From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mx1.redhat.com ([209.132.183.28]:54190 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S936190AbdIZIua (ORCPT ); Tue, 26 Sep 2017 04:50:30 -0400 Date: Tue, 26 Sep 2017 16:50:09 +0800 From: Ming Lei To: Mike Snitzer Cc: Bart Van Assche , "dm-devel@redhat.com" , "hch@infradead.org" , "linux-block@vger.kernel.org" , "axboe@fb.com" , "loberman@redhat.com" Subject: Re: dm-mpath: return DM_MAPIO_REQUEUE in case of rq allocation failure Message-ID: <20170926085007.GD518@ming.t460p> References: <20170922013506.22510-1-ming.lei@redhat.com> <1506092775.2512.5.camel@wdc.com> <20170922174431.GA21108@ming.t460p> <1506102888.2512.9.camel@wdc.com> <20170925030641.GB7090@ming.t460p> <1506352994.2641.4.camel@wdc.com> <20170925161014.GB16470@ming.t460p> <20170925161702.GA29036@redhat.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii In-Reply-To: <20170925161702.GA29036@redhat.com> Sender: linux-block-owner@vger.kernel.org List-Id: linux-block@vger.kernel.org On Mon, Sep 25, 2017 at 12:17:03PM -0400, Mike Snitzer wrote: > On Mon, Sep 25 2017 at 12:10pm -0400, > Ming Lei wrote: > > > On Mon, Sep 25, 2017 at 03:23:16PM +0000, Bart Van Assche wrote: > > > On Mon, 2017-09-25 at 11:06 +0800, Ming Lei wrote: > > > > On Fri, Sep 22, 2017 at 05:54:48PM +0000, Bart Van Assche wrote: > > > > > On Sat, 2017-09-23 at 01:44 +0800, Ming Lei wrote: > > > > > > On Fri, Sep 22, 2017 at 03:06:16PM +0000, Bart Van Assche wrote: > > > > > > > On Fri, 2017-09-22 at 09:35 +0800, Ming Lei wrote: > > > > > > > > + /* > > > > > > > > + * blk-mq's SCHED_RESTART can cover this requeue, so > > > > > > > > + * we needn't to deal with it by DELAY_REQUEUE. More > > > > > > > > + * importantly, we have to return DM_MAPIO_REQUEUE > > > > > > > > + * so that blk-mq can get the queue busy feedback, > > > > > > > > + * otherwise I/O merge can be hurt. > > > > > > > > + */ > > > > > > > > + if (q->mq_ops) > > > > > > > > + return DM_MAPIO_REQUEUE; > > > > > > > > + else > > > > > > > > + return DM_MAPIO_DELAY_REQUEUE; > > > > > > > > } > > > > > > > > > > > > > > This patch is inferior to what I posted because this patch does not avoid > > > > > > > the delay if multiple LUNs are associated with the same SCSI host. Consider > > > > > > > e.g. the following configuration: > > > > > > > * A single SCSI host with two SCSI LUNs associated to that host, e.g. /dev/sda > > > > > > > and /dev/sdb. > > > > > > > * A dm-mpath instance has been created on top of /dev/sda. > > > > > > > If all tags are in use by requests queued to /dev/sdb, no dm requests are in > > > > > > > progress and a request is submitted against the dm-mpath device then the > > > > > > > blk_get_request(q, GFP_ATOMIC) call will fail. The request will be requeued > > > > > > > and the queue will be rerun after a delay. > > > > > > > > > > > > > > My patch does not introduce a delay in this case. > > > > > > > > > > > > That delay may not matter because SCHED_RESTART will run queue just > > > > > > after one request is completed. > > > > > > > > > > Did you understand what I wrote? SCHED_RESTART will be set for /dev/sdb but not > > > > > for the dm queue. That's what I was trying to explain to you in my previous e-mail. > > > > > > > > The patch I posted in this thread will set SCHED_RESTART for dm queue. > > > > > > This is not how communication on an open source mailing list is assumed to work. > > > If you know that you are wrong you are assumed either to shut up or to admit it. > > Code speaks much better than unnecessarily caustic exchanges. Not sure > why Bart persists with that.. but that's for him to sort out. > > > You just mentioned 'This patch is inferior' and never explained my patch > > is wrong, so please go ahead and show me why this patch(the post in this > > thread, also the following link) is wrong. > > > > https://marc.info/?l=linux-block&m=150604412910113&w=2 > > > > I admit both are two ways for the issue, but I don't think my patch > > is wrong. Your approach can be a very big change because .queue_rq() > > will block, and I also mentioned it might cause AIO regression. > > I have no interest in changing DM multipath to block in .queue_rq() > So please consider that approach a dead end. > > Ming, just iterate on your revised patchset, test and post when you're > happy with it. Hi Mike, If you don't consider to change mpath into block in .queue_rq() now, please take this patch first, and I am sure this way is correct, and even it can be thought as a fix. Thanks, Ming