chaostheo

If you are running a multi device real world file server, the benchmark result of gemetric mean are still too good to zfs.

Still ZFS for stability ?? Pull the power plug while writing and you know what your stability is worth in reality. I would say say ZFS is for verify as long as you have a usv in front of your server otherwise with a poweroutage you should have a good backup and be able to do a full restore.

Didn't you still think that zfs is not a bit slow for nfs file service (real clients or for any vm/container) ? 2x Srv, Xeon 62xy, 192G, 13x18T r6 xfs, 100Gb connectx6, ipoib, nfs4: ./elbencho -r -b 1M -t 8 -s 10g --direct F{1..8} OPERATION RESULT TYPE FIRST DONE LAST DONE ========= ================ ========== ========= READ Elapsed time : 7.151s 7.786s IOPS : 10789 10521 Throughput MiB/s : 10789 10521 Total MiB : 77157 81920 --- 09:01:35 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:36 PM ibp216s0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 09:01:36 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:37 PM ibp216s0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 09:01:37 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:38 PM ibp216s0 1500050.00 53390.00 3055651.88 3698.86 0.00 0.00 0.00 25.03 09:01:38 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:39 PM ibp216s0 5653739.00 200654.00 11516727.26 13894.61 0.00 0.00 0.00 94.35 09:01:39 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:40 PM ibp216s0 5705062.00 203018.00 11621316.50 14052.73 0.00 0.00 0.00 95.20 09:01:40 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:41 PM ibp216s0 5620242.00 196400.00 11448537.04 13633.62 0.00 0.00 0.00 93.79 09:01:41 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:42 PM ibp216s0 5699689.00 187514.00 11610387.99 13142.38 0.00 0.00 0.00 95.11 09:01:42 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:43 PM ibp216s0 5713953.00 186006.00 11639418.27 13060.07 0.00 0.00 0.00 95.35 09:01:43 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:44 PM ibp216s0 5725531.00 193176.00 11663012.59 13484.04 0.00 0.00 0.00 95.54 09:01:44 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:45 PM ibp216s0 5610783.00 187174.00 11429270.82 13089.09 0.00 0.00 0.00 93.63 09:01:45 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:46 PM ibp216s0 1942932.00 62188.00 3957775.14 4379.84 0.00 0.00 0.00 32.42 09:01:46 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:47 PM ibp216s0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 09:01:47 PM IFACE rxpck/s txpck/s rxkB/s txkB/s rxcmp/s txcmp/s rxmcst/s %ifutil 09:01:48 PM ibp216s0 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

And still not tried creating snapshots as reflink copies or even tried duperemove ? Then dream on with zfs.

That's true, only mkfs.xfs & mount is required, any tuning options are optional. Write to filesystem while cut host power, last written data is incomplete but filesystem is coming up. With zfs you will get good chance for loosing import pool and all created datasets and even all snapshots at once in that case, so send/receive server should have other power, usv is good for clean shutdown and even backup always good to know :-)

Anytime the day will come where zpool options clear, -f, readonly, -F, -X, -T in any combination wouldn't help importing a pool. The best is to check that out by yourself by a power cut while writing to a testpool dataset and just be prepared for such a case. Theory and reality are two storys and not equal be design.

Yes, without the auto-survive of a power outage zfs will never be accepted by redhat whatever the zfs license is look like in the future, still this simplest thing can bring lots of headache and pain even zfs designed zil and slog but what it's good for when you google cannot import pool ... ?!

These features keeps me solidly without ZFS: 0. Powercut zfs host : cannot import pool without destroy old and create new. 1. For case "0" need minimum 2nd zfs host while backup restore even slower. 2. Extrem slow meta-/data reads with aging, eg try tar any dir to /dev/null. 3. So stay working with other alternatives and have fun.