[Pacemaker] Question on resources' dependency and failover

[Andrew Beekhof]

On Tue, May 11, 2010 at 6:31 PM, Gianluca Cecchi
<gianluca.cecchi at gmail.com> wrote:
> Hello,
> I'm using pacemaker 1.0.8 on rh el 5.5 x86 with clusterlabs repo.
> Based on other posts on linux-ha I'm trying to configure a 2-nodes cluster
> where one of the nodes is nfs-server and the other one is nfs-client of the
> resource exported by the first one.
> The main parts borrowed form the related threads are that I had to set
>  these two constraints:
> configure colocation nfsclient_not_on_nfs-group -inf: nfs-group nfsclient
> configure order nfsclient_after_nfs-group inf: nfs-group nfsclient
> It is all ok if both nodes are ok.
> And I have nfs-server on node1 and nfs-client on node2.
> I successfully configured ping resource for monitoring connectivity (thanks
> again Vadym for the patience for that..).
> At runtime I have:
> Online: [ ha1 ha2 ]
> Full list of resources:
> SitoWeb (ocf::heartbeat:apache):        Started ha1
>  Master/Slave Set: NfsData
>      Masters: [ ha1 ]
>      Slaves: [ ha2 ]
>  Resource Group: nfs-group
>      lv_drbd0   (ocf::heartbeat:LVM):   Started ha1
>      ClusterIP  (ocf::heartbeat:IPaddr2): Started ha1
>      NfsFS (ocf::heartbeat:Filesystem):    Started ha1
>      nfssrv     (ocf::heartbeat:nfsserver):     Started ha1
> nfsclient (ocf::heartbeat:Filesystem):    Started ha2
>  Clone Set: cl-pinggw
>      Started: [ ha1 ha2 ]
> Migration summary:
> * Node ha1:  pingd=200
> * Node ha2:  pingd=200
> and stickiness for cluster is set to 1000
> I can successfully execute
> crm resource move nfs-group
> (and then after a bit unmove to have chance to get it back in the future)
> so that the roles are correctly exchanged between the two nodes at the end
> of the process.
> But I can't do
> crm resource move nfsclient
> because this resource then stops, and the same for nfs-group and nothing
> restarts

Attach the output from cibadmin -Ql when the cluster is in this state
and I'll take a look.

> If I run
> crm resource unmove nfsclient
> then all restarts (with the same roles as before)
> I can live with this... but I would expect only failure of movement of
> nfs-client without impact for nfs-server
> If I prevent node1 (nfs-server role at the moment) to reach the gw, I
> correctly get the resources swapped (because nfs-client is not bound to the
> ping constraint)
> The complete failure of services with my case for movement of nfs-client
> lets me make this general question:
> Suppose I have group of resources named G1 and a resource named R2.
> I define an order R2 after G1 and a colocation constraint of -inf so
> that they run on different nodes (2 nodes overall).
> At runtime I have G1 on node1 and R2 on node2, correctly.
> I do a move of G1, so that I would have at the end G1 on node2 and R2 on
> node1.
>
> But suppose node1 doesn' t satisfy requirements for running R2 (for
> example a file system without the mount point directory on node1).
>
> What Is the expected behaviour:
> - R2 stops correctly on node2
> - G1 starts correctly on node2
> - R2 then, based on constraint and order, tries to start on node1 but fails
> end of the game with G1 still running or
> - G1 too goes in stopped status, so that I now have both G1 and R2 stopped?
>
> And the other case:
> At runtime I have G1 on node1 and R2 on node2, correctly.
> I do a move of R2, so that I would expect at the end G1 on node2 (based on
> anti-location) and R2 on node1.
> Is it expected to have as a result both G1 and R2 in stopped state?
> Thanks for clarifications
>
> _______________________________________________
> 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
>