>>> Tony Stocker <[email protected]> schrieb am 05.06.2013 um 14:09 in Nachricht <alpine.lrh.2.03.1306051156320.11...@tf6102xuryqne.ccf.rbfqvf.anfn.tbi>:
> Is there any way to prioritize what resource groups or resource sets stay > active or are shut down as physical resources become scarce? Did you try " meta priority=..."? Hight priority groups are preferred, it seems. > > I have 8 resource groups. None of them can be run on the same hardware > simultaneously with any of the others. The cluster normally has a total > of 12 physical nodes. So during normal operations, if a physical node > fails, the resource group running on it should move to another one just > fine. That's a different constraint: Did you try "node ..utilization big_thing=1" to define the nodes where you can run one "big thing"? Also set placement-strategy="utilization" then. If you put "utilization big_thing=1" into each fat primitives, your server won't be overloaded, but groups aren't moved if only one primitive cannot be run (a bug IMHO). I'm unsure whether you can put utilizations at the group level. In my understanding the group should summarize all the individual primitives utilizations, but it does not. To complete the non-rationale, stickinesses are summarized for groups, not maximized (i.e.: taking the largest value). You see, ther's some playground there.... > > What happens though if the number 'up' nodes falls to 7 (the quorum > limit)? Is there any way to ensure certain resource groups have priority > over other resource groups? See before. > > Four (4) of my resource groups are related in that one of them is a > process master which farms out jobs to three (3) computing nodes (all of > these are running software designed in-house). Now, without that process > master node the computing nodes are basically worthless. How can I ensure > that the process master will continue to run in a low-availability > scenario, even if that means only running two (2) or even one (1) of the > computing resource groups? Use a high priority for it. > > I'm especially concerned about a scenario in which everything is up and > running normally and then a large subset of physical nodes fail leaving > only the quorum limit nodes. For example: > > Physical Node Resource Group > ------------------------------ > 1 Process Master > 2 Compute 1 > 3 Compute 2 > 4 Compute 3 > 5 Distro 1 > 6 Distro 2 > 7 Segmenter > 8 Developer > 9 -- > 10 -- > 11 -- > 12 -- > > Now assume that physical nodes 1,9,10,11,12 FAIL. The cluster still has 7 > physical nodes, so it meets quorum requirements. But now every node in it > (2-8) has something running on it with a -inf colocation restraint for all > other resource groups. > > Is there a way to kill one of the resource groups that has a lower > priority (for example 'Distro 2' or 'Developer') so that 'Process Master' > can be launched on that physical node instead? priority and utilization both will stop resources to fulfill the constrains (I'm not saying your needs ;-)) Regards, Ulrich > > Thanks, > > Tony > > -- > This message has been scanned for viruses and > dangerous content by MailScanner, and is > believed to be clean. > > _______________________________________________ > 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
