Re: [zfs-discuss] ZFS disk failure question
Thank you for your follow-up. The doc looks great. Having good examples goes a long way to helping others that have my problem. Ideally, the replacement would all happen magically, and I would have had everything marked as good, with one failed disk (like a certain other storage vendor that has it's beefs with Sun does). But, I can live with detaching them if I have to. Another thing that would be nice would be to receive notification of disk failures from the OS via email or SMS (like the vendor I previously alluded to), but I know I'm talking crazy now. Jason On Thu, Oct 22, 2009 at 2:15 PM, Cindy Swearingen cindy.swearin...@sun.com wrote: Hi Jason, Since spare replacement is an important process, I've rewritten this section to provide 3 main examples, here: http://docs.sun.com/app/docs/doc/817-2271/gcvcw?a=view Scroll down the section: Activating and Deactivating Hot Spares in Your Storage Pool Example 4–7 Manually Replacing a Disk With a Hot Spare Example 4–8 Detaching a Hot Spare After the Failed Disk is Replaced Example 4–9 Detaching a Failed Disk and Using the Hot Spare The third example is your scenario. I finally listened to the answer, which is you must detach the original disk if you want to continue to use the spare and replace the original disk later. It all works as described. I see some other improvements coming with spare replacement and will provide details when they are available. Thanks, Cindy On 10/14/09 15:54, Jason Frank wrote: See, I get overly literal when working on failed production storage (and yes, I do have backups...) I wasn't wanting to cancel the in-progress spare replacement. I had a completed spare replacement, and I wanted to make it official. So, that didn't really fit my scenario either. I'm glad you agree on the brevity of the detach subcommand man page. I would guess that the intricacies of the failure modes would probably lend itself to richer content than a man page. I'd really like to see some kind of web based wizard to walk through it I doubt I'd get motivated to write it myself though. The web page Cindy pointed to does not cover how to make the replacement official either. It gets close. But at the end, it detaches the hot spare, and not the original disk. Everything seems to be close, but not quite there. Of course, now that I've been through this once, I'll remember all. I'm just thinking of the children. Also, I wanted to try and reconstruct all of my steps from zpool history -i tank. According to that, zpool decided to replace t7 with t11 this morning (why wasn't it last night?), and I offlined, onlined and detach of t7 and I was OK. I did notice that the history records internal scrubs, but not resilvers, It also doesn't record failed commands, or disk failures in a zpool. It would be sweet to have a line that said something like marking vdev /dev/dsk/c8t7d0s0 as UNAVAIL due to X read errors in Y minutes, Then we can really see what happened. Jason On Wed, Oct 14, 2009 at 4:32 PM, Eric Schrock eric.schr...@sun.com wrote: On 10/14/09 14:26, Jason Frank wrote: Thank you, that did the trick. That's not terribly obvious from the man page though. The man page says it detaches the devices from a mirror, and I had a raidz2. Since I'm messing with production data, I decided I wasn't going to chance it when I was reading the man page. You might consider changing the man page, and explaining a little more what it means, maybe even what the circumstances look like where you might use it. This is covered in the Hot Spares section of the manpage: An in-progress spare replacement can be cancelled by detach- ing the hot spare. If the original faulted device is detached, then the hot spare assumes its place in the confi- guration, and is removed from the spare list of all active pools. It is true that the description for zpool detach is overly brief and could be expanded to include this use case. - Eric -- Eric Schrock, Fishworks http://blogs.sun.com/eschrock ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Re: [zfs-discuss] ZFS disk failure question
