Btrfs "Reserve Flush Emergency" Feature Heading To Linux 6.2

[jbean]

SUSE Enterprise has been using btrfs since a long time.
Oracle Unbreakable Linux supports it with its custom kernel.

As for Redhat, you can find the reasons here: https://news.ycombinator.com/item?id=14907771

Basically they don't have qualified engineers. Which anyway raise the question of why customers don't ask for it so that RH acquires engineers to work on it, but that's it.

Btrfs still has a lot of room for optimizations and improvements before it's really suitable to be a general "set and forget" replacement for enterprise users like MD/LVM RAID + XFS is. Don't get me wrong, for many users, enterprise included it can be great, Facebook afterall has made it work. However facebook has also made it clear that for database workloads, they still use XFS for obvious reasons. They too know the limitations. It can be a *massive* performance slouch compared to just about any other filesystem, ZFS included. Hosting things like databases and VMs can be a massive anti-usecase with Btrfs with the current on disk format and functionality. Sure, it works great for home and small business users. I pretty much use it for everything I have and manage that isn't databases. But as soon as you need to handle a lot of random writes for enterprise level traffic, it bogs down a lot.

When we start seeing a lot of performance improvements land, like extent tree v2, faster quota support/handling, and most importantly better RAID handling (and I'm not just speaking about RAID5/6, I mean basic things, like automatic device resilvering or marking a device dirty or addressing the NOCOW writehole on all RAID profiles), we may see demand go up and in turn Redhat will become more interested in supporting it.

I mean even kernel 6.1 just got a sysfs tunable to make quotas a little bit more usable when deleting snapshots (which right now, are almost unusable performance wise in the presence of snapshots). That is a bit hacky given the extent tree v2 approach though. So as it stands now, it's these "death by a thousand cuts" cases that really just doesn't make it suitable for a 10 year supported enterprise operating system.

If you're a desktop linux user, you should probably use it already for all the benefits it does provide as the performance downsides are not noticeable for this use case. For everyone else in the enterprise space who have a good Btrfs use case and know how to manage it, like Facebook does, there's lots of alternatives to get support in the meantime.

[cynic]

Since Facebook is using it for their fleet doesn't that mean that BTRFS is enterprise ready?
Why isn't BTRFS the default filesystem on all the major distros already? What specifically is holding it back?

SUSE Enterprise has been using btrfs since a long time.
Oracle Unbreakable Linux supports it with its custom kernel.

As for Redhat, you can find the reasons here: https://news.ycombinator.com/item?id=14907771

Basically they don't have qualified engineers. Which anyway raise the question of why customers don't ask for it so that RH acquires engineers to work on it, but that's it.

[jbean]

The news article specifically says that actual on-disk free space is not the problem.

Sort of. My understanding is that it's getting close to running out though at this point (even though it hasn't truly due to reserves), for cases where things are nearing a true ENOSPC and the filesystem is heavily fragmented. Btrfs being a COW filesystem has always struggled with ENOSPC when nearing full, as it's a complicated thing to get right as you don't really always know how much metadata you need to write with each change.

[Developer12]

Hardware RAID doesn't provide data integrity guarantees at all, so I'm not sure how you were able to get to that conclusion about not trusting Btrfs.

Ask facebook why they're doing it this way then.

[pkese]

Performance isn't great for many workloads ?

The fact there are still fixes for BTRFS in every other stable kernel version ?

There are fixes for all filesystems.

This just went into 6.1 kernel for XFS: "a tricky data corruption issue that I still haven't entirely solved" yet nobody is claiming that XFS is not a stable filesystem.

[GingerDog]

Since Facebook is using it for their fleet doesn't that mean that BTRFS is enterprise ready?
Why isn't BTRFS the default filesystem on all the major distros already? What specifically is holding it back?

Performance isn't great for many workloads ?

The fact there are still fixes for BTRFS in every other stable kernel version ?

[intelfx]

Facebook runs it on top of hardware raid. They don't trust it to maintain data integrity. It's sole purpose is to provide snapshotting.

Hardware RAID doesn't provide data integrity guarantees at all, so I'm not sure how you were able to get to that conclusion about not trusting Btrfs.

[Developer12]

Since Facebook is using it for their fleet doesn't that mean that BTRFS is enterprise ready?
Why isn't BTRFS the default filesystem on all the major distros already? What specifically is holding it back?

Facebook runs it on top of hardware raid. They don't trust it to maintain data integrity. It's sole purpose is to provide snapshotting.

[intelfx]

Why is Facebook running their machines with so little free space in the first place?

The news article specifically says that actual on-disk free space is not the problem.

[ATLief]

Why is Facebook running their machines with so little free space in the first place?