ZFS FreeBSD moving to ZFS-on-Linux

There are several linux distros such as devuan, slackware, gentoo, apline linux, etc. that don't use systemd so ZOL developers will have to keep this in mind.

Be sure that will not weigh much. Slackware is brain-dead since 2016, Devuan has too few resources to honor its promise, gentoo and Alpine are niche OS... I evaluate the open source OS landscape at least once in a year and the only decent systemd-free distro I found so far is Void Linux, but though it is gaining momentum, it still has a negligible market share compared to Red Hat and Ubuntu.

Now, systemd's developer is on Red Hat's (i.e. IBM) payroll and Debian has very recently voted to focus on systemd only (with much diplomacy and chosen words, but still), which seals the destiny of Debian-based distributions. Note that the debate at Debian was between systemd and SysV init, they have constantly refused to consider other alternatives such as OpenRC or runit, without even caring to explain why when asked. And when the only allowed alternative to systemd is SysV init, who would choose SysV init?

Nowadays, the Linux ecosystem is clearly ruled by IBM, Intel and Microsoft, just like MS-DOS and Windows in their times, because Linux has appeared as _the_ cloud OS - and the cloud is what generates big bucks today, its a serious business.

This implies that anyone joining a Linux project must agree to shut up and obey. Structuring decisions are made by serious people (i.e. IBM, Intel, Microsoft) without any possible discussion, full stop.
 
Nowadays, the Linux ecosystem is clearly ruled by IBM, Intel and Microsoft, ...
A: You are forgetting a whole lot of other companies that are funding Linux work. And I'm not sure I agree with your list; Microsoft doesn't contribute all that much, compared to the others.
B: And without those companies funding (and therefore ruling) it, there would be much less work on Linux. A serious industrial production OS is not something that can be done by a small number of unpaid volunteers.

This implies that anyone joining a Linux project must agree to shut up and obey.
That as always been the case, since the early days of Linus and Alan Cox. A very small set of people make the decisions, the others do the work. Don't like it? Go work on something else.
 
gentoo and Alpine are niche OS...

Alpine is very common in Docker land, because it keeps Docker images small. Alpine is necessary here since you are shipping a whole OS in Docker and not just an application.

Disclaimer: I dealt with Docker heavily in two courses before I graduated.
 
It's an interesting question. Can you create a zpool without new features so it is compatible between linux and bsd ?

It is possible. I was accessing a single disk zpool from both FreeBSD 12.1 and Gentoo Linux (zol's zfs-kmod 0.8.2). Switching the tank pool to the other os requires zpool export tank in one os followed by zpool import tank in the other. This is only possible by creating the pool in FreeBSD -- a pool created in Linux will have extra "features" not importable into FreeBSD.

I don't recommend trying to export/import a pool between FreeBSD and Linux. For one reason, Linux is less robust in identifying the pool members than is FreeBSD. When I added a couple of new 4 TB drives to my system in preparation for setting up a new 2-disk mirror, and exported and moved the existing single-disk to a hot-swap bay to import and access its data one last time, FreeBSD was fine with the changed port address but Linux crapped and reported that there were no datasets in the pool. This seems characteristic of Linux to me -- rapidly adding more and more features but failing to produce a robust implementation. Being able to move drives around without losing the pool is exactly the sort of feature I expect from zfs. I never could recover access to the datasets in Linux. I exported the "no datasets" pool from Linux one last time and imported it into FreeBSD which handled it just perfectly. My old zpool was safely copied onto the new mirror under FreeBSD.

I'll stick to FreeBSD-only for zfs. FreeBSD can be trusted with my data and Linux can not.
 
Such classic Linus. A lot of it is reasonable (his worries about Oracle copyright and lawsuits). The rest of it is obnoxious bullshit. In particular the statements "It was always more of a buzzword than anything else" and "The benchmarks I've seen do not make ZFS look all that great." are utter garbage. And "it has no real maintenance behind it either any more" is not only a lie, it is even an obnoxious lie.

Good old Linus, always shooting from the hip without aiming. Fortunately, the Linux ecosystem has a lot of sensible people capable of making good decisions, in spite of what the guru says. Unfortunately, there are a lot of Linus fanboys who think everything he says is gospel.
 
Good old Linus, always shooting from the hip without aiming. Fortunately, the Linux ecosystem has a lot of sensible people capable of making good decisions, in spite of what the guru says. Unfortunately, there are a lot of Linus fanboys who think everything he says is gospel.

Am I right to understand that you and the legions of "sensible people capable of making good decisions" in the Linux ecosystem (hereafter collectively referred to as "Red Hat" ) are going to fork the Linux kernel ?

It makes sense since Systemd (hereafter referred to the "1.5M line chaos") has failed big time.

I really like to see that.
 
Am I right to understand that you and the legions of "sensible people capable of making good decisions" in the Linux ecosystem (hereafter collectively referred to as "Red Hat" ) are going to fork the Linux kernel ?

