Stratis Storage 3.5 Released With Encrypted Cache Support

[phoronix]

Phoronix: Stratis Storage 3.5 Released With Encrypted Cache Support

Red Hat continues investing in Stratis Storage as their modern Linux storage solution built atop XFS and LVM with intentions of providing ZFS and Btrfs like functionality but atop a mature and proven base. Released on Tuesday was Stratis 3.5 as the latest version of their Rust-written daemon...

https://www.phoronix.com/news/Stratis-Storage-3.5

[evert_mouw]

with intentions of providing ZFS and Btrfs like functionality

Those intentions are more than four years old. Both OpenZFS and Btrfs look like they have seen more development in those four years than Stratis. Also Hammer2 and Bcachefs are the horizon. I guess it requires a special kind of situation to like Stratis. Maybe Red Had could re-evaluate this choice of technology. E.g. moving development time from Stratis to Bcachefs would maybe speed up things a bit.

[oiaohm]

Those intentions are more than four years old. Both OpenZFS and Btrfs look like they have seen more development in those four years than Stratis. Also Hammer2 and Bcachefs are the horizon. I guess it requires a special kind of situation to like Stratis. Maybe Red Had could re-evaluate this choice of technology. E.g. moving development time from Stratis to Bcachefs would maybe speed up things a bit.

There are cases where both OpenZFS and Btrfs fail. Cluster file systems setups causes interesting problem. Features that are the reason why someone would want to use OpenZFS and Btrfs comes a problem.

Gluster file system is a cluster file systems that redhat/IBM supports. Take all that all the OpenZFS/Btrfs checksum of file data. Gluster and all major cluster file system already do this no matter the backing storage. Now OpenZFS/Btrfs checksums comes unrequited overhead.

Yes this problem with cluster file systems effects Bcachefs as well.

[evert_mouw]

There are cases where both OpenZFS and Btrfs fail. Cluster file systems setups causes interesting problem. Features that are the reason why someone would want to use OpenZFS and Btrfs comes a problem.

Gluster file system is a cluster file systems that redhat/IBM supports. Take all that all the OpenZFS/Btrfs checksum of file data. Gluster and all major cluster file system already do this no matter the backing storage. Now OpenZFS/Btrfs checksums comes unrequited overhead.

Yes this problem with cluster file systems effects Bcachefs as well.

Sure thing, but how does that relate to Stratis?
Does Stratis integrate with Gluster?

[CochainComplex]

Maybe Red Had could re-evaluate this choice of technology. E.g. moving development time from Stratis to Bcachefs would maybe speed up things a bit.

The point is Bcachefs is completly new. Stratis relies on XFS and LVM both of this solutions are very mature and rock solid. Bcachefs is not even ready for production use yet. IMHO it would even make more sense to shift the resources into BTRFS instead of Becachefs.

[skeevy420]

ZFS and Btrfs like functionality but atop a mature and proven base.

Come on now, Michael. How is ZFS not mature or proven?

oiaohm
To be fair, you can disable ZFS's checksumming as well as use ZFS as a clustered file system.

EDIT: My point is that you have to use the right tool for the right job as well as sometimes you have to configure the tool for the job instead of going with the defaults.

[evert_mouw]

The point is Bcachefs is completly new. Stratis relies on XFS and LVM both of this solutions are very mature and rock solid. Bcachefs is not even ready for production use yet. IMHO it would even make more sense to shift the resources into BTRFS instead of Becachefs.

At least Bcache doesn't eat your data. After all those years, Btrfs has lost confidence for a lot of people.

[oiaohm]

Does Stratis integrate with Gluster?

https://grover.fedorapeople.org/stratis-vault-2017-v4.pdf

Page 8 and the answer is yes. Do note that is 2017. Its not a new integration. Yes do notice Ceph is there as well.

Come on now, Michael. How is ZFS not mature or proven?

Depends on define. Redhat requirements OpenZFS is not proven. Causing issues upgrading kernels for security reasons redhat/ibm does not want to tolerate. The argument must use kernel compatible with OpenZFS does not cut it with Redhat requirements because this can equal not being able to update kernel on security grounds.

oiaohm
To be fair, you can disable ZFS's checksumming as well as use ZFS as a clustered file system.

Checksumming is only the start of the issues. Ceph and Gluster have secure delete features. This is kind of where Stratis intergration comes in to go does this file system that we have just deleted a file on anything that would be keeping copy. Fun right ZFS can you disable COW.

Cluster file systems are serous-ally become pain with Btrfs, ZFS, bcachefs... Basically all copy on write file systems as you end up having to fight against the nature of them to remove functionality the cluster file system will be performing.

Redhat reasons for dropping Btrfs that relate to cluster file system issues almost 100% of them apply to ZFS, bcachefs and so on equally.

My point is that you have to use the right tool for the right job as well as sometimes you have to configure the tool for the job instead of going with the defaults.

For cluster file systems the reason why Redhat went Stratis is is that the COW file systems look like completely wrong tool to them and Redhat developers might be right.

You also have to remember Ceph has what is called bluestore that is it own direct to block device and this is done to remove the performance overhead of having a normal file system.

Lot of the setups redhat is paid to support has cluster file systems with security requirements. So redhat saying no OpenZFS and no Btrfs and no bcachefs could be the totally expected outcome for the client base they are servicing.

You could put it as Redhat tasks need nails and they have hammers and when someone attempts to give them a COW file system is like giving them screws yes more advanced yes can be driven in with a hammer but driven in with hammer the screws don't work exactly right. Yes when some attempt to give Redhat OpenZFS is like given the screws with a head shaped not to be simply hammered.

Stratis basically exists because copy on write file systems is not the right tool for everything.

[evert_mouw]

[...]
Stratis basically exists because copy on write file systems is not the right tool for everything.

I doubt it.

Typically, XFS is recommended but it can be used with other filesystems as well. Most commonly EXT4 is used when XFS isn’t, but you can (and many, many people do) use another filesystem that suits you.

Source: https://docs.gluster.org/en/v3/Insta...mmon_criteria/

Why have a complex layering like Stratis when you can just format a bunch of disks with XFS and make them available to Gluster?

[petete]

At least Bcache doesn't eat your data. After all those years, Btrfs has lost confidence for a lot of people.

While i'm quite hopeful for bacachefs, until it gets into the mainline kernel you won't know how reliable it is. I could try it know, but I'm too lazy to compile a custom kernel.