Re: possible raid6 corruption
Christoph Anton Mitterer wrote (ao): May 19 03:25:50 lcg-lrz-dc10 kernel: [903106.581205] sd 0:0:14:0: Device offlined - not ready after error recovery May 28 16:38:43 lcg-lrz-dc10 kernel: [1727488.984810] sd 0:0:14:0: rejecting I/O to offline device May 28 16:39:19 lcg-lrz-dc10 kernel: [1727524.067182] BTRFS: lost page write due to I/O error on /dev/sdm May 28 16:39:19 lcg-lrz-dc10 kernel: [1727524.067426] BTRFS: bdev /dev/sdm errs: wr 1, rd 0, flush 0, corrupt 0, gen 0 May 28 21:03:06 lcg-lrz-dc10 kernel: [1743336.347191] sd 0:0:14:0: rejecting I/O to offline device May 28 21:03:06 lcg-lrz-dc10 kernel: [1743336.369569] BTRFS: lost page write due to I/O error on /dev/sdm Well as I've said,.. maybe it's not an issue at all, but at least it's strange that this happens on brand new hardware only with the btrfs-raid56 node, especially the gazillions of megasas messages. Brand new hardware is most likely to show (hardware) issues as it has no proven track record yet while it was subject to any kind of abuse during transport. I'm sure you will see the same if you put sw raid + ext4 on this server. Nice hardware btw, please share your findings. Sander -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
possible raid6 corruption
Hi. The following is a possible corruption of a btrfs with RAID6,... it may however also be just and issue with the megasas driver or the PERC controller behind it. Anyway since RADI56 is quite new in btrfs, an expert may want to have a look at it whether it's something that needs to be focused on. I cannot mount the btrfs since the incident: I.e. # mount /dev/sd[any disk of the btrfs raid] /mnt/ gives a: [358466.484374] BTRFS info (device sda): disk space caching is enabled [358466.484426] BTRFS: has skinny extents [358466.485421] BTRFS: failed to read the system array on sda [358466.543422] BTRFS: open_ctree failed But no valuable data has been on these devices and I haven't really tried any of the recovery methods. What I did: At the university we run a Tier-2 for the LHC computing grid (i.e. we have loads of storage). Recently we bought a number of Dell nodes each with 16 6TB SATA disks, the disks are connected via a Dell PERC H730P controller (which is based on some LSI Mega*-whatever, AFAIC). Since I had 10 new nodes I wanted to use the opportunity and do some extensive benchmarking, i.e. HW RAID vs. MD RAID, vs btrfs-RAID... + btrfs and ext4, in all reasonable combinations. The nodes which were used for MD/btrfs-RAID obviously used the PERC in pass-through-mode. As said, the nodes are brand new and during the tests the one with btrfs-raid6 had a fs crash (all others continued to run fine). System is Debian jessie, except the kernel from sid (or experimental of that time) 4.0.0 and btrfs-progs 4.0. The fs was created pretty much standard: # mkfs.btrfs -L data-test -d raid6 -m raid6 /dev/sda /dev/sdb /dev/sdc /dev/sdd /dev/sde /dev/sdf /dev/sdg /dev/sdh /dev/sdi /dev/sdj /dev/sdk /dev/sdl /dev/sdm /dev/sdn /dev/sdo /dev/sdp And then there came some heave iozone stressing: # iozone -Rb $(hostname)_1.xls -s 128g -i 0 -i 1 -i 2 -i 5 -j 12 -r 64 -t 1 -F /mnt/iozone Some excerpts from the kerne.log, which might be of interest: May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.479387] Btrfs loaded May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.479680] BTRFS: device label data-test devid 1 transid 3 /dev/sda May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.482080] BTRFS: device label data-test devid 2 transid 3 /dev/sdb May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.484047] BTRFS: device label data-test devid 3 transid 3 /dev/sdc May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.486021] BTRFS: device label data-test devid 4 transid 3 /dev/sdd May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.487892] BTRFS: device label data-test devid 5 transid 3 /dev/sde May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.489849] BTRFS: device label data-test devid 6 transid 3 /dev/sdf May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.491819] BTRFS: device label data-test devid 7 transid 3 /dev/sdg May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.493919] BTRFS: device label data-test devid 8 transid 3 /dev/sdh May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.495761] BTRFS: