Date: Tue, 31 Jan 2017 23:22:22 +0000 From: Steven Hartland <killing@multiplay.co.uk> To: Larry Rosenman <ler@FreeBSD.org> Cc: Freebsd fs <freebsd-fs@freebsd.org> Subject: Re: 16.0E ExpandSize? -- New Server Message-ID: <96534515-4fcb-774e-a599-8d48aec930cd@multiplay.co.uk> In-Reply-To: <ffc24b7bfacd265d637b633566bbaa51@FreeBSD.org> References: <00db0ab7243ce6368c246ae20f9c075a@FreeBSD.org> <1a69057c-dc59-9b78-9762-4f98a071105e@multiplay.co.uk> <ce5a1d39612d694077accda33266a3ab@FreeBSD.org> <ad07e84e-f297-362a-1398-c5503bb56a8d@multiplay.co.uk> <35a9034f91542bb1329ac5104bf3b773@FreeBSD.org> <76fc9505-f681-0de0-fe0c-5624b29de321@multiplay.co.uk> <22e1bfc5840d972cf93643733682cda1@FreeBSD.org> <f2600a53-0dc1-9f41-1405-ed22d96d30cf@multiplay.co.uk> <8a710dc75c129f58b0372eeaeca575b5@FreeBSD.org> <aef02eb0-0888-6fea-a4b8-4033ca56f4a3@multiplay.co.uk> <d3181bd00c827fb99fbcebe6fe097ef8@FreeBSD.org> <a3d78923-5046-11c8-daea-713eacf47bd2@multiplay.co.uk> <ffc24b7bfacd265d637b633566bbaa51@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Yep On 31/01/2017 21:49, Larry Rosenman wrote: > > revert the other patch and apply this one? > > > > On 01/31/2017 3:47 pm, Steven Hartland wrote: > >> Hmm, looks like there's also a bug in the way vdev_min_asize is >> calculated for raidz as it can and has resulted in child min_asize >> which won't provided enough space for the parent due to the use of >> unrounded integer division. >> >> 1981411579221 * 6 = 11888469475326 < 11888469475328 >> >> You should have vdev_min_asize: 1981411579222 for your children. >> >> Updated patch attached, however calculation still isn't 100% >> reversible so may need work, however it does now ensure that the >> children will provide enough capacity for min_asize even if all of >> them are shrunk to their individual min_asize, which I believe >> previously may not have been the case. >> >> This isn't related to the incorrect EXPANDSZ output, but would be >> good if you could confirm it doesn't cause any issues for your pool >> given its state. >> >> On 31/01/2017 21:00, Larry Rosenman wrote: >>> >>> borg-new /home/ler $ sudo ./vdev-stats.d >>> Password: >>> vdev_path: n/a, vdev_max_asize: 0, vdev_asize: 0, vdev_min_asize: 0 >>> vdev_path: n/a, vdev_max_asize: 11947471798272, vdev_asize: >>> 11947478089728, vdev_min_asize: 11888469475328 >>> vdev_path: /dev/mfid4p4, vdev_max_asize: 1991245299712, vdev_asize: >>> 1991245299712, vdev_min_asize: 1981411579221 >>> vdev_path: /dev/mfid0p4, vdev_max_asize: 1991246348288, vdev_asize: >>> 1991246348288, vdev_min_asize: 1981411579221 >>> vdev_path: /dev/mfid1p4, vdev_max_asize: 1991246348288, vdev_asize: >>> 1991246348288, vdev_min_asize: 1981411579221 >>> vdev_path: /dev/mfid3p4, vdev_max_asize: 1991247921152, vdev_asize: >>> 1991247921152, vdev_min_asize: 1981411579221 >>> vdev_path: /dev/mfid2p4, vdev_max_asize: 1991246348288, vdev_asize: >>> 1991246348288, vdev_min_asize: 1981411579221 >>> vdev_path: /dev/mfid5p4, vdev_max_asize: 1991246348288, vdev_asize: >>> 1991246348288, vdev_min_asize: 1981411579221 >>> ^C >>> >>> borg-new /home/ler $ >>> >>> >>> borg-new /home/ler $ sudo zpool list -v >>> Password: >>> NAME