XFS For Linux 4.15 Brings "Great Scads of New Stuff"

[pal666]

I was mentioning present, not past

you mainly mentioned your fantasies

[azdaha]

so is decreasing. it can't decrease size - it does not fulfill requirements for "size is changeable"

lol that made me chuckle; mainly because I understood exactly what you meant, or what you were going to say

[Awesomeness]

redhat lost all btrfs devs.

And by “lost” you mean that they had the choice to work on XFS or leave, right?

I've got impression most of "big names" behind file systems have left RH.

Maybe the "big names" should have done proper work instead of bathing in fame. As I wrote already: Not even btrfs proponent SUSE uses btrfs for /home by default. They use XFS – developed by SGI and Red Hat.

EXT4 ppl gone Google and few other companies, maybe it explains why RH no longer fond of EXT4.

Nobody with half a brain is fond of Ext4. Samsung developed F2FS to get rid of Ext4 on phones.

Their mumbling about trying to turn XFS+LVM+raid+some glue on rust and python to btrfs/zfs/etc like thing does not sounds exciting as well.

Doesn't need to sound exciting. Filesystems need to work reliably.

[azdaha]

Nobody with half a brain is fond of Ext4. Samsung developed F2FS to get rid of Ext4 on phones.

Great counter-arguments in the post; however, I'm not sure what the above means exactly. Do you have further references or info about the implied problems that your very strong statement makes about EXT4? I had not heard of this.

Thanks

[timofonic]

you mainly mentioned your fantasies

Okay, you won. Those are my sexual fantasies. Are you happy now?

Nevermind...

[SystemCrasher]

Maybe the "big names" should have done proper work instead of bathing in fame.

Somehow, when I'm looking on e.g. btrfs internals I'm really in mood to tell that thing still happens. A daring new design, which is meant to address plenty of long standing issues and got enough room for evolution is looking good to me. Sure, it isn't without its weak spots and it always takes time to get advanced designs working right. And XFS... well, its XFS. Just another design of its time, more or less on par with anything of these ages. Sure, it got some strong points, and it is definitely possible to squeeze some extra juice, but well, pigs aren't really meant to fly. So, honestly, I do not get the point of all this buzz.

As I wrote already: Not even btrfs proponent SUSE uses btrfs for /home by default. They use XFS – developed by SGI and Red Hat.

So what? Just yelling big names does not makes things to work better. Nor it magically improves filesystem design. And, honestly, running /home on btrfs and eventually doing few snapshots could be good idea. Just in case I've seriously screwed something up and e.g. rm -rf-ed some valuable dir. What XFS could do about it? Oh, wait, I'm meant to use thing like LVM for snapshots? To the hell with it. Its management is awful, its performance is crap, and no python or rust cruft could fix the later. Not to mention I'm not in mood to deal with codermonkey crap code instead of proper system-level tools. Especially when it comes to my data.

Nobody with half a brain is fond of Ext4. Samsung developed F2FS to get rid of Ext4 on phones.

F2FS has been designed to run on flash storages from scratch. Surely its IO patterns are more convenient to flash-based storages. But still, there is little story.
Recently I've break-n-entered some high-end Android device, running quite new Linux kernel and so on, getting full root. So what do we have here in some practical device? Like a dozen and half of partitions, using EXT4 and F2FS and maybe something else I've forgot. Some zram fun, I like this thing too. Some filesystem-level encryption, etc. It sounds fun but on downside, creating workable backup proven to be quite a challenge. Well, root could screw things up and I've been really up for some system-level adventure. Not exactly safest one. Then there was ExFAT and NTFS3G in FUSE. So dat thing could actually handle nearly any filesystem around, though windows things "enjoy" by speed penalty of FUSE, I guess they want to evade paying patent fees or something. Weirdest thing about it? Well, it reads F2FS and EXT4, but it wouldn't work on SD card. Which is really lame to my taste, since FAT32 implies you can't have files over 4G and ExFAT implies speed penalty of FUSE. So if card is here to expand device memory, F2FS would clearly be best bet, but Google proven to be weird enough to disable it somewhere on filesystem probing level or so.

Doesn't need to sound exciting. Filesystems need to work reliably.

Is this really about XFS? XFS has zeroed files for many years "for the sake of security" each and every crash. Eventually this shit got more or less fixed, but alas, XFS does not even enjoys by "full" journalling, it only cares to journal metadata. This implies XFS does not really gives a fuck what happens to data on crash/powerloss. So I'm not going to buy this marketing BS. It grossly lacks any technical stuff behind it, and I'm not okay with marketing BS instead of proper technical solutions.

[Awesomeness]

A daring new design, which is meant to address plenty of long standing issues

So maybe, just maybe, in 10 years btrfs will finally be usable and not randomly eat data. Great. Looking forward to it.

[SystemCrasher]

So maybe, just maybe, in 10 years btrfs will finally be usable and not randomly eat data. Great. Looking forward to it.

I'm feeling myself like a time traveller, because it does not eats my data here and now. And XFS... well, it just doomed to eat data, unless RH is willing to do something really drastic about full (data+metadata) journalling or some equvivalent of it. But of course, it is possible to ignore the problem and push loud marketing instead of technical solution.

[azdaha]

I'm feeling myself like a time traveller, because it does not eats my data here and now. And XFS... well, it just doomed to eat data, unless RH is willing to do something really drastic about full (data+metadata) journalling or some equvivalent of it. But of course, it is possible to ignore the problem and push loud marketing instead of technical solution.

What's the status of XFS journaling right now? Is the `xfs-dump` feature related to this or equivalent of what you're indicating or does that only deal with metadata?

[insanemal]

I'm feeling myself like a time traveller, because it does not eats my data here and now. And XFS... well, it just doomed to eat data, unless RH is willing to do something really drastic about full (data+metadata) journalling or some equvivalent of it. But of course, it is possible to ignore the problem and push loud marketing instead of technical solution.

HAHAH this is the single most uninformed thing I have read all day.

Do you know ANYTHING about XFS?

XFS is an integral part of a product called DMF. DMF used to be made by SGI and is now made by HPE.

There are litterally 100's of PB of data on XFS filesystems in DMF and as the DMF 'tag line' goes "With ZERO BYTES OF DATA LOST TO DATE"

XFS doesn't eat data.

XFS has actually got some almost-god-tier developers working on it. Have you heard of Dave Chinner?

Also XFS+LVM/DM is not a stupid solution, getting your filesystem doing RAID is a stupid solution. Way to mix your block and file layers up in stupid ways