2.2.0 report on 10.14, 10.11 & 10.9

[nodarkthings]

... so I've put my external drive internal and no more KP with my slice zpool.
It's clearly an issue with USB that only happens in 10.9.

[RJVB]

Instead, folder icons won't show

What folder icons? You mean when you set a custom icon via copy/paste in the Finder's GetInfo dialogs or even when copying the source folder's Icon'$'\r' file (using `ditto --rsrc` to be certain)? In the former case the resulting 'Icon'$'\r'' file will not get the resource fork in my quick testing but it does in the latter case:

Code: Select all

> xattr -l /Volumes/tank/testfolder/'Icon^M' | grep ':$'
com.apple.FinderInfo:
com.apple.ResourceFork:


However, I don't think the resource fork is required; I can use both the above methods to set a folder icon on a folder on an NTFS volume mounted with Paragon's NTFS-for-Mac driver. In fact, the resource fork isn't preserved when I use ditto - AT LEAST NOT as the usual xattr. I'd have to mount the share on Linux to see if another "trick" is used to store resource forks with that driver.

For me folder icons are really icing on the cake; USB stability is definitely not.

Do you have an "old" enclosure or external that is USB2-only and definitely doesn't support UAS(P)? I have seen hangs under Linux when I do large, high-bandwidth transfers over the USB3 port in one of my notebooks. It doesn't KP but I still need to reboot the system to unblock the situation so it boils down a similar annoyance. If transfer rates over the bus have increased significantly in ZFS 2.2x the KP could be a result of the inappropriate handling of the bus reacting in an unexpected manner?

[nodarkthings]

Instead, folder icons won't show

What folder icons? You mean when you set a custom icon via copy/paste in the Finder's GetInfo dialogs or even when copying the source folder's Icon'$'\r' file (using `ditto --rsrc` to be certain)? In the former case the resulting 'Icon'$'\r'' file will not get the resource fork in my quick testing but it does in the latter case:

Code: Select all

> xattr -l /Volumes/tank/testfolder/'Icon^M' | grep ':$'
com.apple.FinderInfo:
com.apple.ResourceFork:


However, I don't think the resource fork is required; I can use both the above methods to set a folder icon on a folder on an NTFS volume mounted with Paragon's NTFS-for-Mac driver. In fact, the resource fork isn't preserved when I use ditto - AT LEAST NOT as the usual xattr. I'd have to mount the share on Linux to see if another "trick" is used to store resource forks with that driver.

For me folder icons are really icing on the cake; USB stability is definitely not.

Do you have an "old" enclosure or external that is USB2-only and definitely doesn't support UAS(P)? I have seen hangs under Linux when I do large, high-bandwidth transfers over the USB3 port in one of my notebooks. It doesn't KP but I still need to reboot the system to unblock the situation so it boils down a similar annoyance. If transfer rates over the bus have increased significantly in ZFS 2.2x the KP could be a result of the inappropriate handling of the bus reacting in an unexpected manner?

You're hitting many (good) points here!
Yes, it's an old USB2 dock (and I don't have a clue of what UAS(P) is ) so I guess your hypothesis about 2.2.x bandwidth could matter (but why is it working in 10.11 and 10.14, then? )
About icons, I've only tried with a folder where I've pasted icons like you say, indeed!
Sure enough, custom icons are icing on the cake... All I can say is that if I put that folder on the zpool via 10.11 or 10.14, 10.9 won't see them, but they are there as I've seen them when I came back to 10.14.

So I guess that in your use case, v2.1.6 could work ok. In my case, v2.2.x could do as I was only using USB for testing, but my real ZFS backups are connected internally, usually...

[RJVB]

This isn't my first rodeo, and my Mac is in fact NOT my machine that only has USB ports for external disks (it still has FireWire 800, plus a Thunderbolt1 port - but that one is occupied). Backing up a pool of a few 100 Gb over gigabit ethernet is just not an option, so USB *has* to work for me.

Yes, it's an old USB2 dock (and I don't have a clue of what UAS(P) is ) so I guess your hypothesis about 2.2.x bandwidth could matter (but why is it working in 10.11 and 10.14, then? )

You misread me, I was thinking that USB3 might be the problem because that's my experience. UAS(P) is a newer "super protocol" that allows better throughput. It can be instable in older Linux kernel versions combined with certain chipsets in the external device but I never heard of problems with it in OS X 10.9 .
However, I would not be surprised if your use of a dock (powered hub?) are increasing your chances for the kind of problems on which the old "don't use ZFS over USB" is based.

I have a dual-device mirrored pool, two M.2 SSDs in USB3 enclosures. I connect those via a small non-powered USB3 dock to one of the ports in my Thunderbolt dock (which gives me my USB3 ports!). I have been using ZFS 2.1.0.4 (latest commit from the 2.1 branch, self-built) until now and have resilvered the mirror device in that pool multiple times now. I think there's about 20Gb of data on the pool without any issues, and resilvering is something that reaches much higher throughput rates than you get from copying a folder (unless it has huge files in it).

Evidently you can mix and match ZFS 2.1 with 2.2 on different machines. I prefer to avoid it because I don't want to see continuous messages about pools that can be upgraded, or run the risk that a pool I create on 1 system won't import on the other. But I just saw that I might get that even with 2.1.6 because someone backported 2.2 features that aren't available in ZoL 2.1.x (like Blake3 checksums).

[nodarkthings]

Evidently you can mix and match ZFS 2.1 with 2.2 on different machines. I prefer to avoid it because I don't want to see continuous messages about pools that can be upgraded, or run the risk that a pool I create on 1 system won't import on the other. But I just saw that I might get that even with 2.1.6 because someone backported 2.2 features that aren't available in ZoL 2.1.x (like Blake3 checksums).

Sorry if I mistake reading things (I'm French ) but as far as my little knowledge goes with ZFS, if you don't upgrade the pools you can safely use them, I'm sure you know it (mine are still at 1.9.x).
Here's something about hiding the message saying pools can be upgraded: https://superuser.com/questions/1768447/how-to-deal-with-the-some-supported-features-are-not-enabled-warning-in-zfs (it's in the OP, apparently the 'compatibility' property)
I guess you could keep a v2.1 somewhere and make sure you always create your pools with that, as long as you don't need more recent features...
EDIT: I haven't read it all, but I guess the gist of it is there https://openzfs.github.io/openzfs-docs/man/master/7/zpool-features.7.html#Compatibility_feature_sets

[RJVB]

Well, you how it goes with that (using an option or using a dedicated version): one day you forget, and you discover your mistake after you've spent a good long while pumping data onto the pool...