On Wed, Apr 15, 2020 at 2:31 PM james <[email protected]> wrote: > > On 4/15/20 11:40 AM, Rich Freeman wrote: > > > I personally use the latest longterm, but not until it has been out > > for a few months. Mainly this is because I use zfs and don't want to > > deal with what versions of the one are compatible with what versions > > of the other. > > Yep, for the main system, but using btrfs with redundant drives. I'd > like zfs, but not certain about it's future being open, open-source, > etc. btrfs has bee great, for what I have done recently. >
So, a few comments here: First, I used to use btrfs and I'd say it is just as important to stick with a longterm using btrfs because that project has a terrible history of introducing regressions in new kernel releases. If I were using btrfs I'm not sure I'd even go with 5.4 over 4.19 as it has only been around a few months and I'd be concerned they haven't worked out all the btrfs bugs yet. Now, I haven't used btrfs recently, so maybe things have gotten much better, but I'm skeptical on that front. I've had to do complete btrfs reinstalls more than once from backups, and this was on btrfs raid1 only. I REALLY like the feature set and design/etc of btrfs and think it definitely could be the future of linux mainstream storage, but for whatever reason QA has been a big problem and it has taken way longer than I expected to mature. It was btrfs QA problems that drove me to pay such close attention to what kernel series I was running. That is why I've mainly moved to zfs as my main general-purpose filesystem on hosts where restoring from backup isn't about popping an SDcard out of a Pi and flashing a couple GB backup image onto it. I'm not entirely happy with some of the limitations of zfs and of course it not being in-mainline is a huge hassle. There really is no risk of it not being FOSS - it is FOSS and of course it always will be as is the case with anything FOSS. Whether anybody is contributing to it in 10 years is another matter, but it isn't like the license has an expiry date on it. The #3 openzfs contributor is a Gentoo dev. I suspect the main risk to zfs is that btrfs finally gets its quality level up sufficiently that people switch over, which would be great. Either that or zfs gets sloppy with QA and people abandon it, which would be terrible, but probably unlikely at this point. For larger-scale storage I'm using Lizardfs and greatly admire Ceph as well in this space. MooseFS is another option (which Lizardfs is a fork of). These distributed filesystems are generally more flexible than zfs and give you redundancy above the host level. Right now the bulk of my storage is on lizardfs with the lizardfs chunkservers being implemented on top of zfs. That gives me the data security benefits of zfs but without the inflexibility, since I don't pool drives so I'm not limited by the ability to add/remove/etc drives from zfs pools. That said, vdev removal has become a thing in v0.8 and perhaps we'll see increased flexibility in the future. Overall zfs and btrfs are actually converging somewhat, just from different target audiences. IMO with the growing importance of distributed filesystems I think that the main niche for zfs and btrfs will be as a general-purpose filesystem similar to ext4 but with additional flexibility (volume management) and robustness (raid/checksums/etc). Once you get bigger than a few drives Ceph will become the gold standard for storage, or at least that is the leading technology right now. Lizardfs is more of a ghetto Ceph that doesn't require dozens of GB of RAM per server. If you haven't been upgrading kernels you may have just missed all the fun of btrfs regressions over recent years. :) In any case unless things have changed a lot I'd seriously consider longterms and then carefully checking for regressions before doing upgrades between major versions. -- Rich
