On Tue, Jun 27, 2006 at 10:23:46PM +0200, Lars Marowsky-Bree wrote: > On 2006-06-26T15:10:05, Dejan Muhamedagic <[EMAIL PROTECTED]> wrote: > > > This has nothing to do with your question, but it is related. The > > colocation constraint with INFINITY can bite. If you have three > > resources, R, A, and B, and both A and B should run together with > > R (R<--INF-->AB), but A and B don't depend on each other, then > > stopping either A or B stops R too. So, it sort of does more than > > needed. According to Andrew Beekhof this won't change before he > > can do some big changes and that certainly won't be in 2.0.6. > > > > The way I solved this kind of setup is by specifying same scores > > for all interdependent resources on a per node basis and with > > always different scores for different nodes. That way you can > > ensure to have all resources running on one node. After adding > > order constraints you should be OK. > > BTW, we just came up with a rather nice hack too. (As nice a work-around > as it gets, I think.) > > If you put all resources you want to be collocated in a group, BUT set > collocated="false" on the group. And create a rsc_location constraint > refering to the group, enumerating all eligible nodes with a unique > score, they'll also be colocated, but the cluster will not stop > resources when one of them fails.
If so, that would make the configuration much simpler. For example, with the above workaround, if one wants to move the resources to another node (as in crm_resource -M), then constraints have to be defined for every resource, so basically crm_resource cannot be used, but one has to create constraints in succession quickly with cibadmin. That makes it quite tricky to manage the cluster. If I understood correctly, it should be something like this: (single l or double l?) <group id="data1" colocated="false"> <primitive id="A"...> ... <primitive id="R1"...> ... <primitive id="R2"...> ... </group> Then, such a group would have the following properties: - resources are started/stopped in the specified order - resources always run on one node Right? Cheers, Dejan > Then you at least don't have to give the location constraint over and > over again, but can just do so for the group. > > > Sincerely, > Lars Marowsky-Br?e > > -- > High Availability & Clustering > SUSE Labs, Research and Development > SUSE LINUX Products GmbH - A Novell Business -- Charles Darwin > "Ignorance more frequently begets confidence than does knowledge" > > _______________________________________________________ > Linux-HA-Dev: [email protected] > http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev > Home Page: http://linux-ha.org/ _______________________________________________________ Linux-HA-Dev: [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev Home Page: http://linux-ha.org/
