2017-12-26
I continue to have strong confidence in ZFS On Linux
Three years ago I wrote about my experiences with ZFS on Linux up to that point, which involved a few months of running it on my office workstation. Given that we're probably going to use it for our new fileservers, I feel it's a good time to give an update with my experiences and views. The short version is that I've had a pretty great time with ZFS On Linux, without any particularly significant problems, and my experiences with it have given me a pretty strong confidence in it. There are two sides to this.
The first side is what I just wrote, with a bit of elaboration. On the one hand, ZFS on Linux has not perfect over the time I've run it. Every so often there have been glitches with new versions (most recently when a new important ZFS systemd unit was accidentally not enabled) and with features (eg this NFS permissions issue). On the other hand, I use the latest Git tip from the master repository, so I'm living a more dangerous life than most people (eg), and I've almost never had any problems. The code that lands in the tip has been remarkably stable and free of issues. In addition I've repeatedly upgraded Fedora versions over the past several years (as well as upgraded kernels plenty of times within Fedora versions) and every time, things have gone smoothly. At this point I've basically stopped worrying that something is going to break in my ZFS On Linux setup.
The other side of this is that I've now had several years of watching ZFS on Linux evolve, since I closely track the state of the master repository (I read the commit message of every commit that gets made to it, since I'm going to be running that change soon and I want to know what I'm getting myself into). Over this time ZFS on Linux has made significant changes and improvements. It has fixed irritations, improved memory handling several times (including a major recent reform), and implemented significant new features such as encryption and sequential scrubs, as well as routinely and frequently incorporating fixes and changes from upstream. In short, ZFS on Linux development is quite active and healthy. ZFS on Linux is not a peripheral platform for ZFS, it's a mainstream, well supported one that is active in overall ZFS development and contributing important changes.
Stability and evolution together are good signs. They mean that you have a project that is not stagnating and that's well run, and further that people care enough about ZFS on Linux (and use it enough) to bring up important features on it first (in the broad open source world). All of this gives me confidence both in running ZFS on Linux today and relying on it in the future. Today there probably won't be problems, and if there are any it's very likely we can get them fixed easily; in the future there will be new fixes and improvements, and we won't be stranded in a backwater watching developments happen elsewhere.
(As far as ZFS versus Btrfs goes, I haven't seen any reason to change my views from early 2015. The Btrfs situation certainly doesn't seem to have gotten better.)