On 2011-07-06 21:58, Florian Haas wrote: >> Attached is a patch for the init.d script which makes the >> drbd-script wait for the arrival of the ll- and md-devices. > > Excellent contribution,
Not too excellent alas, for today I've encountered a sorta-kinda race condition with md, which starts devices with size zero (which is too small for drbd :) to be scaled up to size later. So there needs to be another condition in the while-loop to check the size of the device (I settled for `blockdev --getsize64` -gt 0). > however if you are getting bitten by this here is a practical > workaround: as a general rule, the preferred management framework for > DRBD is Pacemaker, and on Pacemaker managed clusters it is entirely > appropriate to leave the init script disabled (chkconfig drbd off). > At that point Pacemaker will take care of everything (including > loading the drbd.ko module), and all of this will obviously occur > long after your udev has settled. Thank you for pointing me to Pacemaker; however, if one would call the learning curve for understanding drbd 'steep', I would call the curve for Pacemaker, well, 'straight up'? Really, for what I'm using drbd a complete cluster management suite is a bit much. I realize that I'm not being the usual drbd-customer here, but I thought I'd share what I discovered along the way so that maybe more people can benefit. If there's another avenue for that I'd love to hear about it. thanks for replying, Mrten. _______________________________________________ drbd-user mailing list [email protected] http://lists.linbit.com/mailman/listinfo/drbd-user
