#332 consider switching containers to use btrfs driver (instead of overlay)
Opened 2 years ago by chrismurphy. Modified 2 years ago

/etc/containers/storage.conf provided by containers-common defaults to overlay storage driver for podman. There is also a btrfs driver, which instead isolates each container using subvolumes (either read-only or writable snapshots, as the case may be).

Note that this issue explicitly absent in Changes/FedoraCloudBtrfsByDefault. So it should probably get figured out sooner than later.

An advantage of changing both defaults for Fedora 35: we don't have to later worry about what to do about btrfs+overlayfs → btrfs+btrfs upgrades. Switching drivers doesn't yield good results right now; you have to export containers→modify storage.conf→podman system reset→import containers. Otherwise things get messy and/or break.

Related issues:
Cloud: Btrfs as the default file system
Fedora btrfs: cri-o should work out of the box on btrfs
Containers/storage: CRI-O should work out of the box on btrfs
Containers/podman: -closed- automatically use btrfs driver if on btrfs #6563


Metadata Update from @ngompa:
- Issue tagged with: meeting

2 years ago

Metadata Update from @ngompa:
- Issue marked as depending on: #308

2 years ago

We need some regression tests for containers that tend to exercise the graph driver (whether VFS, dm-thin, overlayfs, btrfs) - i.e. the tests aren't driver specific, they exercise the routine container creation, replication, modification, destruction tasks in a way that "hello world" alone does not. Tests probably need to thread these tasks concurrently and in volume, like create 1000 containers while deleting 100 while modifying 50. That sort of thing.

And we need to confirm rootless and rootful cases work, and have expected performance.

If the tests could be timed, that'd be useful performance information to have for regression testing the same driver, but also as a comparison what kinds of workloads favor the btrfs driver.

Oh yeah :neutral_face: does anyone have ideas for what regression tests would be appropriate for validating graph drivers, or who to ask?

Metadata Update from @chrismurphy:
- Issue untagged with: meeting

2 years ago

Login to comment on this ticket.

Metadata