[Pacemaker] cloned IPaddr2 on 4 nodes

Vladimir Legeza vladimir.legeza at gmail.com
Fri Oct 29 12:24:43 UTC 2010


Hello,

On Fri, Oct 29, 2010 at 12:35 PM, Dan Frincu <dfrincu at streamwide.ro> wrote:

>  Hi,
>
>
> Vladimir Legeza wrote:
>
> *Hello folks.
>
> I try to setup four ip balanced nodes but,  I didn't found the right way to
> balance load between nodes when some of them are filed.
>
> I've done:*
>
> [root at node1 ~]# crm configure show
> node node1
> node node2
> node node3
> node node4
> primitive ClusterIP ocf:heartbeat:IPaddr2 \
>     params ip="10.138.10.252" cidr_netmask="32"
> clusterip_hash="sourceip-sourceport" \
>     op monitor interval="30s"
> clone StreamIP ClusterIP \
>     meta globally-unique="true" *clone-max="8" clone-node-max="2"*target-role="Started" notify="true" ordered="true" interleave="true"
> property $id="cib-bootstrap-options" \
>     dc-version="1.0.9-0a40fd0cb9f2fcedef9d1967115c912314c57438" \
>     cluster-infrastructure="openais" \
>     expected-quorum-votes="4" \
>     no-quorum-policy="ignore" \
>     stonith-enabled="false"
>
> *When all the nodes are up and running:*
>
>  [root at node1 ~]# crm status
> ============
> Last updated: Thu Oct 28 17:26:13 2010
> Stack: openais
> Current DC: node2 - partition with quorum
> Version: 1.0.9-0a40fd0cb9f2fcedef9d1967115c912314c57438
> 4 Nodes configured, 4 expected votes
> 2 Resources configured.
> ============
>
> Online: [ node1 node2 node3 node4 ]
>
>  Clone Set: StreamIP (unique)
>      ClusterIP:0    (ocf::heartbeat:IPaddr2):    Started node1
>      ClusterIP:1    (ocf::heartbeat:IPaddr2):    Started node1
>      ClusterIP:2    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:3    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:4    (ocf::heartbeat:IPaddr2):    Started node3
>      ClusterIP:5    (ocf::heartbeat:IPaddr2):    Started node3
>      ClusterIP:6    (ocf::heartbeat:IPaddr2):    Started node4
>      ClusterIP:7    (ocf::heartbeat:IPaddr2):    Started node4
> *
> Everything is OK and each node takes 1/4 of all traffic - wonderfull.
> But we become to 25% traffic loss if one of them goes down:
> *
>
> Isn't this supposed to be normal behavior in a load balancing situation, 4
> nodes receive 25% of traffic each, one node goes down, the load balancer
> notices the failure and directs 33,33% of traffic to the remaining nodes?
>
>
The only way I see to achive 33...% is to decrease  *clone-max *param value
(that should be multiple of online nodes number)
also *clone-max *should be changed on the fly (automaticly).

