Post AQUiGagBBlgZlXfq2C by nestor@mastorol.es
(DIR) More posts by nestor@mastorol.es
(DIR) Post #AQOgGbRmy5jPI53rYu by duponin@udongein.xyz
2022-12-08T12:43:47.377140Z
0 likes, 0 repeats
I dunno if I should go on 4x 4TB (RAIDZ1) or 2x 8TB (mirror)yeah mirror is faster to rebuild but more expensive in the endand that's not my single zpool, I would already have redundancy (zfs send/receive)
(DIR) Post #AQOgKT5lySiyXIbfZA by kaia@brotka.st
2022-12-08T12:44:46.294705Z
0 likes, 0 repeats
@duponin what do you do with that storage?
(DIR) Post #AQOgVfzmL9t1jy9CDI by duponin@udongein.xyz
2022-12-08T12:46:30.769030Z
0 likes, 0 repeats
@kaia forgot writting itDB and S3maybe a few other things such as iSCSI portal
(DIR) Post #AQOgcq7F6xfr9h3QUS by roobre@owo.cafe
2022-12-08T12:46:06Z
0 likes, 0 repeats
@duponin I went for stripe of mirrors a couple years ago when raidz expansion was not a thing. It works nice but it feels very wasteful with weaker guarantees than an expandable raidz-2.
(DIR) Post #AQOgcqiorFjR2EdR3o by duponin@udongein.xyz
2022-12-08T12:47:51.816774Z
0 likes, 0 repeats
@roobre RAIDZ is still not expandable, a ZPOOL can't be changed on the flymirror really sounds waste to me :blobcatpensive:
(DIR) Post #AQOjeezPkVfR5bx6WG by roobre@owo.cafe
2022-12-08T12:58:28Z
1 likes, 0 repeats
@duponin Wait, really? I thought they merged it already :blobCatFacepalm: Mirror is very wasteful especially for low failure rates at non-production workloads. But it scales better IMO, as long as you have enough bays.
(DIR) Post #AQOjhlWLvG68DMq7W4 by duponin@udongein.xyz
2022-12-08T13:22:20.458317Z
0 likes, 0 repeats
@roobre well, you can add vdev to your pool but it won't switch it to RAIDZ2 or anything AFAIK
(DIR) Post #AQP1c3iDAnv24Tob6u by roobre@owo.cafe
2022-12-08T16:42:44Z
1 likes, 0 repeats
@duponin Nope. I am hoping for raidz expansion so I can create a raidz-2 and just keep adding disks without changing the redundancy level.
(DIR) Post #AQP1ehtuiuj3IGYG7U by duponin@udongein.xyz
2022-12-08T16:43:30.186390Z
0 likes, 0 repeats
@roobre some researches should be done, I genuinely don't remember if that can be done
(DIR) Post #AQP1owUhJwrRRwdaTY by roobre@owo.cafe
2022-12-08T16:44:39Z
1 likes, 0 repeats
@duponin I thought it was possible now, but it seems it is not. PR is still unmerged https://github.com/openzfs/zfs/pull/12225
(DIR) Post #AQP2ENXjtFJsa3Bmeu by nestor@mastorol.es
2022-12-08T13:49:31Z
0 likes, 0 repeats
@duponin What are your requirements?If you can afford It, a zraid2 is a good balance between storage, price and not having a bad time you you have to replace a faulty disk.If not, 2x8 mirror is a bit safer just because the more disks in the pool, the more chances of get a faulty one.You mentioned that you have periodic copies on other pool, right? Then It might worth to have more storage capacity, all depends on my initial question. What are your requirements? 😊
(DIR) Post #AQP2EP08T9d16PTyYi by duponin@udongein.xyz
2022-12-08T16:49:51.233612Z
0 likes, 0 repeats
@nestor my requirements are basically having decent IO and not paying thousandsI would do periodic backups to another pool (it's a little mess now)I think this pool would do DB and S3yeah I know it's pretty vague :akko_weary:
(DIR) Post #AQR4JxYU0wENwTFIem by nestor@mastorol.es
2022-12-09T15:28:41Z
0 likes, 0 repeats
@duponin For db, zfs has specific requirements and setups, you should check them before.Between mirror or raidz1 I could recomend the cheapest price/gigabit , probably 4x4 zraid1 for 12 raw TB? + zstd compresión and iI would add fast SSD arc cache. It doesn't need to be specially Big or expensive to improve performance.Of course, as we sapaniards usually say, the devil is in the details. 😁
(DIR) Post #AQR4JyCBdJzRvbp0Xg by nestor@mastorol.es
2022-12-09T15:32:02Z
0 likes, 0 repeats
@duponin For the periodic zfs send/recv there are existent tools, i don't remember them now, ping me if you want the names and i could search them once i reach a computer.
