On Thu, Jun 7, 2018 at 8:58 PM Alfredo Deza <adeza@xxxxxxxxxx> wrote: > > On Thu, Jun 7, 2018 at 2:45 PM, Dan van der Ster <dan@xxxxxxxxxxxxxx> wrote: > > On Thu, Jun 7, 2018 at 6:58 PM Alfredo Deza <adeza@xxxxxxxxxx> wrote: > >> > >> On Thu, Jun 7, 2018 at 12:09 PM, Sage Weil <sweil@xxxxxxxxxx> wrote: > >> > On Thu, 7 Jun 2018, Dan van der Ster wrote: > >> >> On Thu, Jun 7, 2018 at 5:36 PM Dan van der Ster <dan@xxxxxxxxxxxxxx> wrote: > >> >> > > >> >> > On Thu, Jun 7, 2018 at 5:34 PM Sage Weil <sweil@xxxxxxxxxx> wrote: > >> >> > > > >> >> > > On Thu, 7 Jun 2018, Dan van der Ster wrote: > >> >> > > > On Thu, Jun 7, 2018 at 4:41 PM Sage Weil <sweil@xxxxxxxxxx> wrote: > >> >> > > > > > >> >> > > > > On Thu, 7 Jun 2018, Dan van der Ster wrote: > >> >> > > > > > On Thu, Jun 7, 2018 at 4:33 PM Sage Weil <sweil@xxxxxxxxxx> wrote: > >> >> > > > > > > > >> >> > > > > > > On Thu, 7 Jun 2018, Dan van der Ster wrote: > >> >> > > > > > > > Hi all, > >> >> > > > > > > > > >> >> > > > > > > > We have an intermittent issue where bluestore osds sometimes fail to > >> >> > > > > > > > start after a reboot. > >> >> > > > > > > > The osds all fail the same way [see 2], failing to open the superblock. > >> >> > > > > > > > One one particular host, there are 24 osds and 4 SSDs partitioned for > >> >> > > > > > > > the block.db's. The affected non-starting OSDs all have block.db on > >> >> > > > > > > > the same ssd (/dev/sdaa). > >> >> > > > > > > > > >> >> > > > > > > > The osds are all running 12.2.5 on latest centos 7.5 and were created > >> >> > > > > > > > by ceph-volume lvm, e.g. see [1]. > >> >> > > > > > > > > >> >> > > > > > > > This seems like a permissions or similar issue related to the > >> >> > > > > > > > ceph-volume tooling. > >> >> > > > > > > > Any clues how to debug this further? > >> >> > > > > > > > >> >> > > > > > > I take it the OSDs start up if you try again? > >> >> > > > > > > >> >> > > > > > Hey. > >> >> > > > > > No, they don't. For example, we do this `ceph-volume lvm activate 48 > >> >> > > > > > 99fd8e36-fc4d-4bbc-83d9-f5e611cde4b5` several times and its the same > >> >> > > > > > mount failure every time. > >> >> > > > > > >> >> > > > > That sounds like a bluefs bug then, not a ceph-volume issue. Can you > >> >> > > > > try to start the OSD will logging enabled? (debug bluefs = 20, > >> >> > > > > debug bluestore = 20) > >> >> > > > > > >> >> > > > > >> >> > > > Here: https://pastebin.com/TJXZhfcY > >> >> > > > > >> >> > > > Is it supposed to print something about the block.db at some point???? > >> >> > > > >> >> > > Can you dump the bluefs superblock for me? > >> >> > > > >> >> > > dd if=/dev/sdaa1 of=/tmp/foo bs=4K skip=1 count=1 > >> >> > > hexdump -C /tmp/foo > >> >> > > > >> >> > > >> >> > [17:35][root@p06253939y61826 (qa:ceph/dwight/osd*18) ~]# dd > >> >> > if=/dev/sdaa1 of=/tmp/foo bs=4K skip=1 count=1 > >> >> > 1+0 records in > >> >> > 1+0 records out > >> >> > 4096 bytes (4.1 kB) copied, 0.000320003 s, 12.8 MB/s > >> >> > [17:35][root@p06253939y61826 (qa:ceph/dwight/osd*18) ~]# hexdump -C /tmp/foo > >> >> > 00000000 01 01 5d 00 00 00 11 fb be 4d 43 31 4a b5 a4 cb |..]......MC1J...| > >> >> > 00000010 99 be b7 da 72 ca 99 fd 8e 36 fc 4d 4b bc 83 d9 |....r....6.MK...| > >> >> > 00000020 f5 e6 11 cd e4 b5 1d 00 00 00 00 00 00 00 00 10 |................