I finally had a chance to do the e2fsck (ready only) – see below. Unless you say otherwise, we’ll schedule downtime for the filesystem soon and do this on the actual MDT.
It would be nice to figure out why this happened in the first place. Any advice on how to track this down so we can potentially prevent this in the future? [root@hpfs2-eg3-mds0 ~]# e2fsck -fn /dev/Storage/mdt-testsnap e2fsck 1.42.13.wc5 (15-Apr-2016) Pass 1: Checking inodes, blocks, and sizes Deleted inode 14772127 has zero dtime. Fix? no Deleted inode 85060890 has zero dtime. Fix? no Pass 2: Checking directory structure Pass 3: Checking directory connectivity Pass 4: Checking reference counts Inode 23596587 ref count is 1, should be 2. Fix? no Inode 23596588 ref count is 1, should be 2. Fix? no Inode 25254104 ref count is 1, should be 2. Fix? no Inode 25254105 ref count is 1, should be 2. Fix? no Inode 25254138 ref count is 1, should be 2. Fix? no Inode 25254139 ref count is 1, should be 2. Fix? no Inode 25776685 ref count is 1, should be 2. Fix? no Inode 25776686 ref count is 1, should be 2. Fix? no Inode 25776719 ref count is 1, should be 2. Fix? no Inode 26294667 ref count is 1, should be 2. Fix? no Inode 192124340 ref count is 1, should be 3. Fix? no Pass 5: Checking group summary information Inode bitmap differences: -14772127 -85060890 Fix? no [QUOTA WARNING] Usage inconsistent for ID 0:actual (6880575488, 2980) != expected (11033616384, 3960) [QUOTA WARNING] Usage inconsistent for ID 33152:actual (323584, 468) != expected (323584, 469) [QUOTA WARNING] Usage inconsistent for ID 3260:actual (16705683456, 8380739) != expected (16705683456, 8380748) Update quota info for quota type 0? no [QUOTA WARNING] Usage inconsistent for ID 0:actual (6880399360, 2022) != expected (11033440256, 3002) [QUOTA WARNING] Usage inconsistent for ID 3000:actual (31232651264, 41964064) != expected (31232663552, 41963909) Update quota info for quota type 1? no hpfs2eg3-MDT0000: ********** WARNING: Filesystem still has errors ********** hpfs2eg3-MDT0000: 45080951/213319680 files (0.2% non-contiguous), 36292306/106658816 blocks [root@hpfs2-eg3-mds0 ~]# From: "Dilger, Andreas" <andreas.dil...@intel.com> Date: Thursday, September 1, 2016 at 4:25 PM To: Darby Vicker <darby.vicke...@nasa.gov> Cc: "lustre-discuss@lists.lustre.org" <lustre-discuss@lists.lustre.org> Subject: Re: [lustre-discuss] Inaccessible directory Doing a file-level backup of the MDT is mostly useful when migrating from one backing storage to another backing storage device. It isn't really useful to restore individual "files" from an MDT-only backup since they only contain the file metadata. Restoring all of the inodes from the MDT file backup means that all of the inode numbers are changed and a full LFSCK is needed to rebuild the Object Index (OI) files so that the FID->inode mappings are correct. Depending on the size of the filesystem this may take several hours. For disaster recovery purposes, a device-level backup (dd) is more "plug and play" in that the whole image is restored from the backup and the LFSCK phase only needs to handle files that have been modified since the time the backup was created. Cheers, Andreas -- Andreas Dilger Lustre Principal Architect Intel High Performance Data Division On 2016/09/01, 14:49, "Vicker, Darby (JSC-EG311)" <darby.vicke...@nasa.gov> wrote: Thanks. This is happening on all the clients so its not a DLM lock problem. We’ll try the fsck soon. If that come back clean is this likely a hardware corruption? Could this have anything to do with our corruption? Aug 29 12:13:34 hpfs2-eg3-mds0 kernel: LustreError: 0-0: hpfs2eg3-MDT0000: trigger OI scrub by RPC for [0x2000079cd:0x3988:0x0], rc = 0 [1] We make a file level backup of the MDT using the procedure in the lustre manual on a daily basis and keep a history of those. We’ve never had a problem so we’ve never had to restore anything from the backups. Is the device level backup (dd) necessary or is file level sufficient? From: "Dilger, Andreas" <andreas.dil...