So far we have upgraded more tahn 70 instances using our repo and also upgraded all previously 2.3.0 to 2.3.1. Also we have made it compatible with PVE 8.4 and Kernel 6.14. About 20 instances already running PVE 8.4 + Kernel 6.14.
right, we are all impatiently waiting. i keep checking the updates daily, sometimes multiple times, would be nice to see in the test branch even if it isn't certain for production yet. i am still very tempted to install the ZFS from this repo. haha, if it takes much longer i will probably do it.
hopefully, haha. we totally need those features, it will solve a lot of problems and probably have a lot fewer people in the forum complaining about zfs issues with it being slow, i know ive contributed to that trying to fix it... hahaI could bet on it zfs 2.3.2 would be released this week ... maybe that's any next time the first 2.3 edition for pve ...
Haha, yeah, we will see ... interesting about any perf boosts to "normal" zfs usage yet (hopefully some will report improvements) as I don't think there are many guys and girls who ie use deduped zvols today and switch to fast-dedupe ... And it brings new problems too like zraid expansion which takes very long which ie was missing problem/feature until now ...
true many avoid dedup because its so painfully slow and not worth it usually, but more probably will try fast dedup from what ive seen on benchmarks the performance loss is very minimal compared tot he old method which almost completely destroys all performance. but i think direct I/O will also really help for many, especially for nvme users, (i am hoping it helps with the buffer issues and stall outs i get when writing a large amount of files directly to the drive), for me at least i think if dedup didnt have such an impact and stall outs stopped zfs would be far more usable. i could definitely see zraid expansion being potentially problematic and also take forever to test properly, seems like a really great/needed feature though, even if it comes with some bugs at first.Haha, yeah, we will see ... interesting about any perf boosts to "normal" zfs usage yet (hopefully some will report improvements) as I don't think there are many guys and girls who ie use deduped zvols today and switch to fast-dedupe ... And it brings new problems too like zraid expansion which takes very long which ie was missing problem/feature until now ...
definitely just a joke, we're i am sure, all quite happy with and thankful for, the wonderful work the proxmox devs do for us all, we are just excited for the new shiny, so to speak, haha.
oh yeah i wouldnt use it on consumer NVMEs for sure, its apart of why i still am not using it as a boot disk, mine would be toast, but its holding up well as an LVM-Thin volume, only at 8%, but i have removed all swap files from it and tried to minimize logging to it, and it was at 6% i believe when i installed proxmox and ive used a number of vms on it for well over a year now. so pretty good at least.Maybe ... the new faster direct I/O will going to make cheap consumer ssd's/nvme's nearly complete unuseably while toasting these instantly then ... ??
For fast-dedupe you need more mem+cpu resources which are hold gladly small on pve's as wanted for the vm's which than accidently could go into opposite than expected if not provided ... try and test extensively
Hi!
I upgraded for one reason: I ran out of space and wanted to expand the raidz vdev.
Thank you for making this available to everyone. Mostly my concerns areZFS 2.3.1 has been officially released – by the OpenZFS team! – though, is is simply not yet included in PVE delivery.
For me personally, there's no reason not to use it. My team and I have spent about 7 man-weeks testing, migrating, and preparing repositories. While anyone could technically download the official release and compile it, we had to do this on multiple systems for our own infrastructure and for customers. That’s why we decided to invest some extra time (and money) to make prebuilt packages available for everyone – simply to make it easier to adopt.
I don’t believe stability is the issue here – support might be. That said, getting everything working properly was a lot more effort than we expected. Someone commented that “it can’t be that much work” – well, it is, if you want to do it properly ;-)
I’m not trying to push anyone to use it. All I’m saying is: it’s released – and not only includes the improvements that have been discussed, but also a lot of bug fixes, making ZFS more stable and reliable overall.
As for performance: hard to generalize. On Gen5 NVMe and under demanding workloads, we’re seeing speedups of around 1.5x to 2x. On older hardware? Likely less – we haven’t tested that in detail.
Happy to discuss more if anyone has specific questions
– M.
i am definitely leaning towards waiting for the official one even though this is really temping, only because i cannot afford to risk any kind of future instability or data loss, or i would absolutely try it.
