Nixpkgs: Does beesd actually work?

[davidak]

Describe the bug

I have a NixOS system with 250 LXC containers (running NixOS) from the same LXC image. They are initially about 600 MB and content is identical. So i thought that this is a perfect scenario for deduplication, but only 10 GB are saved from 170 after i let beesd do its thing for 4 days.

To Reproduce

I might try to reproduce it with a simpler setup when i have time.

Steps to reproduce the behavior:

1. ...
2. ...
3. ...

Expected behavior
250 x 600 MB are deduplicated to 1x 600MB, since the nix store is read only and never changes

Additional context
Add any other context about the problem here.

Notify maintainers
@charles-dyfis-net

Metadata
Please run nix-shell -p nix-info --run "nix-info -m" and paste the result.

Maintainer information:

# a list of nixpkgs attributes affected by the problem
attribute:
# a list of nixos modules affected by the problem
module:

[charles-dyfis-net]

It works for me -- I use it in practice. That said -- typically, if you're running a bunch of lxc containers, you'd want to directly mount a shared Nix store into all of their namespaces, so that shared store wouldn't be duplicated (and require deduplication) in the first place.

bees really stands out in scenarios where file-by-file deduplication isn't good enough -- if you _can_ do file-by-file, using xfs_io dedupe or even cp --reflink=always (caveat that using cp loses the guarantees that a dedupe operation will be nondestructive) is your friend.

Anyhow -- if you have a lot of content you know isn't being deduplicated, the first place I'd start is inspecting /.beeshome/beesstats.txt, and taking any remedial action implied (if the hash table is full, f/e, you know it needs to be larger); followed by reading through https://github.com/Zygo/bees/blob/master/docs/wrong.md, making sure you aren't doing something like btrfs defrag that undoes bees' work (or leaving autodefrag turned on), etc.

The second place I'd start is talking to upstream.

[davidak]

thank you very much for the fast and helpful response!

if you're running a bunch of lxc containers, you'd want to directly mount a shared Nix store into all of their namespaces, so that shared store wouldn't be duplicated (and require deduplication) in the first place.

can you give me a hint how to do that in practice?

[charles-dyfis-net]

if you're running a bunch of lxc containers, you'd want to directly mount a shared Nix store into all of their namespaces, so that shared store wouldn't be duplicated (and require deduplication) in the first place.

can you give me a hint how to do that in practice?

Something like:

lxc.mount.entry = /nix/store nix/store none defaults,bind.ro 0.0

[charles-dyfis-net]

Closing this ticket for now. Wish you luck in getting something that works for you.