| > >> >> > 00000030 00 00 01 01 2b 00 00 00 01 80 80 40 00 00 00 00 |....+......@....| > >> >> > 00000040 00 00 00 00 00 02 00 00 00 01 01 07 00 00 00 eb |................| > >> >> > 00000050 b2 00 00 83 08 01 01 01 07 00 00 00 cb b2 00 00 |................| > >> >> > 00000060 83 20 01 61 6d 07 be 00 00 00 00 00 00 00 00 00 |. .am...........| > >> >> > 00000070 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................| > >> >> > * > >> >> > 00001000 > >> >> > > >> >> > > >> >> > >> >> Wait, we found something!!! > >> >> > >> >> In the 1st 4k on the block we found the block.db pointing at the wrong > >> >> device (/dev/sdc1 instead of /dev/sdaa1) > >> >> > >> >> 00000130 6b 35 79 2b 67 3d 3d 0d 00 00 00 70 61 74 68 5f |k5y+g==....path_| > >> >> 00000140 62 6c 6f 63 6b 2e 64 62 09 00 00 00 2f 64 65 76 |block.db..../dev| > >> >> 00000150 2f 73 64 63 31 05 00 00 00 72 65 61 64 79 05 00 |/sdc1....ready..| > >> >> 00000160 00 00 72 65 61 64 79 06 00 00 00 77 68 6f 61 6d |..ready....whoam| > >> >> 00000170 69 02 00 00 00 34 38 eb c2 d7 d6 00 00 00 00 00 |i....48.........| > >> >> 00000180 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................| > >> >> > >> >> It is similarly wrong for another broken osd.53 (block.db is /dev/sdc2 > >> >> instead of /dev/sdaa2). > >> >> And for the osds that are running, that block.db is correct! > >> >> > >> >> So.... the block.db device is persisted in the block header? But after > >> >> a reboot it gets a new name. (sd* naming is famously chaotic). > >> >> ceph-volume creates a softlink to the correct db dev, but it seems not used? > >> > > >> > Aha, yes.. the bluestore startup code looks for the value in the > >> > superblock before the on in the directory. > >> > > >> > We can either (1) reverse that order, (and/)or (2) make ceph-volume use a > >> > stable path for the device name when creating the bluestore. And/or (3) > >> > use ceph-bluestore-tool set-label-key to fix it if it doesn't match (this > >> > would repair old superblocks... permanently if we use the stable path > >> > name). > >> > >> ceph-volume does not require a stable/persistent path name at all. For > >> partitions we store the partuuid, and *always* make sure that we have > >> the right device, because we query blkid. > >> > > > > ceph-disk didn't require stable path names either, so it's good to > > hear that this feature is still there in ceph-volume lvm. > > The only requirement being the PARTUUID has to be present (if wanting > to use a partition). > > > It's been a bit unclear how to map the concept of a partitioned SSD > > for filestore journals to the bluestore world. > > As I understand it now, what we did is correct and fully supported? > > (see the parted routine earlier in the thread...) > > It is fully supported, and we do test this, we just weren't able to > hit it. Again, another option that > would prevent this sort of issue would be to use an LV instead of a partition. Sage mentioned earlier that normally ceph-volume creates an LV for the db or wal device. How do we trigger that feature? I've understood from the docs that a LV for db or wal would be up to the operator to pre-create and pass to c-v lvm create... Cheers, Dan > > > > > Thanks! > > > > Dan > > > > > >> In