Thank you, that did the trick. That's not terribly obvious from the man page though. The man page says it detaches the devices from a mirror, and I had a raidz2. Since I'm messing with production data, I decided I wasn't going to chance it when I was reading the man page. You might consider changing the man page, and explaining a little more what it means, maybe even what the circumstances look like where you might use it. Actually, an official and easily searchable What to do when you have a zfs disk failure with lots of examples would be great. There are a lot of attempts out there, but nothing I've found is comprehensive. Jason On Wed, Oct 14, 2009 at 4:23 PM, Eric Schrock eric.schr...@sun.com wrote: On 10/14/09 14:17, Cindy Swearingen wrote: Hi Jason, I think you are asking how do you tell ZFS that you want to replace the failed disk c8t7d0 with the spare, c8t11d0? I just tried do this on my Nevada build 124 lab system, simulating a disk failure and using zpool replace to replace the failed disk with the spare. The spare is now busy and it fails. This has to be a bug. You need to 'zpool detach' the original (c8t7d0). - Eric Another way to recover is if you have a replacement disk for c8t7d0, like this: 1. Physically replace c8t7d0. You might have to unconfigure the disk first. It depends on the hardware. 2. Tell ZFS that you replaced it. # zpool replace tank c8t7d0 3. Detach the spare. # zpool detach tank c8t11d0 4. Clear the pool or the device specifically. # zpool clear tank c8t7d0 Cindy On 10/14/09 14:44, Jason Frank wrote: So, my Areca controller has been complaining via email of read errors for a couple days on SATA channel 8. The disk finally gave up last night at 17:40. I got to say I really appreciate the Areca controller taking such good care of me. For some reason, I wasn't able to log into the server last night or in the morning, probably because my home dir was on the zpool with the failed disk (although it's a raidz2, so I don't know why that was a problem.) So, I went ahead and rebooted it the hard way this morning. The reboot went OK, and I was able to get access to my home directory by waiting about 5 minutes after authenticating. I checked my zpool, and it was resilvering. But, it had only been running for a few minutes. Evidently, it didn't start resilvering until I rebooted it. I would have expected it to do that when the disk failed last night (I had set up a hot spare disk already). All of the zpool commands were taking minutes to complete while c8t7d0 was UNAVAIL, so I offline'd it. When I say all, that includes iostat, status, upgrade, just about anything non-destructive that I could try. That was a little odd. Once I offlined the drive, my resilver restarted, which surprised me. After all, I simply changed an UNAVAIL drive to OFFLINE, in either case, you can't use it for operations. But no big deal there. That fixed the login slowness and the zpool command slowness. The resilver completed, and now I'm left with the following zpool config. I'm not sure how to get things back to normal though, and I hate to do something stupid... r...@datasrv1:~# zpool status tank pool: tank state: DEGRADED scrub: scrub stopped after 0h10m with 0 errors on Wed Oct 14 15:23:06 2009 config: NAME STATE READ WRITE CKSUM tank DEGRADED 0 0 0 raidz2 DEGRADED 0 0 0 c8t0d0 ONLINE 0 0 0 c8t1d0 ONLINE 0 0 0 c8t2d0 ONLINE 0 0 0 c8t3d0 ONLINE 0 0 0 c8t4d0 ONLINE 0 0 0 c8t5d0 ONLINE 0 0 0 c8t6d0 ONLINE 0 0 0 spare DEGRADED 0 0 0 c8t7d0 REMOVED 0 0 0 c8t11d0 ONLINE 0 0 0 c8t8d0 ONLINE 0 0 0 c8t9d0 ONLINE 0 0 0 c8t10d0 ONLINE 0 0 0 spares c8t11d0 INUSE currently in use Since it's not obvious, the spare line had both t7 and t11 indented under it. When the resilver completed, I yanked the hard drive on target 7. I'm assuming that t11 has the same content as t7, but that's not necessarily clear from the output above. So, now I'm left with the following config. I can't zfs remove t7, because it's not a hot spare or a cache disk. I can't zfs replace t7 with t11, I'm told that t11 is busy. And I didn't see any other zpool subcommands that look likely to fix the problem. Here are my system details: SunOS datasrv1 5.11 snv_118 i86pc i386 i86xpv Solaris This system is currently running ZFS pool version 16. Pool 'tank' is already formatted using the current version. How do I tell the system that t11
Re: [zfs-discuss] ZFS disk failure question
See, I get overly literal when working on failed production storage (and yes, I do have backups...) I wasn't wanting to cancel the in-progress spare replacement. I had a completed spare replacement, and I wanted to make it official. So, that didn't really fit my scenario either. I'm glad you agree on the brevity of the detach subcommand man page. I would guess that the intricacies of the failure modes would probably lend itself to richer content than a man page. I'd really like to see some kind of web based wizard to walk through it I doubt I'd get motivated to write it myself though. The web page Cindy pointed to does not cover how to make the replacement official either. It gets close. But at the end, it detaches the hot spare, and not the original disk. Everything seems to be close, but not quite there. Of course, now that I've been through this once, I'll remember all. I'm just thinking of the children. Also, I wanted to try and reconstruct all of my steps from zpool history -i tank. According to that, zpool decided to replace t7 with t11 this morning (why wasn't it last night?), and I offlined, onlined and detach of t7 and I was OK. I did notice that the history records internal scrubs, but not resilvers, It also doesn't record failed commands, or disk failures in a zpool. It would be sweet to have a line that said something like marking vdev /dev/dsk/c8t7d0s0 as UNAVAIL due to X read errors in Y minutes, Then we can really see what happened. Jason On Wed, Oct 14, 2009 at 4:32 PM, Eric Schrock eric.schr...