opendevreview | Takashi Kajinami proposed openstack/swift master: Update SAIO document https://review.opendev.org/c/openstack/swift/+/950951 | 07:35 |
---|---|---|
opendevreview | Takashi Kajinami proposed openstack/swift master: Update SAIO document https://review.opendev.org/c/openstack/swift/+/950951 | 07:49 |
timburke | DHE, let us know how it goes! i've often wondered about whether xfs is still the right move -- i've played with btrfs a little, but hadn't gotten to trying zfs. i hadn't even gotten around to trying to measure performance differences | 15:52 |
DHE | I'm mostly trying to take advantage of ZFS' metadata disk. but modifying the auditor to skip an MD5 check would also be a nice touch since ZFS does self-verification of checksums... | 16:16 |
DHE | the other thing... ZFS xattr support, as I understand it, will embed the value inside the inode record if it can (size permitting), otherwise it assigns a new inode to carry the payload... so it should support large xattrs... and those SHOULD go onto the SSD? I think it will... so HEAD requests and rsyncs should be real fast | 17:41 |
timburke | DHE, yeah, that'd be my expectation, too. fwiw, xfs should support a similar data/metadata disk split -- have a look at the logdev option: https://wiki.archlinux.org/title/XFS#External_XFS_Journal | 18:32 |
DHE | that's just the journal. this is the entire metadata tree. `find [...]` should be able to sweep the disk hitting only the SSD, barring the SSD being full and spilling onto the main spinning disk... | 20:24 |
DHE | I did see XFS was also working on support for zoned devices (aka Shingled disks where the host manages them) and a metadata disk was in the works hijacking a feature called the "realtime" section, but it seems to be backwards where you put the spinning disk as realtime data, the SSD as the main disk, and mark all data as realtime... | 20:27 |
opendevreview | Tim Burke proposed openstack/swift master: ring: Introduce a v2 ring format https://review.opendev.org/c/openstack/swift/+/834261 | 20:46 |
timburke_ | DHE, eek, yeah, that doesn't sound near as good... | 20:51 |
DHE | yeah... I love ZFS in general, but non-redundant storage is not its strong point. so this might be interesting for the wrong reasons... | 20:55 |
opendevreview | Yan Xiao proposed openstack/swift master: Object-server: change labeled timing metrics sample rate for debugging requests https://review.opendev.org/c/openstack/swift/+/951042 | 21:18 |
opendevreview | Merged openstack/swift feature/mpu: mpu: append link to dependant mpu resource to override etag https://review.opendev.org/c/openstack/swift/+/938189 | 21:31 |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!