addition to that, in the case of bluestore, we go by each device > >> and ensure that whatever link was done by bluestore is corrected > >> before attempting to start the OSD [0] > >> > >> IMO bluestore should do #1, because this is already solved in the > >> ceph-volume code (we knew dev names could change), but #2 and #3 are > >> OK to help with this issue today. > >> > >> Another option would be to just avoid using the partition for block.db > >> and just use an LV. > >> > >> [0] https://github.com/ceph/ceph/blob/master/src/ceph-volume/ceph_volume/devices/lvm/activate.py#L155-L168 > >> > >> > >> > > >> > sage > >> > > >> > > >> >> > >> >> ... > >> >> Dan & Teo > >> >> > >> >> > >> >> > >> >> > >> >> > >> >> > > >> >> > -- dan > >> >> > > >> >> > > >> >> > > >> >> > > >> >> > > >> >> > > >> >> > > >> >> > > >> >> > > >> >> > > Thanks! > >> >> > > sage > >> >> > > > >> >> > > > > >> >> > > > Here's the osd dir: > >> >> > > > > >> >> > > > # ls -l /var/lib/ceph/osd/ceph-48/ > >> >> > > > total 24 > >> >> > > > lrwxrwxrwx. 1 ceph ceph 93 Jun 7 16:46 block -> > >> >> > > > /dev/ceph-34f24306-d90c-49ff-bafb-2657a6a18010/osd-block-99fd8e36-fc4d-4bbc-83d9-f5e611cde4b5 > >> >> > > > lrwxrwxrwx. 1 root root 10 Jun 7 16:46 block.db -> /dev/sdaa1 > >> >> > > > -rw-------. 1 ceph ceph 37 Jun 7 16:46 ceph_fsid > >> >> > > > -rw-------. 1 ceph ceph 37 Jun 7 16:46 fsid > >> >> > > > -rw-------. 1 ceph ceph 56 Jun 7 16:46 keyring > >> >> > > > -rw-------. 1 ceph ceph 6 Jun 7 16:46 ready > >> >> > > > -rw-------. 1 ceph ceph 10 Jun 7 16:46 type > >> >> > > > -rw-------. 1 ceph ceph 3 Jun 7 16:46 whoami > >> >> > > > > >> >> > > > # ls -l /dev/ceph-34f24306-d90c-49ff-bafb-2657a6a18010/osd-block-99fd8e36-fc4d-4bbc-83d9-f5e611cde4b5 > >> >> > > > lrwxrwxrwx. 1 root root 7 Jun 7 16:46 > >> >> > > > /dev/ceph-34f24306-d90c-49ff-bafb-2657a6a18010/osd-block-99fd8e36-fc4d-4bbc-83d9-f5e611cde4b5 > >> >> > > > -> ../dm-4 > >> >> > > > > >> >> > > > # ls -l /dev/dm-4 > >> >> > > > brw-rw----. 1 ceph ceph 253, 4 Jun 7 16:46 /dev/dm-4 > >> >> > > > > >> >> > > > > >> >> > > > --- Logical volume --- > >> >> > > > LV Path > >> >> > > > /dev/ceph-34f24306-d90c-49ff-bafb-2657a6a18010/osd-block-99fd8e36-fc4d-4bbc-83d9-f5e611cde4b5 > >> >> > > > LV Name osd-block-99fd8e36-fc4d-4bbc-83d9-f5e611cde4b5 > >> >> > > > VG Name ceph-34f24306-d90c-49ff-bafb-2657a6a18010 > >> >> > > > LV UUID FQkRxS-No7X-ajkP-5L3N-K22a-IXg6-QLceZC > >> >> > > > LV Write Access read/write > >> >> > > > LV Creation host, time p06253939y61826.cern.ch, 2018-03-15 10:57:37 +0100 > >> >> > > > LV Status available > >> >> > > > # open 0 > >> >> > > > LV Size <5.46 TiB > >> >> > > > Current LE 1430791 > >> >> > > > Segments 1 > >> >> > > > Allocation inherit > >> >> > > > Read ahead sectors auto > >> >> > > > - currently set to 256 > >> >> > > > Block device 253:4 > >> >> > > > > >> >> > > > --- Physical volume --- > >> >> > > > PV Name /dev/sda > >> >> > > > VG Name ceph-34f24306-d90c-49ff-bafb-2657a6a18010 > >> >> > > > PV Size <5.46 TiB / not usable <2.59 MiB > >> >> > > > Allocatable yes (but full) > >> >> > > > PE Size 4.00 MiB > >> >> > > > Total PE 1430791 > >> >> > > > Free PE 0 > >> >> > > > Allocated PE 1430791 > >> >> > > > PV UUID WP0Z7C-ejSh-fpSa-a73N-H2Hz-yC78-qBezcI > >> >> > > > > >> >> > > > (sorry for wall o' lvm) > >> >> > > > > >> >> > > > -- dan > >> >> > > > > >> >> > > > > >> >> > > > > >> >> > > > > >> >> > > > > >> >> > > > > >> >> > > > > >> >> > > > > >> >> > > > > Thanks! > >> >> > > > > sage > >> >> > > > > > >> >> > > > > > >> >> > > > > > -- dan > >> >> > > > > > > >> >> > > > > > > >> >> > > > > > > > >> >> > > > > > > sage > >> >> > > > > > > > >> >> > > > > > > > >> >> > > > > > > > > >> >> > > > > > > > Thanks! > >> >> > > > > > > > > >> >> > > > > > > > Dan > >> >> > > > > > > > > >> >> > > > > > > > [1] > >> >> > > > > > > > > >> >> > > > > > > > ====== osd.48 ====== > >> >> > > > > > > > > >> >> > > > > > > > [block] /dev/ceph-34f24306-d90c-49ff-bafb-2657a6a18010/osd-block-99fd8e36-fc4d-4bbc-83d9-f5e611cde4b5 > >> >> > > > > > > > > >> >> > > > > > > > type block > >> >> > > > > > > > osd id 48 > >> >> > > > > > > > cluster fsid dd535a7e-4647-4bee-853d-f34112615f81 > >> >> > > > > > > > cluster name ceph > >> >> > > > > > > > osd fsid 99fd8e36-fc4d-4bbc-83d9-f5e611cde4b5 > >> >> > > > > > > > db device /dev/sdaa1 > >> >> > > > > > > > encrypted 0 > >> >> > > > > > > > db uuid 3381a121-1c1b-4e45-a986-c1871c363edc > >> >> > > > > > > > cephx lockbox secret > >> >> > > > > > > > block uuid FQkRxS-No7X-ajkP-5L3N-K22a-IXg6-QLceZC > >> >> > > > > > > > block device > >> >> > > > > > > > /dev/ceph-34f24306-d90c-49ff-bafb-2657a6a18010/osd-block-99fd8e36-fc4d-4bbc-83d9-f5e611cde4b5 > >> >> > > > > > > > crush device class None > >> >> > > > > > > > > >> >> > > > > > > > [ db] /dev/sdaa1 > >> >> > > > > > > > > >> >> > > > > > > > PARTUUID 3381a121-1c1b-4e45-a986-c1871c363edc > >> >> > > > > > > > > >> >> > > > > > > > > >> >> > > > > > > > > >> >> > > > > > > > [2] > >> >> > > > > > > > -11> 2018-06-07 16:12:16.138407 7fba30fb4d80 1 -- - start start > >> >> > > > > > > > -10> 2018-06-07 16:12:16.138516 7fba30fb4d80 1 > >> >> > > > > > > > bluestore(/var/lib/ceph/osd/ceph-48) _mount path /var/lib/ceph/os > >> >> > > > > > > > d/ceph-48 > >> >> > > > > > > > -9> 2018-06-07 16:12:16.138801 7fba30fb4d80 1 bdev create path > >> >> > > > > > > > /var/lib/ceph/osd/ceph-48/block type kernel > >> >> > > > > > > > -8> 2018-06-07 16:12:16.138808 7fba30fb4d80 1 bdev(0x55eb46433a00 > >> >> > > > > > > > /var/lib/ceph/osd/ceph-48/block) open path /v > >> >> > > > > > > > ar/lib/ceph/osd/ceph-48/block > >> >> > > > > > > > -7> 2018-06-07 16:12:16.138999 7fba30fb4d80 1 bdev(0x55eb46433a00 > >> >> > > > > > > > /var/lib/ceph/osd/ceph-48/block) open size 60 > >> >> > > > > > > > 01172414464 (0x57541c00000, 5589 GB) block_size 4096 (4096 B) rotational > >> >> > > > > > > > -6> 2018-06-07 16:12:16.139188 7fba30fb4d80 1 > >> >> > > > > > > > bluestore(/var/lib/ceph/osd/ceph-48) _set_cache_sizes cache_size > >> >> > > > > > > > 134217728 meta 0.01 kv 0.99 data 0 > >> >> > > > > > > > -5> 2018-06-07 16:12:16.139275 7fba30fb4d80 1 bdev create path > >> >> > > > > > > > /var/lib/ceph/osd/ceph-48/block type kernel > >> >> > > > > > > > -4> 2018-06-07 16:12:16.139281 7fba30fb4d80 1 bdev(0x55eb46433c00 > >> >> > > > > > > > /var/lib/ceph/osd/ceph-48/block) open path /v > >> >> > > > > > > > ar/lib/ceph/osd/ceph-48/block > >> >> > > > > > > > -3> 2018-06-07 16:12:16.139454 7fba30fb4d80 1 bdev(0x55eb46433c00 > >> >> > > > > > > > /var/lib/ceph/osd/ceph-48/block) open size 60 > >> >> > > > > > > > 01172414464 (0x57541c00000, 5589 GB) block_size 4096 (4096 B) rotational > >> >> > > > > > > > -2> 2018-06-07 16:12:16.139464 7fba30fb4d80 1 bluefs > >> >> > > > > > > > add_block_device bdev 1 path /var/lib/ceph/osd/ceph-48/blo > >> >> > > > > > > > ck size 5589 GB > >> >> > > > > > > > -1> 2018-06-07 16:12:16.139510 7fba30fb4d80 1 bluefs mount > >> >> > > > > > > > 0> 2018-06-07 16:12:16.142930 7fba30fb4d80 -1 > >> >> > > > > > > > /home/jenkins-build/build/workspace/ceph-build/ARCH/x86_64/AVAILA > >> >> > > > > > > > BLE_ARCH/x86_64/AVAILABLE_DIST/centos7/DIST/centos7/MACHINE_SIZE/huge/release/12.2.5/rpm/el7/BUILD/ceph-12.2.5/src/o > >> >> > > > > > > > s/bluestore/bluefs_types.h: In function 'static void > >> >> > > > > > > > bluefs_fnode_t::_denc_finish(ceph::buffer::ptr::iterator&, __u8 > >> >> > > > > > > > *, __u8*, char**, uint32_t*)' thread 7fba30fb4d80 time 2018-06-07 > >> >> > > > > > > > 16:12:16.139666 > >> >> > > > > > > > /home/jenkins-build/build/workspace/ceph-build/ARCH/x86_64/AVAILABLE_ARCH/x86_64/AVAILABLE_DIST/centos7/DIST/centos7/MACHINE_SIZE/huge/release/12.2.5/rpm/el7/BUILD/ceph-12.2.5/src/os/bluestore/bluefs_types.h: > >> >> > > > > > > > 54: FAILED assert(pos <= end) > >> >> > > > > > > > > >> >> > > > > > > > ceph version 12.2.5 (cad919881333ac92274171586c827e01f554a70a) > >> >> > > > > > > > luminous (stable) > >> >> > > > > > > > 1: (ceph::__ceph_assert_fail(char const*, char const*, int, char > >> >> > > > > > > > const*)+0x110) [0x55eb3b597780] > >> >> > > > > > > > 2: (bluefs_super_t::decode(ceph::buffer::list::iterator&)+0x776) > >> >> > > > > > > > [0x55eb3b52db36] > >> >> > > > > > > > 3: (BlueFS::_open_super()+0xfe) [0x55eb3b50cede] > >> >> > > > > > > > 4: (BlueFS::mount()+0xe3) [0x55eb3b5250c3] > >> >> > > > > > > > 5: (BlueStore::_open_db(bool)+0x173d) [0x55eb3b43ebcd] > >> >> > > > > > > > 6: (BlueStore::_mount(bool)+0x40e) [0x55eb3b47025e] > >> >> > > > > > > > 7: (OSD::init()+0x3bd) [0x55eb3b02a1cd] > >> >> > > > > > > > 8: (main()+0x2d07) [0x55eb3af2f977] > >> >> > > > > > > > 9: (__libc_start_main()+0xf5) [0x7fba2d47b445] > >> >> > > > > > > > 10: (()+0x4b7033) [0x55eb3afce033] > >> >> > > > > > > > NOTE: a copy of the executable, or `objdump -rdS <executable>` is > >> >> > > > > > > > needed to interpret this. > >> >> > > > > > > > _______________________________________________ > >> >> > > > > > > > ceph-users mailing list > >> >> > > > > > > > ceph-users@xxxxxxxxxxxxxx > >> >> > > > > > > > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > >> >> > > > > > > > > >> >> > > > > > > > > >> >> > > > > > _______________________________________________ > >> >> > > > > > ceph-users mailing list > >> >> > > > > > ceph-users@xxxxxxxxxxxxxx > >> >> > > > > > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com > >> >> > > > > > > >> >> > > > > > > >> >> > > > > >> >> > > > > >> >> > >> >> > >> > _______________________________________________ > >> > ceph-users mailing list > >> > ceph-users@xxxxxxxxxxxxxx > >> > http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com _______________________________________________ ceph-users mailing list ceph-users@xxxxxxxxxxxxxx http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com