device label data-test devid 9 transid 3 /dev/sdi May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.497645] BTRFS: device label data-test devid 10 transid 3 /dev/sdj May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.499477] BTRFS: device label data-test devid 11 transid 3 /dev/sdk May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.501307] BTRFS: device label data-test devid 12 transid 3 /dev/sdl May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.503208] BTRFS: device label data-test devid 13 transid 3 /dev/sdm May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.505037] BTRFS: device label data-test devid 14 transid 3 /dev/sdn May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.506837] BTRFS: device label data-test devid 15 transid 3 /dev/sdo May 10 00:26:39 lcg-lrz-dc10 kernel: [115511.508800] BTRFS: device label data-test devid 16 transid 3 /dev/sdp May 10 00:27:34 lcg-lrz-dc10 kernel: [115566.351260] BTRFS info (device sdp): disk space caching is enabled May 10 00:27:34 lcg-lrz-dc10 kernel: [115566.351307] BTRFS: has skinny extents May 10 00:27:34 lcg-lrz-dc10 kernel: [115566.351333] BTRFS: flagging fs with big metadata feature May 10 00:27:34 lcg-lrz-dc10 kernel: [115566.354089] BTRFS: creating UUID tree Literally gazillions of these: May 19 02:39:19 lcg-lrz-dc10 kernel: [900318.402678] megasas:span 0 rowDataSize 1 May 19 02:39:19 lcg-lrz-dc10 kernel: [900318.402705] megasas:span 0 rowDataSize 1 Wile I saw the above lines on all the other nodes as well, there were only like 30 once, and that's it. But on the one node with btrfs the log file was flooded to 1,6GB with these. At some point I've had this: May 19 03:25:19 lcg-lrz-dc10 kernel: [903075.511076] megasas: [ 0]waiting for 1 commands to complete for scsi0 May 19 03:25:24 lcg-lrz-dc10 kernel: [903080.526184] megasas: [ 5]waiting for 1 commands to complete for scsi0 May 19 03:25:29 lcg-lrz-dc10 kernel: [903085.541375] megasas: [10]waiting for 1 commands to complete for scsi0 May 19 03:25:34 lcg-lrz-dc10 kernel:
Re: possible raid6 corruption
I'm seeing three separate problems: May 19 03:25:39 lcg-lrz-dc10 kernel: [903095.585150] megasas: megasas_aen_polling waiting for controller reset to finish for scsi0 May 19 03:25:50 lcg-lrz-dc10 kernel: [903106.581205] sd 0:0:14:0: Device offlined - not ready after error recovery I don't know if that's controller related or drive related. In either case it's hardware related. And then: May 28 16:40:43 lcg-lrz-dc10 kernel: [1727608.170703] BTRFS: bdev /dev/sdm errs: wr 12, rd 0, flush 0, corrupt 0, gen 0 May 28 16:40:50 lcg-lrz-dc10 kernel: [1727615.608552] BTRFS: bdev /dev/sdm errs: wr 12, rd 1, flush 0, corrupt 0, gen 0 ... May 28 16:43:16 lcg-lrz-dc10 kernel: [1727761.077607] BTRFS: bdev /dev/sdm errs: wr 28, rd 21596, flush 0, corrupt 0, gen 0 This is just the fs saying it can't write to one particular drive, and then also many read failures. And then: May 28 21:03:06 lcg-lrz-dc10 kernel: [1743336.369569] BTRFS: lost page write due to I/O error on /dev/sdm May 28 21:03:07 lcg-lrz-dc10 kernel: [1743337.093299] sd 0:0:14:0: rejecting I/O to offline device May 28 21:03:07 lcg-lrz-dc10 kernel: [1743337.094348] BTRFS (device sdp): bad tree block start 3328214216270427953 3448651776 So another lost write to the same drive, sdm, and then new problem which is bad tree block on a different drive sdp. And then: May 28 21:03:07 lcg-lrz-dc10 kernel: [1743337.096927] BTRFS: error -5 while searching for dev_stats item for device /dev/sdm! May 28 21:03:07 lcg-lrz-dc10 kernel: [1743337.097314] BTRFS warning (device sdp): Skipping commit of aborted transaction. It still hasn't given up on sdm (which seems kinda odd by now that there are thousands of read errors and the kernel thinks it's offline anyway), but then now has to deal with problems with sdp. The resulting stack trace though suggests a umount was in progress? May 28 22:55:56 lcg-lrz-dc10 kernel: [1750099.616565] CPU: 4 PID: 134844 Comm: umount Tainted: GW 4.0.0-trunk-amd64 #1 Debian 4.0-1~exp1 https://bugs.launchpad.net/ubuntu/+source/linux/+bug/891115 That's an old bug, kernel 3.2 era. But ultimately it looks like it was hardware related. Chris Murphy -- To unsubscribe from this list: send the line unsubscribe linux-btrfs in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html