@intel.com> Date: Thursday, September 1, 2016 at 11:48 AM To: Darby Vicker <darby.vicke...@nasa.gov> Cc: "lustre-discuss@lists.lustre.org" <lustre-discuss@lists.lustre.org> Subject: Re: [lustre-discuss] Inaccessible directory The first thing to try is just cancel the DLM locks on this client, to see if the problem is just a stale lock cached there: lctl set_param ldlm.namespaces.*.lru_size=clear That likely won't help if this same problem is being hit on all of your clients. Next to try is running e2fsck on the MDT. It is recommended to use the most recent e2fsprogs-1.42.12-wc5, since this fixes a number of bugs in older versions. It is also strongly recommended to make a "dd" backup of the whole MDT before the e2fsck run, and to do this on a regular basis. This is convenient in case of MDT corruption (now or in the future), and you can restore only the MDT instead of having to restore 400TB of data as well. I was just updating the Lustre User Manual about this http://review.whamcloud.com/21726 . Running "e2fsck -fn" first can also give you an idea of what kind of corruption is present before Making the fix. Cheers, Andreas On Aug 31, 2016, at 10:54, Vicker, Darby (JSC-EG311) <darby.vicke...@nasa.gov> wrote: Hello, We’ve run into a problem where an entire directory on our lustre file system has become inaccessible. # mount | grep lustre2 192.52.98.142@tcp:/hpfs2eg3 on /lustre2 type lustre (rw,flock) # ls -l /lustre2/mirrors/cpas ls: cannot access /lustre2/mirrors/cpas: Stale file handle # ls -l /lustre2/mirrors/ ls: cannot access /lustre2/mirrors/cpas: Stale file handle 4 drwxr-s--- 5 root g27122 4096 Dec 23 2014 cev-repo/ ? d????????? ? ? ? ? ? cpas/ 4 drwxrwxr-x 5 root eg3 4096 Aug 21 2014 sls/ Fortunately, we have a backup of this directory from about a week ago. However, I would like to figure out how this happened to prevent any further damage. I’m not sure if we’re dealing with corruption in the LFS, damage to the underlying RAID or something else and I’d appreciate some help figuring this out. Here’s some info on our lustre servers: CentOS 6.4 2.6.32-358.23.2.el6_lustre.x86_64 Lustre 2.4.3 (I know - we need to upgrade...) Hardware RAID 10 MDT (Adaptec 6805Q – SSD’s) (19x) OSS’s - Hardware RAID 6 OST’s (Adaptec 6445) 1 27TB OST per OSS, ldiskfs Dual homed via Ethernet and IB Most Ethernet clients (~50 total) are CentOS 7 using lustre-client-2.8.0-3.10.0_327.13.1.el7.x86_64.x86_64. Our compute nodes (~400 total) connect over IB and are still CentOS 6 using lustre-client-2.7.0-2.6.32_358.14.1.el6.x86_64.x86_64. The lustre server hardware is about 4 years old now. All RAID arrays are reported as healthy. Searched JIRA and the mailing lists and couldn’t find anything related. This sounded close at first: https://jira.hpdd.intel.com/browse/LU-3550 But, as shown above, the issue shows up on a native lustre client, not an NFS export. We are exporting this LFS via SMB but I don’t think that’s related. I think the next step is to run an e2fsck but we haven’t done that yet and would appreciate advise on stepping through this. Thanks, Darby _______________________________________________ lustre-discuss mailing list lustre-discuss@lists.lustre.org http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org _______________________________________________ lustre-discuss mailing list lustre-discuss@lists.lustre.org http://lists.lustre.org/listinfo.cgi/lustre-discuss-lustre.org
