On Thu, Nov 15, 2012 at 7:44 AM, Robinson, Eric <[email protected]> wrote: >> I totally agree. I try to use HA setups in production >> environments but I only do 2 or so a year and meanwhile I >> have a complete zoo of versions, tools, shells etc. > > > I was trying fairly hard not to say something like that for fear of > alienating helpful members of the list, but I have to admit that when I saw > the above in writing, my reaction was relief. The phrase, "complete zoo of > versions, tools, shells etc." captures my feelings and it felt good to know I > was not alone. I only have 5 Linux-HA clusters in production, plus one > recently retired and one coming online as we speak. Probably it would help if > I did them more often. At a rate of 1 or 2 per year, things do change rapidly > enough that there's usually a new learning curve and I'm faced with the > choice to maintain different stacks and tools or try to retrofit the old > clusters to keep everything consistent, neither of which is easy with a small > staff. I don't mean to complain. If I wanted things to change less often I > could have gone into beet farming. But I have to agree with Lars that, > overall, the story could be improved, leading to broader acceptance.
Most of the churn has been in the interaction between Pacemaker and Corosync. The tools and configuration syntax is largely unchanged since 1.0. I can think of 3 tooling changes: - ptest/crm_simulate - hb_report/crm_report - standalone crmsh Thats not /too/ bad in 4 years. Ironically the reason for the Pacemaker/Corosync differences is mostly the "things don't change" principle that you mentioned... - SLES11 is tied to the custom plugin because we didn't understand the CMAN architecture at the time - RHEL6 is tied to CMAN because thats what rgmanager supports - Debian still thinks its the 1980s and relies on overlay repos And of course they're all on different release schedules, so lining them up is near impossible and Pacemaker ends up trying to support and document 5 different types of Corosync interactions (plus heartbeat). The only way to get consistency (until everyone is using corosync 2.x) is to drop support for all but one "enterprise" distro. Once the cluster is up though, the difference should be minimal. CTS can certainly test all 5 with very few differences. But perhaps I'm forgetting something important. _______________________________________________ Linux-HA mailing list [email protected] http://lists.linux-ha.org/mailman/listinfo/linux-ha See also: http://linux-ha.org/ReportingProblems
