ZFS/Zsys Improvements Are Already Underway For Ubuntu 20.04 LTS

[k1e0x]

Technically speaking, those Suse ZFS packages aren't included in their default repositories and require using either their Official Experimental Filesystems or some random Community Provided repository and all those Fedora ones are all Community Provided. All Fedora has OOTB is ZFS-Fuse, libguestfs-zfs, and some GoLang ZFS wrappers.

Out of the box, Ubuntu and Manjaro....and apparently Debian from Old Stable to Sid too. Been a while since I searched the Debian repos.

I ran SuSE for a while.. pretty sure I just zypper install zfs. Maybe they pulled it.. either way it wasn't hard. I put it on CentOS pretty easily too. (on root)

[skeevy420]

I ran SuSE for a while.. pretty sure I just zypper install zfs. Maybe they pulled it.. either way it wasn't hard. I put it on CentOS pretty easily too. (on root)

Perhaps, but when I used Suse for a month around March this year, it wasn't available via their official, on by default, repos and the ZoL FAQs all link to that same page you did for Suse support.

As far as Fedora/RHEL/CentOS all go, ZoL provides repositories for those distributions with links available on both their GitHub repo and homepage.

[starshipeleven]

Um, Manjaro has ZFS in their first party repos and had a ZFS installer over two years ago (twas hidden and activated by launching their installer with, IIRC, --experimental...run their installer with --help to see if it is still available).

Yeah ok, ok, Ubuntu AND Manjaro.

I thought only Antergos (now dead) had ZFS support.

[skeevy420]

Yeah ok, ok, Ubuntu AND Manjaro.

I thought only Antergos (now dead) had ZFS support.

For a time I believe they both had the same installer with the same hidden settings and, these days, some of their community isos have basic ZFS on Root support.

That said, I didn't care for it on Antergos. Something about updating to the latest mainline while using zfs-dkms lead to too damn many oh noes upon reboot and I learned to just always keep a LiveUSB plugged into my motherboard's internal USB port (that's something I do regardless these days) for quick chroot kernel downgrades (I had to learn the hard way not to use the restroom during a pacman -Syu ).

And Debian has it too (not via the installer on root though )

[starshipeleven]

Here is SuSE (and Fedora) ZFS packages in base. https://software.opensuse.org/package/zfs

That's not "in base". That's OpenSUSE OBS, which is a package building service similar to Ubuntu's PPA for third party repositories.

Debian's had it for about the past 7 versions also.

Ok they do, it's in contrib repo but it's at least official and enabled by default. Not available in the installer I think, so it's still treated as second-class filesystem, available as an option for those who need it after installation.

It looks like *every* major Linux distro has ZFS available in default repos

No they don't. So far you have only managed to correct me on Debian and Manjaro isn't really a major distro.

Do they not want good things?

Apparently not.

Nobody seems to complain much about the CLOSED SOURCE & OUT OF TREE graphics driver.

You haven't been much on Phoronix right? Every thread about NVIDIA there is someone complaining about that, and how out-of-tree drivers do break when you update some component they rely on.

So is this just a pissing contest between "My open source license is better than your open source license"?

No, why every time someone pulls ZFS out then it must be a license zealot.

I don't even think that is true

Your opinion on legal matters isn't worth much. Companies with legal teams have decided, even Canonical's legal team didn't tell them they could try to push ZFS to mainline, the best they are allowed to do is to ship it in the initramfs as a module.

And... btw.. have you looked at the features of zsys they are implementing?

I agree on having a decent interface for zfs functionality. One of the reasons I'm on OpenSUSE is that it provides some interface for btrfs advanced functionality.

You are legitimately hurting Linux users and Linux adoption by trying to suppress this technology

I'm just stating facts. I'm not responsible if distros aren't all embracing ZFS.

just remember.. we could have had those features 10 years ago when ZFS was opensourced.

What I said still applies. Next-gen filesystems are a killer feature. Major distros that are also a commercial product are all using different next-gen filesystem to lock-in their customers, and it's not strange.

The "killer" effect of the "killer feature" will go into effect once Ubuntu's ZFS support and management utilities are stabilized and production ready (won't take long, even for fucking noobs like Canonical), just watch how Illumos finally dies off and FreeNAS begins to struggle for air.

[starshipeleven]

I ran SuSE for a while.. pretty sure I just zypper install zfs. Maybe they pulled it.. either way it wasn't hard. I put it on CentOS pretty easily too. (on root)

This is a OpenSUSE Tumbleweed system with default repos + packman for the proprietary/licensed media codecs and such

sudo zypper search zfs
[sudo] password for root:
Retrieving repository 'packman' metadata ..............................................................................................................................................[done]
Building repository 'packman' cache ...................................................................................................................................................[done]
Retrieving repository 'repo-debug' metadata ...........................................................................................................................................[done]
Building repository 'repo-debug' cache ................................................................................................................................................[done]
Retrieving repository 'repo-non-oss' metadata .........................................................................................................................................[done]
Building repository 'repo-non-oss' cache ..............................................................................................................................................[done]
Retrieving repository 'repo-oss' metadata .............................................................................................................................................[done]
Building repository 'repo-oss' cache ..................................................................................................................................................[done]
Loading repository data...
Reading installed packages...

