Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
please do not hijack a thread. On Mon, Sep 15, 2008 at 09:45:25AM -0500, nathan at robotics.net wrote: > > My nice cisco 3750 switch crashed yesterday, both DRBD servers noticed > the lack of connectivity, then came back when the switch rebooted and > auto resolved the split brain. However I am running CLVM on top of DRBD > and fencing was not happy with the loss of connectivity and each server > decided to shoot each other in the head. > > Each server is connected with gig ethernet as well as infiniband and the > cluster and I believe fencing can be configured to use two interfaces. Is > there any way to configure DRBD to use BOTH IP address? The ability to > use two networks for DRBD would help in many ways. no. you can use bonding, though, which is recommend. > The 2nd question is, assuming that DRBD can not be configured to use two > networks, how safe is it for me to configure fencing to use both network > and have DRBD run as primary/primary for a while in split brain? please define "safe". > The boxes are used for XEN and each server only reads/writes to a > unique file using tapaio. My guess is not long before blocks would get > messed up beyond repair even if they are each accessing unique files, > but I am not sure on that. as long as the files stay the same size and you are not using some rebalancing, auto defragmenting or log structured file system (you don't), that is just fine. problem is: drbd has no three (not even two!) way merge. so after you connect them again and let them resync, its either the yellow, or the blue. the respective other is warped back in time to where it has been before the split brain. -- : Lars Ellenberg : LINBIT | Your Way to High Availability : DRBD/HA support and consulting http://www.linbit.com DRBD® and LINBIT® are registered trademarks of LINBIT, Austria. __ please don't Cc me, but send to list -- I'm subscribed