philipp.reisner at linbit.com
Fri Jan 15 14:22:21 CET 2010
The two highlights of drbd-8.3.7:
* New RPM packaging, which is very close to Fedora's packaging rules.
In case you are using RPMs make sure that you install the needed
subset of RPMs.
The drbd-X.X.X-1.ARCH.rpm is now a virtual package that
has dependencies on all the components: drbd-utils, drbd-udev,
drbd-xen, drbd-bash-completion, drbd-heartbeat and drbd-pacemaker.
Those in turn might depend on the component they bridge to drbd.
I.e. drbd-udev depends on udev.
For a hassle free upgrade we omitted a few dependencies:
drbd-xen does not depend on xen
drbd-pacemaker does not depend on pacemaker
drbd-heartbeat does not depend on heartbeat
For new installations I recommend to install drbd-utils and drbd-udev,
and the other integration package(s) you need, and to not install
the virtual package (drbd).
* New default config that suggests to use config-snippets in /etc/drbd.d/
Adding new drbd resources to a pacemaker cluster with the new
OCF based resource agent can be a challenging task. The issue
is that pacemaker might execute a monitor operation every
second. Since the RA uses drbdadm, it will fail if you have a
broken config on your machine. -- Broken configuration files
may happen when adding new resources.
One way out is to split the config into snippets:
/etc/drbd.d/global_common.conf and some /etc/drbd.d/*.res
A new resource is first created as /etc/drbd.d/something.new
Then you check the syntax of the new resource with
drbdadm -t /etc/drbd.d/something.new dump something
and if it is successful activate that resource by renaming
it's config snippet: mv something.new something.res
* Lots of fixes to the new RPM packaging
* Lots of fixes to the autoconfig stuff
* Following the rename of CONFIG_LBD to CONFIG_LBDAF
* Silenced an assert. Could trigger after changing write ordering (Bugz 261)
* Fixed a race condition between detach and ongoing IO. Very hard to
trigger, caused an OOPS in make_request/drbd_make_request. (Bugz 262)
* Fixed a regression in the resync handshake code introduced before 8.3.3.
That bug causes DRBD to block during the initial handshake when a partial
resync is not possible but a full resync is necessary. Happens very rarely.
* Do not drop into StandAlone mode when connection is lost during authentication
* Corrected a null test in the authentication code, found by conccinelle,
thanks to upstream integration. The chance to trigger that was probably 10^-9.
* crm-fence-peer.sh is now also usable if DRBD is managed from the xen block
* Fixes to the init script's dependencies
* Backported cleanups that where contributed to the in kernel DRBD
* Allow online resizing of disconnected devices, new option to drbdsetup
drbdsetup /dev/drbdX resize --assume-peer-has-space
* Allow multiple after options in the syncer section for stacked setups
* Correctly process relative paths in include statements in drbd.conf
* New option (-t) for drbdadm to test syntax of config snippets
* Following Linux upstream changes 2.6.32 (SHASH and in_flight issues)
* New /etc/drbd.conf example that suggests the use of /etc/drbd.d/xxx.res
With this release we have a new policy, that LINBIT's pre build RPMs and
DPKGs are available 7 days after the release. This is Friday January 22,
: Dipl-Ing Philipp Reisner
: LINBIT | Your Way to High Availability
: Tel: +43-1-8178292-50, Fax: +43-1-8178292-82
DRBD(R) and LINBIT(R) are registered trademarks of LINBIT, Austria.
More information about the drbd-announce