** Description changed:

- This is a companion to https://bugs.launchpad.net/snapd/+bug/2165972.
- However it does not seem to be fixed by PR#17626.
+ [Impact]
+ 
+  *
+ 
+ [Test Plan]
+ 
+  *
+ 
+ [Where problems could occur]
+ 
+  *
+ 
+ [Other Info]
+ 
+  *
+ 
+ ---original---
+ This is a companion to https://bugs.launchpad.net/snapd/+bug/2165972. However 
it does not seem to be fixed by PR#17626.
  
  I've been working on improving Workshop VM launch times, see this draft:
  https://github.com/canonical/workshop/pull/1042. I found that (at least
  sometimes) a significant amount of the boot is spent waiting for snapd
  to refresh the catalog. In my draft I simply create an empty file at
  /var/cache/snapd/names using cloud-init; this saves about 2s on average
  (with my NZ internet).
  
  Here is a quick reproducer:
  ```
  #!/usr/bin/env bash
  set -e
  
  trap 'lxc delete -f v1 v2' EXIT
  
  workaround='#cloud-config
  bootcmd:
  - install -D --mode=0644 /dev/null /var/cache/snapd/names
  '
  
  for vm in v1 v2; do
        lxc init --vm ubuntu:24.04 "$vm"
        if [ "$vm" = v2 ]; then
                lxc config set v2 cloud-init.user-data="$workaround"
        fi
        lxc start "$vm"
        until lxc exec "$vm" -- true &>/dev/null; do sleep 1; done
        lxc exec "$vm" -- systemctl is-system-running --quiet --wait
        lxc exec "$vm" -- systemd-analyze --no-pager blame
        lxc exec "$vm" -- systemd-analyze --no-pager critical-chain
  done
  ```
  
  In one run, the workaround took snapd.seeded.service from 3.615s to
  2.098s. I'm using an Ubuntu 24.04 host with LXD 6.9 and a ZFS storage
  pool backed by a fast SSD. YMMV.
  
  I don't really want to use this workaround, because I think it probably
  misleads snapd into thinking the catalog was refreshed recently. I just
  want to move the catalog refresh off the critical path. I think a lot of
  workshops don't really use snaps, since they can use the equivalent SDK
  instead, so it's not worth paying the cost of refreshing a catalog that
  might never be used. I slopped together a couple of patches here:
  https://github.com/jonathan-conder/snapd/commits/fast12/ which might
  help with that.
  
  I have a more involved reproducer which validates these patches against
  a baseline of PR#17626 rebased on 2.77.1; see attached.
  
  Both patches give roughly similar speed improvements and combining them
  doesn't seem to have a big impact over just one of them. But I've been
  running my benchmarks in a noisy environment. I'll leave it to your
  judgement as to whether either of them is actually a good idea; maybe
  you can come up with something more suitable.

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2167816

Title:
  snapd network requests slow down LXD VM boot

To manage notifications about this bug go to:
https://bugs.launchpad.net/snapd/+bug/2167816/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to