philipp.reisner at linbit.com
Sat Apr 7 10:18:22 CEST 2007
It is time to get all the post 8.0.1 fixes out of the door:
* Removed a bug that could cause an OOPS in drbd_al_to_on_disk_bm()
* Improved the robustness of the UUID based algorithm that decides
about the resync direction.
* Fixed the error handling in case a the open() of a backing
* Fixed a race condition that could cause a "drbdadm disconnect" to hang.
* More verbosity in we can not claim a backing block device.
* More verbosity and paranoia in the bitmap area.
* On some vendor kernels 8.0.1 did not load because of kzalloc. fixed.
* Fault injection can now not only be turned on or off, but can be
enabled on a per device basis.
* Fixed the scripts and files needed to build drbd into a kernel.
The first on the list is the real nasty one. It might only bite you if
you do a switchover after a lot of IO-activity, and that IO must
be 100% scattered, no sequential accesses. (the AL must be full)
(Actually I am still impressed that this issue was triggered
by an user at all.)
The bug of fix number two can be reproduces by resetting a SyncTarget
node, while it is in the "WFBitMapT" state. It is a small time-
window, but it can happen...
Nothing hair-raising, but users of 8.0.x should definitely upgrade
More information about the drbd-announce