On 2007-04-10T11:56:59, Michael Schwartzkopff <[EMAIL PROTECTED]> wrote:
> > > At the moment I would like to have that parameter since interop between > > > LVS and CLUSTERIP is not tested at all. After these tests we can drop it. > > One can't simply drop a parameter once introduced. > Why not? My script is still marked alpha and not intended for production use. > If it getts better variables get fixed. True, but that means I can't just pull your changes in ;-) For the final version, this should probably be removed. It seems we agree on this one. > > LVS vs clusterip is a good point, but that doesn't depend on this > > setting, does it? As long as it's detected by the script as above. > > > > LVS doesn't work with the clusterip anyway, so that's not really a > > problem. > But I did not test what happens if the user configures both. Just reject it and error out then - lvs_support doesn't work in clone mode, that is fine. The load_share parameter doesn't convey any additional meaning than configuring it as a clone (or not). Cloning an IPaddr2 resource only makes sense as a clusterip target. > Per-node ordering might be not the right way. Think about the following. Four > clones on four machines. Machine 3 dies and the first one has to take over. > 1) Resource 1 is shout down > 2) Resource 1 is started > 3) Resource 3 is started. > > Between 1) and 2) there is no connectivity for resource 1. Uhm. Of course. Machine 3 has just died. Of course there's no connectivity until it is restarted. > This might be up to several seconds, which is not acceptable in my > oppinion. But, there also won't be any IP rejects/denied being send, as none of the other node replies to that specific hash bucket yet. So, it'll look like a brief delay only. > Is would be better just to start resource 3 on node 1. That's what happens. If node3 crashed, we obviously don't stop the resource - that's implied in the node hosting the resource being gone & fenced. > Basically there is no big difference between per-node ordering and global > ordering. There's a big difference for your case, though. No concurrent invocations for the same clone on the same node, so you don't need to do any locking in your script. I don't see how your suggestion improves availability. Thanks, Lars -- Teamlead Kernel, SuSE Labs, Research and Development SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg) "Experience is the name everyone gives to their mistakes." -- Oscar Wilde _______________________________________________ Linux-HA mailing list Linux-HA@lists.linux-ha.org http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems