I will be substituting that card with another areca, on the running storage, soon. I then will take that card and put it on a test storage machine and make it work a lot to reproduce the problem, so I will be able to run your other checks. And no, I can't generate a crashdump, because the command to do it never returns after going up to 100%...... Thanx a lot. Gabriele. Da: George Wilson A: Gabriele Bulfon Cc: [email protected] Data: 11 dicembre 2012 15.13.05 CET Oggetto: Re: [discuss] Again: illumos based ZFS storage failure So this tells us that the pool is not suspending because of any errors. The next thing to look at is whether or not the I/Os are just not returning. Are you able to generate a crash dump? Here are some other things to try: # echo "::stacks -m zfs" | mdb -kzfs_threads.out # echo "::stacks -c spa_sync" | mdb -k # echo "::zio_state -r" | mdb -kzios.out - George On 12/11/12 6:15 AM, Gabriele Bulfon wrote: I got the problem again today, I had moved the iscsi to the Areca, left only CIFS on the adaptec card, and I have only CIFS not running. Probably this is a problem of the adaptec card itself, but I don't know if it's the combination of the illumos kernel + this card. I have no signal of error anywhere, but zfs command not responding on the filesystem over the adaptec. Even df -h do note returns. I tried the "walk spa" command during failure, but output the same: sonicle@xstorage:~# echo "::walk spa | ::print spa_t spa_name spa_suspended" | mdb -k spa_name = [ "adaptec" ] spa_suspended = 0 spa_name = [ "areca" ] spa_suspended = 0 spa_name = [ "rpool1" ] spa_suspended = 0 Da: George Wilson A: Gabriele Bulfon Cc: [email protected] Data: 27 novembre 2012 15.10.45 CET Oggetto: Re: [discuss] Again: illumos based ZFS storage failure Was this data you provided below from a time when the server was hung? If not, try running this next time you see the issue. - George On 11/23/12 9:44 AM, Gabriele Bulfon wrote: Hi, here is the output. Looks sane: sonicle@xstorage:~# echo "::walk spa | ::print spa_t spa_name spa_suspended" | mdb -k spa_name = [ "adaptec" ] spa_suspended = 0 spa_name = [ "areca" ] spa_suspended = 0 spa_name = [ "rpool1" ] spa_suspended = 0 Da: George Wilson A: [email protected] Cc: Gabriele Bulfon Data: 23 novembre 2012 14.46.55 CET Oggetto: Re: [discuss] Again: illumos based ZFS storage failure It's possible that the adaptec pool has suspended because of some error on the storage. Can you run the following as root and provide the output: echo "::walk spa | ::print spa_t spa_name spa_suspended" | mdb -k - George On 11/23/12 5:43 AM, Gabriele Bulfon wrote: Hi, I got the same problem this morning. Thanks to Alasdair suggestion, I commented out the zfs "quota" command from /etc/profile, so I could enter the bash prompt and investigate. As a summary of the problem: - 3 zfs pools (rpool, areca, adpatec) each on a different controller: rpool as a mirror of internal disks, areca as raidz on 7 sata disks of the areca controller plus half space of SD as zlog, adaptec as raidz on 8 disks of the adaptec controller plus half space of SD as zlog. - areca space is used for NFS sharing to unix servers, and always responds. - adaptec space is used for CIFS sharing and an iScsi volume for the Windows PDC. Then we have a vmware server with an iscsi resource store, given to the virtualized PDC as a secondary disk for sqlserver data and some more. PDC boots directly from the vmware server disks. At once, both CIFS from the storage and PDC iscsi disk do not respond. CIFS fails probably because the PDC AD is not responding, probably busy checking the iscsi disk, in a loop. Going into the storage bash: - zpool status shows everything find, every pool is correct. - /var/adm/messages shows only smbd time outs with the PDC, no hardware or zfs problem. - at the time of failure, fmdump -eVvm showed same previously found errors, 3 days earlier - after rebooting all the infrastructure, fmdump -eVvm showed same previously found errors around the time of rebooting the storage, not at the time of the experienced failures. We find one stated error for each disk of the adaptec controller (cut &paste at the end) - a zfs list areca/* showed all the areca filesystems - a zfs list adaptec blocked never returning - any access to the zfs structure of the adaptec would block - as suggested by Alasdair, I ran savecore -L (I checked to have the dump device and with enough space) - the savecore command ran for sometime until reaching 100%, then blocked never returning. - I could not "init 5" the storage, never returning - I tried sending the poweroff signal with the power button, console showed its intention to power off, but never did it - I forced powered off via the power button. - Once everything was powered on again, everything ran fine. - I looked for the dump into /var/crash, but I had no /var/crash at all. How can I investigate further this problem? Do I have any chance to find the savecore output in the dump device, even if I have not /var/crash? Here is the fmdump output: Nov 23 2012 09:25:04.422282821 ereport.io.scsi.cmd.disk.dev.uderr nvlist version: 0 class = ereport.io.scsi.cmd.disk.dev.uderr ena = 0x2b08d604c300001 detector = (embedded nvlist) nvlist version: 0 version = 0x0 scheme = dev device-path = /pci@0,0/pci8086,3595@2/pci8086,370@0/pci9005,2bc@e/disk@7,0 devid = id1,sd@TAdaptec_3805____________8366BAF3 (end detector) devid = id1,sd@TAdaptec_3805____________8366BAF3 driver-assessment = fail op-code = 0x1a cdb = 0x1a 0x0 0x8 0x0 0x18 0x0 pkt-reason = 0x0 pkt-state = 0x1f pkt-stats = 0x0 stat-code = 0x0 un-decode-info = sd_get_write_cache_enabled: Mode Sense caching page code mismatch 0 __ttl = 0x1 ---------------------------------------------------------------------------------- Da: Alasdair Lumsden A: [email protected] Data: 10 novembre 2012 14.37.20 CET Oggetto: Re: [discuss] illumos based ZFS storage failure I haven't read the whole thread, but the next time it happens you'll want to invoke a panic and make the dump file available. You'll want to ensure that: 1. Multithreaded dump is disabled in /etc/system with: * Disable MT dump set dump_plat_mincpu=0 Without this there is a risk of your dump not saving correctly. 2. That you have a dump device and that it's big enough to capture your kernel size (zfs set volsize=X rpool/dump) 3. That dumpadm is happy and set to save cores etc: dumpadm -y -z on -c kernel -d /dev/zvol/dsk/rpool/dump There's lots of good info here: http://wiki.illumos.org/display/illumos/How+To+Report+Problems You can also inspect things with mdb while the system is up, but if it's a production system normally you want to get it rebooted and into production again ASAP. So in that situation, you can take a dump of the running system with: savecore -L One thing to keep in mind is /etc/profile runs /usr/sbin/quota, which can screw over logins when the zfs subsystem is unhappy. I really think it should be removed by default since on most systems quotas aren't even used. So comment it out - we do so on all our systems. This will give you a better chance of logging in when things go wrong. I think there's a way to SSH in bypassing /etc/profile but I can't remember what it is - perhaps someone can chime in. Good luck. Centralised storage is difficult to do and when it goes wrong everything that depends on it goes down. It's a "all your eggs in one giant failbasket". Doing it homebrew with ZFS is cost effective and can be fast, but it is also risky. This is why there are companies like Nexenta out there with certified combinations of hardware and software engineered to work together. This extends to validating firmware combinations of disks/HBAs/etc. Cheers, Alasdair ------------------------------------------- illumos-discuss Archives: https://www.listbox.com/member/archive/182180/=now RSS Feed: https://www.listbox.com/member/archive/rss/182180/21175541-02f10c6f Modify Your Subscription: https://www.listbox.com/member/?&; Powered by Listbox: http://www.listbox.com illumos-discuss | Archives | Modify Your Subscription
------------------------------------------- illumos-discuss Archives: https://www.listbox.com/member/archive/182180/=now RSS Feed: https://www.listbox.com/member/archive/rss/182180/21175430-2e6923be Modify Your Subscription: https://www.listbox.com/member/?member_id=21175430&id_secret=21175430-6a77cda4 Powered by Listbox: http://www.listbox.com
