Add a Blame tab to Services Page

[makinikm]

Hi,
As part of performance optimizations or after upgrades and installations I sometimes use 'systemd-analyze time' and 'systemd-analyze blame'. I thought it could be a nice feature for Cockpit to have on it's Services page as a tab - a list of systemd units sorted by their boot time launch timers, styled the same as the main Services tab - along with description, preset and state, and a link to the service' page for a name. The output of 'analyze time' could go as a header under the tab name, above filter names.

The data from 'systemd-analyze plot --json=short' could be used for the units launch timers, but I have no idea how to integrate that into Cockpit's sorting. After taking a short look at the code I'd rather leave that here as an idea to be considered for Core, rather than try vibehacking anything presentable or as a plugin...

Does it make any sense?
Requesting for comments here.
👀

[jelly]

I was looking at systemd-analyze plot > svg` again, and that seems wholly unusable by how wide the kernel time is :-)

But that was with the idea of presenting the same sort of data in Cockpit, your idea seems more integrated but for really knowing why something is slow you'll need to know why something is blocked by another unit/dependency. Just presenting "649ms NetworkManager.service" is slow isn't too useful. The relationship is important and that things run in parallel.

[jelly]

It is definitively interesting but not super trivial to integrate, it needs some more thought.

[makinikm]

It all started from plot >svg, of course ;) I even considered adding it as a transparent desktop background layer for some time, but it is indeed unusable and scaling it doesn't help. On that note there's this new gem:
systemd-analyze dot 'avahi-daemon.*' | dot -Tsvg > ~/avahi.svg ...

In regards to the NetworkManager.service - all true, it is indeed the "wait for network" service... on desktop systems, or maybe just systems with NetworkManager;) On a system without it, the output of 'blame' appears like this:

[root@gppi01:~] # systemd-analyze blame
10.143s apt-show-versions.service
 7.249s fstrim.service
 6.191s ifup@eth0.service
  839ms postfix.service
  805ms ifupdown-pre.service
  578ms lvm2-monitor.service
  566ms pmlogger.service
  432ms dev-nvme0n1p1.device
...

Jokes aside, I think the output of systemd-analyze critical-chain [UNIT...] is more alike to a boot times dependency graph, and somehow, visually - such format could be used for something similar in Cockpit's UI design, for example if added below (or above) the 'Service logs' section on each Service' page for when opened via the Blame tab. I have no idea how to get that data out of systemd for this, since the output is formatted there and it has no data export options (so far).
E.g. systemd-analyze critical-chain apt-show-versions.service:

apt-show-versions.service +10.143s
└─basic.target @1.742s
  └─sockets.target @1.742s
    └─cockpit.socket @1.649s +91ms
      └─sysinit.target @1.638s
        └─apparmor.service @1.497s +140ms
          └─local-fs.target @1.490s
            └─mnt-storage.mount @8.634s +241ms
              └─mnt-storage.automount @1.024s
                └─local-fs-pre.target @1.023s
                  └─lvm2-monitor.service @444ms +578ms
                    └─systemd-journald.socket @422ms
                      └─-.mount @375ms
                        └─-.slice @375ms