hmm... Idea is very interesting. =8- )
*
*

> Just out of curiosity.
>
> [root at node1 ~]# crm node standby node1
> [root at node1 ~]# crm status
> ============
> Last updated: Thu Oct 28 17:30:01 2010
> Stack: openais
> Current DC: node2 - partition with quorum
> Version: 1.0.9-0a40fd0cb9f2fcedef9d1967115c912314c57438
> 4 Nodes configured, 4 expected votes
> 2 Resources configured.
> ============
>
> Node node1: standby
> Online: [ node2 node3 node4 ]
>
>  Clone Set: StreamIP (unique)
> *     ClusterIP:0    (ocf::heartbeat:IPaddr2):    Stopped
>      ClusterIP:1    (ocf::heartbeat:IPaddr2):    Stopped *
>      ClusterIP:2    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:3    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:4    (ocf::heartbeat:IPaddr2):    Started node3
>      ClusterIP:5    (ocf::heartbeat:IPaddr2):    Started node3
>      ClusterIP:6    (ocf::heartbeat:IPaddr2):    Started node4
>      ClusterIP:7    (ocf::heartbeat:IPaddr2):    Started node4
>
> *I found the solution (to prevent loosing) by set clone-node-max to 3*
>
> [root at node1 ~]# crm resource meta StreamIP set clone-node-max 3
> [root at node1 ~]# crm status
> ============
> Last updated: Thu Oct 28 17:35:05 2010
> Stack: openais
> Current DC: node2 - partition with quorum
> Version: 1.0.9-0a40fd0cb9f2fcedef9d1967115c912314c57438
> 4 Nodes configured, 4 expected votes
> 2 Resources configured.
> ============
>
> *Node node1: standby*
> Online: [ node2 node3 node4 ]
>
>  Clone Set: StreamIP (unique)
> *     ClusterIP:0    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:1    (ocf::heartbeat:IPaddr2):    Started node3*
>      ClusterIP:2    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:3    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:4    (ocf::heartbeat:IPaddr2):    Started node3
>      ClusterIP:5    (ocf::heartbeat:IPaddr2):    Started node3
>      ClusterIP:6    (ocf::heartbeat:IPaddr2):    Started node4
>      ClusterIP:7    (ocf::heartbeat:IPaddr2):    Started node4
>
> *The problem is that nothing gonna changed when node1 back online.*
>
> [root at node1 ~]# crm node online node1
> [root at node1 ~]# crm status
> ============
> Last updated: Thu Oct 28 17:37:43 2010
> Stack: openais
> Current DC: node2 - partition with quorum
> Version: 1.0.9-0a40fd0cb9f2fcedef9d1967115c912314c57438
> 4 Nodes configured, 4 expected votes
> 2 Resources configured.
> ============
>
> Online: [ *node1* node2 node3 node4 ]
>
>  Clone Set: StreamIP (unique)
> *     ClusterIP:0    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:1    (ocf::heartbeat:IPaddr2):    Started node3*
>      ClusterIP:2    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:3    (ocf::heartbeat:IPaddr2):    Started node2
>      ClusterIP:4    (ocf::heartbeat:IPaddr2):    Started node3
>      ClusterIP:5    (ocf::heartbeat:IPaddr2):    Started node3
>      ClusterIP:6    (ocf::heartbeat:IPaddr2):    Started node4
>      ClusterIP:7    (ocf::heartbeat:IPaddr2):    Started node4
> *
> There are NO TRAFFIC on node1.
> If I back clone-node-max to 2  - all nodes revert to the original state.*
>
>
>
> So, My question is How to avoid such "hand-made" changes ( or is it
> possible to automate* clone-node-max* adjustments)?
>
> Thanks!
>
> You could use location constraints for the clones, something like:
>
> location StreamIP:0 200: node1
> location StreamIP:0 100: node2
>
> This way if node1 is up, it will run there, but if node1 fails it will move
> to node2. And if you don't define resource stickiness, when node1 comes back
> online, the resource migrates back to it.
>

I already tried to do so, but such configuration is not seems to be
acceptable:

crm(live)configure# location location_marker_0 StreamIP:0 200: node1
crm(live)configure# commit
element rsc_location: Relax-NG validity error : Expecting an element rule,
got nothing
element rsc_location: Relax-NG validity error : Element constraints has
extra content: rsc_location
element configuration: Relax-NG validity error : Invalid sequence in
interleave
element configuration: Relax-NG validity error : Element configuration
failed to validate content
element cib: Relax-NG validity error : Element cib failed to validate
content
crm_verify[20887]: 2010/10/29_16:00:21 ERROR: main: CIB did not pass
*DTD/schema
validation*
Errors found during check: config not valid



> I haven't tested this, but it should give you a general idea about how it
> could be implemented.
>
> Regards,
>
> Dan
>

> ------------------------------
>
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.orghttp://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs: http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>
>
> --
> Dan FRINCU
> Systems Engineer
> CCNA, RHCE
> Streamwide Romania
>
>
> _______________________________________________
> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>
> Project Home: http://www.clusterlabs.org
> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
> Bugs:
> http://developerbugs.linux-foundation.org/enter_bug.cgi?product=Pacemaker
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20101029/709e49c3/attachment-0002.htm>


More information about the Pacemaker mailing list