zfs-macOS-2.4.1rc2

[lundman]

I have just posted rc1 for zfs-2.4.1. It has a lot of commits and new features. I have done a zfs-test run on Intel/Catalina (last place we can run zfs-test), and tried it on macmini tahoe.

Let's test it and report back issues. Might be worth checking in on some of the open issues as well

I did notice Blake3 is still commented out, should check if we can get that to run again.

[a7z]

I don't know if it is too early to ask, but I might test it on 10.11 and 10.14 if you are so kind and make a pkg for those.

[lundman]

oh yeah, its a good time to make sure it compiles on old systems still

[lundman]

done

[a7z]

Thanks for compiling.

Installation from 2.4.1rc1 was much faster than with 2.3.1 on 10.11.
Installation on 10.14 took a very long time.

rsync, cp, rm, Finder, all work at normal speed for zfs.

import/export work as expected.

Snapshotting, diff, rollback and destroy work.

FreeFileSync is just as fast.

scrub performance is within the normal range.
scrub -S -E is scrubbing the whole pool ("when the date is prior to the date where this feature is enabled").
But where do I enable it?

As "cp" from the OS (10.11 and 10.14) does not have --reflink I installed GNU coreutils long time ago.
But gcp (GNU coreutils 9.5) with --reflink only returns "... could not be cloned ... Operation not supported".
It has never worked even with older versions of zfs.
Although the feature is enabled.

Uninstalling 2.4.1rc1 led to strange errors (zfs-uninstall-cli.txt) although the uninstall seems to work (zfsuninstaller.LVktlD.txt).
I made a diff between the uninstall scripts 1.9.4 vs 2.4.1rc1. You have changed some lines.
Maybe these changes are only for 10.15 or newer.

zfs-uninstall-cli.txt
zfsuninstaller.LVktlD.txt

Disk Speed Test (Blackmagicdesign) shows fluctuating performance from 30mb/s - 250mb/s (250mb/s would be normal) while AJA System Test Lite shows constant performance but only 150-170mb/s write and cause of arc a higher read.

Beside this there were no crashes or panic on both OSs.

I tested only the normal daily using parts I already use or wanted to use (like block cloning).
For DirectIO and the other more fancy parts I have no use case for or are not possible for me to test cause of not having a NVMe SSD.

My System is a MacMini 2009, Samsung SSD and 8gb.

[lundman]

snort - I see I compiled for 14 (Sonoma). It had a 14 in it!

[lundman]

Uninstalling 2.4.1rc1 led to strange errors

Ah yep, the newer macOS does need us to run that rebuild command, I thought I wrapped it in a test for the command, but probably messed it up.

Cloning could possibly be made to work, but it is one of those questions if it is worth it for old OS.(I assume it still works for newer macOS).

[lundman]

OK there is a panic problem with "new" arm64 macOS. Something goes wrong in vnop_pageout. (Any macOS when sptm enabled).

[lundman]

77a544b ed88da2

Will be in rc2 build