SIZE ALLOC FREE EXPANDSZ FRAG CAP DEDUP HEALTH ALTROOT >>> zroot 10.8T 94.3G 10.7T 16.0E 0% 0% 1.00x ONLINE - >>> raidz1 10.8T 94.3G 10.7T 16.0E 0% 0% >>> mfid4p4 - - - - - - >>> mfid0p4 - - - - - - >>> mfid1p4 - - - - - - >>> mfid3p4 - - - - - - >>> mfid2p4 - - - - - - >>> mfid5p4 - - - - - - >>> borg-new /home/ler $ >>> >>> >>> On 01/31/2017 2:37 pm, Steven Hartland wrote: >>> >>> In that case based on your zpool history I suspect that the >>> original mfid4p4 was the same size as mfid0p4 (1991246348288) >>> but its been replaced with a drive which is (1991245299712), >>> slightly smaller. >>> >>> This smaller size results in a max_asize of 1991245299712 * 6 >>> instead of original 1991246348288* 6. >>> >>> Now given the way min_asize (the value used to check if the >>> device size is acceptable) is rounded to the the nearest >>> metaslab I believe that replace would be allowed. >>> https://github.com/freebsd/freebsd/blob/master/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c#L4947 >>> >>> Now the problem is that on open the calculated asize is only >>> updated if its expanding: >>> https://github.com/freebsd/freebsd/blob/master/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c#L1424 >>> >>> The updated dtrace file outputs vdev_min_asize which should >>> confirm my suspicion about why the replace was allowed. >>> >>> On 31/01/2017 19:05, Larry Rosenman wrote: >>> >>> I've replaced some disks due to failure, and some of the >>> pariition sizes are different. >>> >>> >>> autoexpand is off: >>> >>> borg-new /home/ler $ zpool get all zroot >>> NAME PROPERTY VALUE SOURCE >>> zroot size 10.8T - >>> zroot capacity 0% - >>> zroot altroot - default >>> zroot health ONLINE - >>> zroot guid 11945658884309024932 default >>> zroot version - default >>> zroot bootfs zroot/ROOT/default local >>> zroot delegation on default >>> zroot autoreplace off default >>> zroot cachefile - default >>> zroot failmode wait default >>> zroot listsnapshots off default >>> zroot autoexpand off default >>> zroot dedupditto 0 default >>> zroot dedupratio 1.00x - >>> zroot free 10.7T - >>> zroot allocated 94.3G - >>> zroot readonly off - >>> zroot comment - default >>> zroot expandsize 16.0E - >>> zroot freeing 0 default >>> zroot fragmentation 0% - >>> zroot leaked 0 default >>> zroot feature@async_destroy enabled local >>> zroot feature@empty_bpobj active local >>> zroot feature@lz4_compress active local >>> zroot feature@multi_vdev_crash_dump enabled local >>> zroot feature@spacemap_histogram active local >>> zroot feature@enabled_txg active local >>> zroot feature@hole_birth active local >>> zroot feature@extensible_dataset enabled local >>> zroot feature@embedded_data active local >>> zroot feature@bookmarks enabled local >>> zroot feature@filesystem_limits enabled local >>> zroot feature@large_blocks enabled local >>> zroot feature@sha512 enabled local >>> zroot feature@skein enabled local >>> borg-new /home/ler $ >>> >>> >>> borg-new /home/ler $ gpart show >>> => 40 3905945520 mfid0 GPT (1.8T) >>> 40 1600 1 efi (800K) >>> 1640 1024 2 freebsd-boot (512K) >>> 2664 1432 - free - (716K) >>> 4096 16777216 3 freebsd-swap (8.0G) >>> 