Another non sequitur. We don't really care whether ZoL is merged into Linux or not. Actually, it's quite convenient for FreeBSD that it can't be merged, that way ZoL is forced to maintain a well defined boundary with the Linux kernel, which is (comparatively) good for portability.

Oh, and screw OpenBSD and their opinions on operating system design.
 
Another non sequitur.
You think the Guru has gone cuckoo so why do you still follow the Guru? I don't see any non sequitur.

We don't really care whether ZoL is merged into Linux or not.

There is not such thing as Linux. There is only a Linux kernel and there will be an integration at some point in time unless ZoL hits a roadblock.

Oh, and screw OpenBSD and their opinions on operating system design.

My opinion is that every approach to operating system design is interesting and edifying.
 
Am I right to understand that you and the legions of "sensible people capable of making good decisions" in the Linux ecosystem (hereafter collectively referred to as "Red Hat" ) are going to fork the Linux kernel ?
I am in no position to do that, as I am not a Linux kernel decision maker. But I know a little bit about storage and file systems, and if you think about it for a while, you will find that ZFS has great advantages, which Linus happens to ignore in his rant. I hope that the better mousetrap will eventually succeed in the marketplace, and ZFS will be fully supported on some Linux distribution and in the kernel. I've explained why I think ZFS is good several times on this forum; in a nutshell: built in RAID (which much shorter MTTR, which translates directly into higher durability), complete on-disk checksum coverage, support for multiple RAID levels and relatively dynamic transition between them, and a mostly workable user interfaces (sometimes a little wonky, sometimes missing a few features, but better than everything else that's free).

It makes sense since Systemd (hereafter referred to the "1.5M line chaos") has failed big time.
A: Systemd has nothing to do with ZFS. You can use both, one of them, or neither. If you use both, you can integrate them somewhat, but you don't have to.

B: The claim that systemd has failed is ludicrous. It is the default init system on the largest Linux distribution (measured by installed paying customers, not kids with their laptops in dorm rooms). Matter-of-fact, I use it all the time, I have written services for it, and it is perfectly workable. My fear is that systemd is succeeding too well, and will take over the market for init systems so completely (simply because it is better) that we will lose intellectual diversity.

C: Does this mean that I like systemd, or like its creator Lennart, or think that he is an upstanding person or good software designer and engineer? Hell no. I don't like working with systemd, I think Lennart is a sociopath, and his idea of software design is to pile his old solutions on each other and ignore other people and real-world requirements. But for better or worse, the result works well enough, and is taking over. Having an apoplexy over it won't make it go away, and lying and spreading fake news really won't help.

Honestly, your fears might still come to fruition. It's possible that people actually listen to Linus and ZFS gets unsupported from the Linux kernel, and therefore the Linux ecosystem. To me, that would be irrelevant, because I run ZFS on FreeBSD. It is also possible that ZFS on FreeBSD will wither away for lack of volunteers to maintain it, in particular if too much of the volunteer energy gets spent on battles between Linux and other operating systems. If ZFS becomes unusable for lack of maintenance or platform, I would have to migrate my home server to another file system. While painful (it would take several weekends), that's perfectly possible. And given that I have friends in the storage industry, I would probably end up with a technically good solution, even if it requires buying licenses. There are other good file systems, but they are not free, but you can find them.
 
If ZFS becomes unusable for lack of maintenance or platform, I would have to migrate my home server to another file system. While painful (it would take several weekends), that's perfectly possible. And given that I have friends in the storage industry, I would probably end up with a technically good solution, even if it requires buying licenses. There are other good file systems, but they are not free, but you can find them.

What filesystems would these be? ZFS covers quite the range of what a user would want/need in a robust filesystem/block device manager.
 
I see lots of advantages for keeping ZoF and ZoL in one repo ... mostly because I think all those Linux companies provide more automated testing which FreeBSD can also benefit then (at least for zfs matters)
 
C: Does this mean that I like systemd, or like its creator Lennart, or think that he is an upstanding person or good software designer and engineer? Hell no. I don't like working with systemd, I think Lennart is a sociopath, and his idea of software design is to pile his old solutions on each other and ignore other people and real-world requirements. But for better or worse, the result works well enough, and is taking over. Having an apoplexy over it won't make it go away, and lying and spreading fake news really won't help.

I never considered the definition of sociopath as applied to Mr Poettering, but it fits every posting I've seen him make quite well. (Whether the actual person is, in fact, a sociopath is a matter for those who know him personally.)

That being said, I find the assertion "works well enough" quite ironic being posted on the forum of an OS who has rarely accepted that same idea. I have various technical objections to systemd such that I will not use it unless paid rather large sums of money. I will also note that many people often use what they are fed, rather than what they might know is better, and there's no changing basic behavior of the common human.

Still, after reading threads like this, I often wonder: Why do people insist on mindshare as a required criterion for validity of software?

