attributes>
- <operations id="sbd-stonith-operations">
  <op id="sbd-stonith-op-monitor-15" interval="15" name="monitor"
start-delay="15" timeout="15" />
  </operations>
- <instance_attributes id="sbd-stonith-instance_attributes">
  <nvpair id="nvpair-724e62b3-0b26-4778-807e-a457dbd2fe42"
name="sbd_device" value="/dev/dm-1" />
  </instance_attributes>
  </primitive>
  </clone>
  </resources>


After some failover test, we found something strange. Said that if the
oracle service is running on node1. First we pull out the node1's
network cable, we can see the node1 is fenced by node2, resulting
node1 rebooting. Second the oracle db service will failover to node2,
and this is what we expected. But third, after the node1 boot up and
the network is up again, node2 is fenced by node1, and finally oracle
db service is failback to node1---this is NOT what we want.

We found after node1 rebooting, it seems that node1 cannot communicate
with node2 via TOTEM. And node1 uses sbd to fence node2 and get the
resource back. Is there something wrong with my settings? Or someone
could give me some advice about this situation?

I've attached the log on both nodes and the pacemaker's settings, if
you can, please pay some attention in node1's log below (tibcodb is
node1 and tibcodb2 is node2) :

Jun  1 11:30:39 tibcodb openais[5018]: [TOTEM] Token Timeout (5000 ms)
retransmit timeout (490 ms)
Jun  1 11:30:39 tibcodb openais[5018]: [TOTEM] The network interface
[10.224.1.89] is now up.
.......................................................
......................................................
Jun  1 11:30:39 tibcodb openais[5018]: [CLM  ] CLM CONFIGURATION CHANGE
Jun  1 11:30:39 tibcodb openais[5018]: [CLM  ] New Configuration:
Jun  1 11:30:39 tibcodb openais[5018]: [CLM  ]    r(0) ip(10.224.1.89)
 <======================== Why node2 ( tibcodb2 : 10.224.1.90 ) is not
recognized, and only node1 is recognized?
Jun  1 11:30:39 tibcodb openais[5018]: [CLM  ] Members Left:
Jun  1 11:30:39 tibcodb openais[5018]: [CLM  ] Members Joined:
Jun  1 11:30:39 tibcodb openais[5018]: [CLM  ]    r(0) ip(10.224.1.89)
Jun  1 11:30:39 tibcodb openais[5018]: [crm  ] notice:
pcmk_peer_update: Stable membership event on ring 104: memb=1, new=1,
lost=0
Jun  1 11:30:39 tibcodb openais[5018]: [crm  ] info: pcmk_peer_update:
NEW:  tibcodb 1
Jun  1 11:30:39 tibcodb openais[5018]: [crm  ] info: pcmk_peer_update:
MEMB: tibcodb 1
Jun  1 11:30:39 tibcodb openais[5018]: [MAIN ] info: update_member:
Node tibcodb now has process list: 00000000000000000000000000053312
(340754)
Jun  1 11:30:39 tibcodb openais[5018]: [SYNC ] This node is within the
primary component and will provide service.
........................................
.........................................
Jun  1 11:30:57 tibcodb stonithd: [5026]: info: client tengine [pid:
5031] requests a STONITH operation RESET on node tibcodb2 <==== Why
node1 want to fence node2 ?
Jun  1 11:30:57 tibcodb stonithd: [5026]: info:
stonith_operate_locally::2683: sending fencing op RESET for tibcodb2
to sbd-stonith:0 (external/sbd) (pid=5398)
Jun  1 11:30:57 tibcodb sbd: [5400]: info: tibcodb2 owns slot 1
Jun  1 11:30:57 tibcodb sbd: [5400]: info: Writing reset to node slot tibcodb2
Jun  1 11:31:07 tibcodb sbd: [5400]: info: reset successfully
delivered to tibcodb2


Please, any help would be appreciated.




More information about the Pacemaker mailing list