From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: From: Philipp Reisner To: drbd-dev@lists.linbit.com Subject: Re: [Drbd-dev] DRBD-8: Update to handle I/O failures during resync Date: Tue, 10 Oct 2006 13:42:05 +0200 References: <342BAC0A5467384983B586A6B0B3767103C3B0DB@EXNA.corp.stratus.com> In-Reply-To: <342BAC0A5467384983B586A6B0B3767103C3B0DB@EXNA.corp.stratus.com> MIME-Version: 1.0 Content-Type: text/plain; charset="iso-8859-15" Content-Transfer-Encoding: quoted-printable Content-Disposition: inline Message-Id: <200610101342.05647.philipp.reisner@linbit.com> List-Id: Coordination of development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Am Montag, 9. Oktober 2006 20:05 schrieb Graham, Simon: Hi Simon, Apart from a little whitespace change to make it to conform to Linus' Linux coding style. I applied the patch as you sent it. I read through it, and I could not spot anything wrong... > 1. New fields are added to the mdev and bm_extent structs to count the > number of failed requests during resync - the field in bm_extent is > used to determine when we are finished with the extent at which > time it is zeroed so that next time we process the same extent, > we'll try to resync everything again. > > 2. Resync is considered done when the remaining count of block to sync > is <=3D the count of failed requests. > > 3. When a failure occurs, a new routine, drbd_rs_failed_io() is called > to handle this - it takes care of incrementing the failed counts > and determining if the resync operation is now done. This is called > when a NegRSDreply is sent or received and also when a NegAck is > received for a resync operation (this is necessary to ensure that > the SyncSource correctly knows when the resync is done). > > 4. When a resync operation completes and there have been failure, the > new code does NOT rotate the UUIDs - I felt this was correct > because the resync really hasn't completed in this case (also, I > found that if I did not do this, the code would end up resyncing > the entire disk next time!) > > I did not introduce any new states for this - let me know if you think > there needs > to be one; with this patch, we simply end up Connected and Inconsistent. > Right. I will test it here in my environment as well. What more and more gets obvious that we need a method to start resync without setting the=20 bitmap, only to resync what is currently set. Maybe "drbdadm resync". Right now I do not see a reason for more states... Thanks or the patch, great work! =2DPhilipp =2D-=20 : Dipl-Ing Philipp Reisner Tel +43-1-8178292-50 : : LINBIT Information Technologies GmbH Fax +43-1-8178292-82 : : Sch=F6nbrunnerstr 244, 1120 Vienna, Austria http://www.linbit.com :