On Wed, Mar 16, 2016 at 11:20 AM, Neal Becker <ndbecker2(a)gmail.com> wrote:
If I had known that, I might not have chosen btrfs for my
filesystems. Came
as a rude surprise when I suddenly needed more swap space, and couldn't do
it.
You might check the FAQ and also the Gotchas list to make sure you're
not missing something else.
https://btrfs.wiki.kernel.org/index.php/FAQ
https://btrfs.wiki.kernel.org/index.php/Gotchas
Swap expects to get a contiguous set of sectors, whether it's on a
partition or a filesystem. But Btrfs doesn't work this way at all, all
metadata and data extents are logical addresses internally, those get
translated by a chunk tree to reference device(s) and sector the data
is located on. The way balancing and allocation work, those chunks can
be moved at anytime to any device or even across multiple devices.
Getting swapfiles support means work for swap as well as Btrfs.
https://btrfs.wiki.kernel.org/index.php/Project_ideas#Swap_file_support
https://lwn.net/Articles/625412/
But it's an open question what limitations this will have for a
snapshottable cow file system even if it gets finished. I'm kinda
thinking maybe the file would need to go in its own subvolume that
gets some kind of nosnapshot attribute? I'm not really sure but
snapshotting swap seems like a bad idea.
--
Chris Murphy