OpenZFS 2.3-rc2 Brings CPU Pinning & Optimized Kernel Same-Page Merging

[unwind-protect]

I believe the CPU pinning and KSM thing is only related to the qemu test environment but not to ZFS itself. So, there's nothing exciting about this merge.

Yeah. And it would be stunning if those things were introduced between two release candidates.

[mbod]

So core existential question... is OpenZFS worth the risk of running in enterprise production given its out of tree and occasionally breaking kernel update issues?

Absolutely. Out of tree modules exist since Linux exists. This is the reason why DKMS was invented. And it is good. With regard to stability DKMS modules have an advantage over in-tree modules: With DKMS you can stick to a specific ZFS version when you update the kernel. We all know that btrfs and even ext4 had regressions in the past. This can always happen when a new kernel brings a filesystem update as well. With ZFS you can keep the ZFS version flat.

With regard to kernel version compatibility I must say that this is not an issue at all. True, the ZFS developers need time to update ZFS for changes with new kernels. Linux changes the API from patchlevel to patchlevel, from 6.10 to 6.11 for example. This ZFS update can take several weeks/month. BUt ZFS is not made for bleeding edge kernels anyways. That contradicts the goal of stability. ZFS is more for LTS kernels. And with LTS kernels you do not have any update issue for years resp. never.

[Developer12]

So core existential question... is OpenZFS worth the risk of running in enterprise production given its out of tree and occasionally breaking kernel update issues?

Ask all the millions of people doing it. Who did you think was funding OpenZFS development all this time? A portion of that funding goes into the adaptation that OpenZFS does to run on top of the linux kernel (and the windows kernel, and the macOS kernel, and the BSD kernels, and the illumos kernels). Much of the rest goes into new features like those in the articles.

Not that it really matters for enterprise. They don't run bleeding-edge kernels. OpenZFS support for LTS kernels often arrives before the LTS is cut. If you're extra worried, run Ubuntu Server or another distro where the distro guarentees that releases and updates will be kept in step. It's a basic part of what distros do.

[zexelon]

So core existential question... is OpenZFS worth the risk of running in enterprise production given its out of tree and occasionally breaking kernel update issues?

[pWe00Iri3e7Z9lHOX2Qx]

Looks like the mentioned performance / CPU pinning stuff is for the test suite runs in VMs, not for the filesystem.

[Polaris]

I believe the CPU pinning and KSM thing is only related to the qemu test environment but not to ZFS itself. So, there's nothing exciting about this merge.
Also, one feature that has been overlooked is the json output support for all commands, which is massive for anyone doing scripting.

[phoronix]

Phoronix: OpenZFS 2.3-rc2 Brings CPU Pinning & Optimized Kernel Same-Page Merging

OpenZFS 2.3-rc1 released last week with RAIDZ expansion, fast deduplication, and direct IO support among other changes for this ZFS file-system implementation for use on Linux and FreeBSD systems. OpenZFS 2.3-rc2 is out today with a few more interesting changes...

https://www.phoronix.com/news/OpenZFS-2.3-rc2-Released