16781312 3889162240 4 freebsd-zfs (1.8T) >>> 3905943552 2008 - free - (1.0M) >>> >>> => 40 3905945520 mfid1 GPT (1.8T) >>> 40 1600 1 efi (800K) >>> 1640 1024 2 freebsd-boot (512K) >>> 2664 1432 - free - (716K) >>> 4096 16777216 3 freebsd-swap (8.0G) >>> 16781312 3889162240 4 freebsd-zfs (1.8T) >>> 3905943552 2008 - free - (1.0M) >>> >>> => 40 3905945520 mfid2 GPT (1.8T) >>> 40 1600 1 efi (800K) >>> 1640 1024 2 freebsd-boot (512K) >>> 2664 1432 - free - (716K) >>> 4096 16777216 3 freebsd-swap (8.0G) >>> 16781312 3889162240 4 freebsd-zfs (1.8T) >>> 3905943552 2008 - free - (1.0M) >>> >>> => 40 3905945520 mfid3 GPT (1.8T) >>> 40 1600 1 efi (800K) >>> 1640 1024 2 freebsd-boot (512K) >>> 2664 16777216 3 freebsd-swap (8.0G) >>> 16779880 3889165680 4 freebsd-zfs (1.8T) >>> >>> => 40 3905945520 mfid5 GPT (1.8T) >>> 40 1600 1 efi (800K) >>> 1640 1024 2 freebsd-boot (512K) >>> 2664 1432 - free - (716K) >>> 4096 16777216 3 freebsd-swap (8.0G) >>> 16781312 3889162240 4 freebsd-zfs (1.8T) >>> 3905943552 2008 - free - (1.0M) >>> >>> => 40 3905945520 mfid4 GPT (1.8T) >>> 40 1600 1 efi (800K) >>> 1640 1024 2 freebsd-boot (512K) >>> 2664 1432 - free - (716K) >>> 4096 16777216 3 freebsd-swap (8.0G) >>> 16781312 3889160192 4 freebsd-zfs (1.8T) >>> 3905941504 4056 - free - (2.0M) >>> >>> borg-new /home/ler $ >>> >>> >>> this system was built last week, and I **CAN** rebuild it if >>> necessary, but I didn't do anything strange (so I thought :) ) >>> >>> >>> >>> >>> On 01/31/2017 12:30 pm, Steven Hartland wrote: >>> >>> Your issue is the reported vdev_max_asize > vdev_asize: >>> vdev_max_asize: 11947471798272 >>> vdev_asize: 11947478089728 >>> >>> max asize is smaller than asize by 6291456 >>> >>> For raidz1 Xsize should be the smallest disk Xsize * >>> disks so: >>> 1991245299712 * 6 = 11947471798272 >>> >>> So your max asize looks right but asize looks too big >>> >>> Expand Size is calculated by: >>> if (vd->vdev_aux == NULL && tvd != NULL && >>> vd->vdev_max_asize != 0) { >>> vs->vs_esize = P2ALIGN(vd->vdev_max_asize - >>> vd->vdev_asize, >>> 1ULL << tvd->vdev_ms_shift); >>> } >>> >>> So the question is why is asize too big? >>> >>> Given you seem to have some random disk sizes do you >>> have auto expand turned on? >>> >>> On 31/01/2017 17:39, Larry Rosenman wrote: >>> >>> vdev_path: n/a, vdev_max_asize: 11947471798272, >>> vdev_asize: 11947478089728 >>> >>> >>> -- >>> Larry Rosenman http://people.freebsd.org/~ler >>> <http://people.freebsd.org/%7Eler>; >>> Phone: +1 214-642-9640 E-Mail: >>> ler@FreeBSD.org <mailto:ler@FreeBSD.org> >>> US Mail: 17716 Limpia Crk, Round Rock, TX 78664-7281 >>> >>> >>> -- >>> Larry Rosenman http://people.freebsd.org/~ler >>> <http://people.freebsd.org/%7Eler>; >>> Phone: +1 214-642-9640 E-Mail: ler@FreeBSD.org >>> <mailto:ler@FreeBSD.org> >>> US Mail: 17716 Limpia Crk, Round Rock, TX 78664-7281 > > > -- > Larry Rosenman http://people.freebsd.org/~ler > <http://people.freebsd.org/%7Eler>; > Phone: +1 214-642-9640 E-Mail: ler@FreeBSD.org > <mailto:ler@FreeBSD.org> > US Mail: 17716 Limpia Crk, Round Rock, TX 78664-7281
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?96534515-4fcb-774e-a599-8d48aec930cd>