Hi, On Fri, Sep 26, 2008 at 02:41:54PM +0200, Luis Motta Campos wrote: > Fellows, > > My Heartbeat set up is almost good to get into production. We just > need to decide and implement a better solution for the STONITH > procedure. We've been using the "dummy" solution so far. > > I have a somewhat unusual set up, with two clusters running MySQL > database servers: a "Replication Master Cluster" and a "Replication > Slave Cluster", both consisting of two identical hosts. > > I call my solution unusual because every node in both clusters is > located in a different data centre (I can not afford to stop my service > because I lost of connectivity with a single data centre). > > That means that there's little or no possibility of getting serial > access to the power console that controls the peer of a given host as > it's physically impossible to pull the necessary cables (for several > reasons). > > At this moment, I am working with mainly two alternatives: meatware or > a custom driver for STONITH, that will make a single SSH call with a > specially crafted key to trigger some action from the remote power console. > > As a solution, meatware is a proven, well-tested solution that is > distributed as part of the heartbeat standard software. It also doesn't > relies on the network, which is a good thing. On the down side, it > requires a human operator, causing long delays for the service take > over. It also will require more training for the operators, which is not > desirable (they have a lot in their plate, already). > > On the other hand, the promise of fast and automatic service recovery > without the intervention of a human operator, the lack of need for > operator specific training and know-how may compensate for the fact that > the solution depends on the Network and a custom driver, and it's not > largely tested (as all good home-grown solutions). > > Given that, I would like to have your technical advice. What solution > would you recommend me, given the aforementioned restrictions?
Either way would work as you described it. The "custom driver", which would just be a new stonith plugin doesn't look like very difficult to implement. We're happy to see contributions :) Whether that solution would work depends largely on your network connectitivy. How likely it is that neither node can fence the other side? Also, if that happens, which would result in both nodes waiting forever for the fencing operation, does that actually make any difference to the users? Hth and sorry for the late reply. Dejan > Thanks a lot in advance. > Regards. > -- > Luis Motta Campos is a software engineer, > Perl Programmer, foodie and photographer. > _______________________________________________ > 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