(DIR) Post #AQR4JygJpI5pR3v4Vc by duponin@udongein.xyz
2022-12-09T16:22:50.467369Z
0 likes, 0 repeats
@nestor after small benchmarks I won't use ZFS for DB, performances are terrible (around 80 TPS)I followed this guide on a few parts but having too much will to do everything and diving in performances tuninghttps://vadosware.io/post/everything-ive-seen-on-optimizing-postgres-on-zfs-on-linux/however, for file storage I'll just go for 4×4 raidz1, why not L2ARC laterlet's not go too much in the details now :akko_wink: for send/recv, there are two options available with NixOS, syncoid and zfs-replicateI haven't made my choice yet
(DIR) Post #AQUeCKdwT8sj3WLDAe by duponin@udongein.xyz
2022-12-11T09:48:51.426397Z
0 likes, 0 repeats
@nestor yup, :nixowos: FTWyou have to take time to start using it, but once you are started, you build a nice momentumfor DB, I use PostgreSQLI think I won't use ZFS, too much workone of my base need 40minutes to be dumpedthat's a bit long but nothing unmanageable
(DIR) Post #AQUeFf1p8MO19F6Bzk by nestor@mastorol.es
2022-12-11T09:39:32Z
1 likes, 0 repeats
@duponin Regaring syncoid, I believe that I tested it and it was fine but it was long time ago.Personally I don't run any send/recv in my domestic systems, mostly because I haven´t the hardware. At work I use my own scripts for rsync and zfs snapshot , send/recv using rundeck, and several instances of truenas.
(DIR) Post #AQUeP8Bi5a50BtEsOu by duponin@udongein.xyz
2022-12-11T09:51:18.461401Z
0 likes, 0 repeats
@nestor the send/recv would mostly be for backup purpose, from my desktop to the storage serverand maybe for externalising my backups on another server too (unless I use borgbackup to not put all eggs in the same basket)
(DIR) Post #AQUgz2Rn6rfUih0hsG by nestor@mastorol.es
2022-12-11T10:11:38Z
1 likes, 0 repeats
@duponin Yep, I was talking about really huge dbs, when the backups might take to much time to be practical, even in a master/slave environment.Nixos is, definitively in my new year queue of nice to do in 2023 :D
(DIR) Post #AQUh57CEMkEDBbd6Se by duponin@udongein.xyz
2022-12-11T10:21:12.648242Z
0 likes, 0 repeats
@nestor I'm considering doing a DB replica (a cluster basically) to have a sort of backup and dumping data without doing too much I/O on production DBhehe, have fun!
(DIR) Post #AQUiGagBBlgZlXfq2C by nestor@mastorol.es
2022-12-11T10:25:33Z
0 likes, 0 repeats
@duponin I'll do.If you end going with the replica, you should keep in mind that replica aren't really backups.It will save you if the master server dies, but it will not protect you from user errors like deleting or updating the wrong data. Also recovery point in time will need extra work.Have a great day!
(DIR) Post #AQUiGbJso9RdkgFXv6 by duponin@udongein.xyz
2022-12-11T10:34:32.833807Z
0 likes, 0 repeats
@nestor sure it's not backup, just a replicaI would run backups from the replica