Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On 10-05-21 09:02 AM, Philipp Reisner wrote: > Hi, > > This again is a call for all DRBD users, that have some testing equipment > available, and are ready to contribute a bit of their time to testing > DRBD. > > This is a release candidate, do not use it for production. > This is a release candidate, do not package it for distributions. > > The most outstanding change with this release is, that DRBD now > can use multiple BIOs (single IO requests) for processing a signle > data packat on the receiving side. This allows DRBD in many more > cases than before (on LVM's LVs) to process BIOs of a size of up > to 32k. (Before that DRBD, in many cases, forced the whole stack > down to a max BIO size of 4k). > > This is important for hardware setups where both disk IO subsystem > and connecting network operate in the area of 10GBit/sec. Those become > rapidly more common. > > > 8.3.8rc1 (api:88/proto:86-94) > -------- > * Do not expose failed local READs to upper layers, regression introduced > in 8.3.3 > * Fixed support for devices with 4k hard sector size (again) > * Fixed a potential Oops in the disconnect code > * Fixed a race condition that could cause DRBD to consider the peers disk > as Inconstent after resync instead of UpToDate (Bugz 271) > * Fixed a reace condition that could cause DRBD to consider the peers disk > as Outdated instead of Inconsistent during resync (Bugz 277) > * Disallow to start a resync with invalidate / invalidate-remote when the > source disk is not UpToDate > * Forcing primary works now also for Consistent, not only for Outdated and > Inconsistent (Bugz 266) > * Improved robustness against corrupt or malicous sector addresses when > receiving data > * Added the initial-split-brain, it gets called also if the split-brain gets > automatically resolved > * Added the --assume-clean option for the resize command, it causes drbd to > not resync the new storage after an online grow operation > * drbdadm: Do not segfault if stacked-on-top-of refers to an undefined res > * drbdadm: Do not consider configs with invalid after statements as invalid > * drbdadm: Do not segfault if the peer's proxy section is missing > * drbdadm: Allow nullglob in include statement > * drbdadm: Fixed the use of waitpid > * init script: fix insserv headers (Debian 576901) > * Gave the receiving code the ability to use multiple BIOs for writing a > single data packet; now DRBD works with BIOs up to 32kByte also on LVM > devices; from now on the use_bmbv config option does nothing > * New command check-resize, that allows DRBD to detect offline resizing > and to move internal meta-data accordingly > * Added a control loop, that allows DRBD to find auto tune the resync > speed, on connections with large queues (drbd-proxy) > * Create new current UUIDs late, after loosing connection to the peer > * --dry-run option for connect; disconnects after sync handshake > * --overwrite-data-of-peer got an alias named --force > * Improvements to crm-fence-peer > * Fixed option parsing and stacking in snapshot-resync-target-lvm.sh > * Compiles on 2.6.33 and 2.6.34 > > http://git.drbd.org/?p=drbd-8.3.git;a=commit;h=drbd-8.3.8rc1 > http://oss.linbit.com/drbd/8.3/drbd-8.3.8rc1.tar.gz > > Happy testing! > > -Phil If a normal 1Gbit link is enough to test, and if you can give me an idea of what tests I could run to best test this, I'll be happy to load it up on my test cluster this weekend. -- Digimer E-Mail: linux at alteeve.com AN!Whitepapers: http://alteeve.com Node Assassin: http://nodeassassin.org