I see that if people believe that software "works well enough, and is taking over" that people will use it. However the hidden implication there is that there is Only One Way To Rule Them All. Why is that even a thing given the intellectual diversity on our planet? Do we have too many operating systems? I don't think we have enough, and not that I use 10 of them either.

Addressing the inevitable "but we don't have enough developers" idea, I do get that people believe you need a critical mass of developers to make forward progress in "some reasonable timeframe", but what ever happened to "it takes as long as it takes"? Why can't the current amount of developers ever be enough? Does anyone (other than me) think that it's more important to have diversity than developer count?

ZFS is easily the best filesystem I've ever used (and I've been around). FreeBSD has done an empirically observed awesome job in implementing it; the best example I have is watching a controller timeout not affect data integrity at all. I'd hate to lose that just because fear of lack of developers ultimately allowed less professional Linux code into FreeBSD. I'm not saying this is happening, or going to happen either. It's just that I'd hate for that to happen given any non-zero chance of that happening, even if that non-zero is epsilon.
 
B: The claim that systemd has failed is ludicrous. It is the default init system on the largest Linux distribution (measured by installed paying customers, not kids with their laptops in dorm rooms). Matter-of-fact, I use it all the time, I have written services for it, and it is perfectly workable. My fear is that systemd is succeeding too well, and will take over the market for init systems so completely (simply because it is better) that we will lose intellectual diversity.
The largest Linux distribution by far is Android, and they do not use systemd. The largest desktop distribution is probably Chromeos (Certainly in the US. I can't find worldwide figures), and they use upstart. I seriously doubt projects like Openwrt or DD-WRT will ever adopt it. So where is it "succeeding"? Arguably on the server, which is deliciously ironic since the original rationale for hatching the monstrosity was to reduce boot times, presumably on machines where someone cares because they're waiting at the console.

I seriously wonder if the likely custom distros places like Google, Facebook and Amazon use feature systemd. It's unlikely we'll ever find out, unfortunately. I'd find it very strange if Google eschewed systemd on Chromeos, but decided to use it in the host OS in their cloud, though.

Those "kids in the dorm room" are likely working on the stuff we'll be using in 10 years. I sincerely hope they're on systemd-free systems, but that's unfortunately unlikely.

I'm glad it works for you, it has made my life miserable, and like Nasrudin I will not work on it unless handsomely compensated to do so.

Edit: Alpine Linux, most commonly-used distro for docker containers: Openrc.
 
That being said, I find the assertion "works well enough" quite ironic being posted on the forum of an OS who has rarely accepted that same idea. I have various technical objections to systemd such that I will not use it unless paid rather large sums of money. I will also note that many people often use what they are fed, rather than what they might know is better, and there's no changing basic behavior of the common human.

Well, its about being able to keep two (or more) ideas / thoughts in one's mind at the same time, instead of just one.
There are many systems / things in today's world and society that I don't like - does that mean that I must abstain from using all of them? Hell no, some of these systems are so useful that living without them would make life very hard (not impossible, just very hard). So it is like this for me: where I have a choice (and remember, this is different from person to person) I use only the systems I prefer, and abstain from using systems I don't like. Where I must (or find life without too hard) - I use systems that I don't like.

The good thing about that is that I know many systems, both good and bad, and are able to discuss their pros and cons based on my actual experience.

It is interesting how you can have technical objections to a system like systemd, yet you claim that you will not use it unless paid "rather large sums of money". How can you have technical objections (as opposed to political / ideological objections) to a system if you have not used it? Baffles my mind.
 
So, on the practical side of things, are we going to see native encryption in ZFS on FreeBSD any time soon? Or never, for some license-related reasons? :)
 
It is interesting how you can have technical objections to a system like systemd, yet you claim that you will not use it unless paid "rather large sums of money". How can you have technical objections (as opposed to political / ideological objections) to a system if you have not used it? Baffles my mind.

In the same way that I know that a car with square wheels is not worth the effort...I find that "using" systemd would be something I do not need to endure to know that it is worth avoiding at most costs.

Usually, a mind that is baffled lacks proper information. If you actually knew my background and experience, I claim you would be less baffled. Then again, presuming that "you think like I do" is the root of most of human misunderstanding. :)

So I've been in computer science a very long time. In my experience systemd has security flaws of the highest order and the needless complexity issue is troubling to say the least. Systemd, for example, violates an observable principle that applications should do their own DNS requests -after the system is booted- and the OS proper should not. (Feel free to challenge those claims scientifically, but perhaps do that in another thread?). So given those assumptions, I've decided to avoid systemd. Besides, this is FreeBSD and I daresay the FreeBSD devs will never go down that road.

I will also claim that using some technology blindly just to avoid worldview/political/ideological bias is just as bad as using or not using the very same technology because of said bias. There isn't a rule based methodology here, use something if you need it, avoid it if it causes you significant concern. That's just what I do.
 
Just a fun fact: the myth busters did a take on square wheels once. I think they got paid for that.
 
Back
Top