From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753058AbbBZCwh (ORCPT ); Wed, 25 Feb 2015 21:52:37 -0500 Received: from mta8.srv.hcvlny.cv.net ([167.206.4.203]:56752 "EHLO mta8.srv.hcvlny.cv.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752535AbbBZCwg (ORCPT ); Wed, 25 Feb 2015 21:52:36 -0500 X-Greylist: delayed 1800 seconds by postgrey-1.27 at vger.kernel.org; Wed, 25 Feb 2015 21:52:36 EST Date: Wed, 25 Feb 2015 21:22:34 -0500 (EST) From: james owens Subject: Continually increasing inflight IO statistics with md mirror and blk-mq enabled on 3.19 release X-Originating-IP: [207.237.198.101] To: linux-kernel@vger.kernel.org Message-id: <3a774fb2.e494.14bc3b15a3f.Webtop.41@optonline.net> MIME-version: 1.0 Content-type: text/plain; charset=UTF-8; format=flowed; delsp=no Content-transfer-encoding: 7BIT Content-disposition: inline X-Priority: 3 X-SID: 41 X-Authuserid: jamesowens@optonline.net User-Agent: Laszlo Mail 3 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Please cc me personally on any replies as I am not a subscriber to the list. I currently have a 2 device md raid 1 on my Linux workstation where the slave devices are usb 3.0 external drives 4 TB each which I use for backup purposes. I also have a 3ware 9750 RAID controller with a HW RAID 1 SSD mirror and 12 drive RAID 6. I recently installed the 3.19 release kernel from openSuSE current to try out the block-mq optimizations. The new mq subsystem is clearly enabled by the presence of the mq subdirectory under /sys/block/foo. The performance on the 3ware is great, especially the SSD mirror; however, when using the md RAID 1 with external USB 3.0 devices, I noticed the inflight IO counters at /sys/block/md127/inflight are continuously increasing. This is not normal as these should go to 0 once any pending reads/writes are finalized. Is this simply an accounting problem I can safely ignore, or is it an unsafe situation that I need to be concerned about? Note that this does not occur with the exported disks on the 3ware array /dev/sda and /dev/sdb. I could try and see if this happens with a plain USB drive as I have some available. If you dismount the volume and stop the md array and then reassemble it, the statistics reset (as one would expect). I use a bitmap file to speed up resyncs, and the bitmap file is clearing, and there is no filesystem corruption that I can see, so the writes are clearly getting to the disks correctly. I will provide more information as best I can if needed. My guess is that this will be pretty easy to reproduce. Best regards, Jim Owens