[Drbd-dev] [GIT PULL] drbd-8.4.2 for the linux-3.8 merge window
axboe at kernel.dk
Fri Nov 9 12:40:15 CET 2012
On 2012-11-09 12:36, Philipp Reisner wrote:
> Hi Jens,
> Please pull drbd-8.4.2 into your for-3.8/drivers branch.
> The most noticeable change is the support for multiple replicated
> volumes in a single DRBD connection.
> This release requires new drbd userland tools >= 8.4.0 (available
> since July 2011).
> 8.4.x is network protocol compatible with all previous releaes.
> This release brings a new meta-data format. Forward (8.3 -> 8.4)
> conversion happens complete seamless. Backward conversion
> is done by a single command (drbdadm apply-al res).
> The "recent changes" chapter of our user's guide describes all
> changes in detail:
> 8.4.2 (api:genl1/proto:86-100)
> * Go into inconsistent disk state with on-io-error=pass-on policy
> * Timeouts for requests processing on the peer (previously that
> worked only if the data socket was congested)
> * Conflicting write detection is now based on an interval tree,
> removed the hash-tables (necessary for the unlimited BIO sizes)
> * Support for multiple volumes (minors, block devices) per connection;
> up to 65536 volumes per connection supported
> * Reduced IO latencies during some state changes (esp. start resync)
> * New on disk format for the AL: double capacity; 4k aligned IO; same space
> * Multiple AL changes in a single transaction (precondition for
> unlimited BIO sizes)
> * DRBD no longer imposes any limit on BIO sizes
> * Removed DRBD's limits on the number of minor devices
> * DRBD's minors can now be removed (not only unconfigured)
> * Switched the user space interface form connector to generic netlink
> * The wire-protocol is now a regular connection option, which can be
> changed while the device is online
> * IO freezing/thawing is done on connection (all volumes) level
> * fencing is done on connection (all volumes) level
> * Enforce application of activity log after primary crash in user space
> * New default values (compared to drbd-8.3) for: minor-count, ko-count, al-extents,
> c-plan-ahead, c-fill-target, c-min-rate, use-rle, on-io-error
> * Optional load balancing for read requests: new keyword "read-balance"
> * New option 'al-updates no' to disable writing transactions into the
> activity log. It is use full if you prefer a full sync after a primary
> crash, for improved performance of a spread out random write work load
> * Expose the data generation identifies via sysfs
> Jens, regarding the code:
> It has the sysfs bits in again. The reason for that is that we want to
> expose more information by that, and remove the /proc/drbd with the
> next evolutionary step. -- In case this is a show stopper, let me
> remove the sysfs bits.
The exact same sysfs bits I complained about last time? If yes, then I
don't understand why you haven't changed yet. Or why you are trying to
push the same bits again that got rejected last time.
> Here is the git-pull-request test:
> (The patch subjects removed to make the mail more digestible)
Please don't do that, it basically makes the pull request useless! A few
hundred extra lines is not an issue.
More information about the drbd-dev