@sun.com wrote: On 10/14/09 14:26, Jason Frank wrote: Thank you, that did the trick. That's not terribly obvious from the man page though. The man page says it detaches the devices from a mirror, and I had a raidz2. Since I'm messing with production data, I decided I wasn't going to chance it when I was reading the man page. You might consider changing the man page, and explaining a little more what it means, maybe even what the circumstances look like where you might use it. This is covered in the Hot Spares section of the manpage: An in-progress spare replacement can be cancelled by detach- ing the hot spare. If the original faulted device is detached, then the hot spare assumes its place in the confi- guration, and is removed from the spare list of all active pools. It is true that the description for zpool detach is overly brief and could be expanded to include this use case. - Eric -- Eric Schrock, Fishworks http://blogs.sun.com/eschrock ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
[zfs-discuss] ZFS disk failure question
So, my Areca controller has been complaining via email of read errors for a couple days on SATA channel 8. The disk finally gave up last night at 17:40. I got to say I really appreciate the Areca controller taking such good care of me. For some reason, I wasn't able to log into the server last night or in the morning, probably because my home dir was on the zpool with the failed disk (although it's a raidz2, so I don't know why that was a problem.) So, I went ahead and rebooted it the hard way this morning. The reboot went OK, and I was able to get access to my home directory by waiting about 5 minutes after authenticating. I checked my zpool, and it was resilvering. But, it had only been running for a few minutes. Evidently, it didn't start resilvering until I rebooted it. I would have expected it to do that when the disk failed last night (I had set up a hot spare disk already). All of the zpool commands were taking minutes to complete while c8t7d0 was UNAVAIL, so I offline'd it. When I say all, that includes iostat, status, upgrade, just about anything non-destructive that I could try. That was a little odd. Once I offlined the drive, my resilver restarted, which surprised me. After all, I simply changed an UNAVAIL drive to OFFLINE, in either case, you can't use it for operations. But no big deal there. That fixed the login slowness and the zpool command slowness. The resilver completed, and now I'm left with the following zpool config. I'm not sure how to get things back to normal though, and I hate to do something stupid... r...@datasrv1:~# zpool status tank pool: tank state: DEGRADED scrub: scrub stopped after 0h10m with 0 errors on Wed Oct 14 15:23:06 2009 config: NAME STATE READ WRITE CKSUM tank DEGRADED 0 0 0 raidz2 DEGRADED 0 0 0 c8t0d0 ONLINE 0 0 0 c8t1d0 ONLINE 0 0 0 c8t2d0 ONLINE 0 0 0 c8t3d0 ONLINE 0 0 0 c8t4d0 ONLINE 0 0 0 c8t5d0 ONLINE 0 0 0 c8t6d0 ONLINE 0 0 0 spare DEGRADED 0 0 0 c8t7d0 REMOVED 0 0 0 c8t11d0 ONLINE 0 0 0 c8t8d0 ONLINE 0 0 0 c8t9d0 ONLINE 0 0 0 c8t10d0ONLINE 0 0 0 spares c8t11d0 INUSE currently in use Since it's not obvious, the spare line had both t7 and t11 indented under it. When the resilver completed, I yanked the hard drive on target 7. I'm assuming that t11 has the same content as t7, but that's not necessarily clear from the output above. So, now I'm left with the following config. I can't zfs remove t7, because it's not a hot spare or a cache disk. I can't zfs replace t7 with t11, I'm told that t11 is busy. And I didn't see any other zpool subcommands that look likely to fix the problem. Here are my system details: SunOS datasrv1 5.11 snv_118 i86pc i386 i86xpv Solaris This system is currently running ZFS pool version 16. Pool 'tank' is already formatted using the current version. How do I tell the system that t11 is the replacement for t7, and how to I then add t7 as the hot spare (after I replace the disk)? Thanks -- This message posted from opensolaris.org ___ zfs-discuss mailing list zfs-discuss@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/zfs-discuss