S | Name              | Summary                                            | Type  
--+-------------------+----------------------------------------------------+--------
  | lzfse             | Reference C implementation of the LZFSE compressor | package
  | lzfse-debuginfo   | Debug information for package lzfse                | package
  | lzfse-debugsource | Debug sources for package lzfse                    | package
  | lzfse-devel       | Reference C implementation of the LZFSE compressor | package

I don't see any zfs in there I'm afraid.

[k1e0x]

No they don't. So far you have only managed to correct me on Debian and Manjaro isn't really a major distro.

I'm not correcting you and I'm not talking to you. You're not an intelligent person and you're comments have very little value other than troll humor. Why would I correct a fool?


You're like that nerd in the conversation that is all "Welll... Actually.." in the most nasally annoying voice ever.

[kgardas]

If so much clients cared about checksumming, Windows server would not exist, RHEL would not exist, SUSE enterprise would not exist and everyone would be using Solaris and FreeBSD as those were the only ones that had ZFS support for a long while.

The entire point of RAID is to protect agains disk failure, data integrity provided by hardware ECC in the drives themselves is enough for most single server customers.

Seriously, how did the world survive when ZFS didn't exist? Servers didn't blow up, data wasn't lost and so on, it matters only in very rare cases.

Actually SuSE supports btrfs and you may be surprised but Window Server supports something called ReFS. W.r.t. your claim about not needing checksuming -- this is just your claim. I know that ZFS saved me from a lot of corruptions when dealing with not so good PSU and not so good (old) SATA cables. The funny part is: you will probably never see issue like that as you don't checksum everything automatically...

[skeevy420]

Actually SuSE supports btrfs and you may be surprised but Window Server supports something called ReFS. W.r.t. your claim about not needing checksuming -- this is just your claim. I know that ZFS saved me from a lot of corruptions when dealing with not so good PSU and not so good (old) SATA cables. The funny part is: you will probably never see issue like that as you don't checksum everything automatically...

You just told that to the biggest BTRFS and SUSE supporter on Phoronix

EDIT: ROFL

[starshipeleven]

Actually SuSE supports btrfs and you may be surprised but Window Server supports something called ReFS.

ReFS is kind of meh and only barely comparable to true next-gen filesystems like btrfs and ZFS.

No compression, as if NTFS compression wasn't complete and utter shit that divides write performance by a factor of 10 even on a RAID card with 512MB cache.
Windows server provides deduplication regardless of the filesystem but it's not the same thing, OK it does shrink down stuff like VM virtual drives and that's the least they could do, but we are still talking of an active process doing this job, similar to an online checkdisk, it's not done trasparently by the filesystem. And of course I can do it on NTFS too so what's the point of ReFS.

No dedicated snapshotting feature (Volume Shadow Copy works on NTFS as well, as they implemented this in software already, it's not as instantaneous as with real filesystems like btrfs or ZFS of course), no quotas. Again, a feature that actually matters on a server. Not saying VSS is bad, but it's not a hot babe either.

It is CopyOnWrite so it can survive sudden power-off or being disconnected while writing, but when does that even come into the real life of an actual server? If you have a Windows server doing anything important without an UPS you are a moron asking for trouble anyway, and the datacenter isn't a place where a device can get disconnected randomly while being written on.

It's not usable outside of Server (where it would be most useful), nor usable for boot partition (same).

It has some advanced features for bigger server usecases, and some more stuff for virtualization servers, but that's it. It's not supposed to replace NTFS, it's supposed to be used in bigger servers that need special things.

W.r.t. your claim about not needing checksuming -- this is just your claim. I know that ZFS saved me from a lot of corruptions when dealing with not so good PSU and not so good (old) SATA cables. The funny part is: you will probably never see issue like that as you don't checksum everything automatically...

No it isn't just my claim. Server hardware has significantly higher quality than normal consumer hardware, SATA is just a pallid crappy subset of what SAS protocol does (enterprise drives are usually SAS, and even when they are not they are usually attached to a SAS controller anyway), and RAID cards with dedicated cache have always been good enough in a system where you have UPS and stuff.
SAS drives are on a whole another level of reliability too.

On servers you commonly just throw away drives at the first sign of potential future failure, like for example when there is a bad sector. You know how many years I've used drives with 1-5 bad sectors? 9 years. Still perfectly fine. I have some Seagate 1.5 TB drives that have from 50 to 100 bad sectors (yes it's normal, they are seagate drives, I still love them), and their SMART tells me they have been online for 8 years now. Still perfectly fine, the bad sectors aren't increasing in years, SMART self-test comes out clean.

On servers you get alarms and all hell breaking lose if one of the two (or more) redundant PSUs goes out of spec, and the board just shuts it off and keeps running on the other one(s) and waits for someone to come and hotswap in a new one.

On Windows servers you have Windows (so there is the first issue), and on them you will have crappy applications that can and do act weird and sometimes lose data because they are coded like shit, so the one-in-a-thousand time that hardware corrupts your data will not be noticed and the same usual shit application will take the blame.

This is the same kind of discussion we can have about ECC, do you really need it? probably not. Most low and midrange servers don't need it either. But it's better to have it than not having it. It's not like you are paying a fortune for your RAM anyway, so why settle for less than the best to save 10-20$ per RAM bank.

So yes the type of hardware is completely different, the environment is completely different, and the user is also different.

That said, yes, for home use where reliability is lower, and there isn't someone always checking that everything is OK, checksumming is great, and I use it every day.

I did use btrfs to make a large array (10+) of random mismatched drives I have laying around and would not otherwise use, with its own ATX power supply and it's all accessed by a SAS expander (and I have a SAS card in my PC).