Open strategicpause opened 3 weeks ago
Not sure if this is helpful, but this is the same command when evaluating Bottlerocket a couple years ago. It's pretty old, but at least shows some precedence for faster boot up times.
bash-5.1# systemd-analyze
Startup finished in 2.015s (kernel) + 7.070s (userspace) = 9.086s
multi-user.target reached after 7.069s in userspace
bash-5.1# systemd-analyze blame
1.839s wicked.service
538ms settings-applier.service
467ms docker.service
439ms host-containerd.service
multi-user.target reached after 29.098s in userspace.
This seems like a significant outlier for an aws-dev
boot. At the other extreme, under KVM locally, I get numbers like this:
bash-5.2# systemd-analyze
Startup finished in 616ms (kernel) + 2.226s (userspace) = 2.842s
multi-user.target reached after 2.226s in userspace.
bash-5.2# systemd-analyze blame
443ms systemd-networkd-wait-online.service
331ms docker.service
265ms dev-dm\x2d0.device
261ms dev-vda3.device
229ms load-crash-kernel.service
167ms systemd-udev-trigger.service
141ms systemd-resolved.service
137ms activate-configured.service
132ms systemd-journald.service
122ms soci-snapshotter.service
117ms systemd-logind.service
111ms settings-applier.service
103ms containerd.service
There may be some combination of CPU or I/O throttling at play in the 29 second case.
Some data points for a t3.small launch in EC2:
multi-user.target reached after 7.168s in userspace.
docker.service +1.132s
└─containerd.service @5.675s +315ms
└─soci-snapshotter.service @5.208s +414ms
multi-user.target reached after 9.413s in userspace.
docker.service +1.525s
└─containerd.service @6.797s +1.034s
└─soci-snapshotter.service @5.769s +1.002s
multi-user.target reached after 7.835s in userspace.
docker.service +1.065s
└─containerd.service @6.488s +227ms
└─soci-snapshotter.service @6.061s +383ms
The recent addition of the SOCI snapshotter dependency definitely delays the containerd
start, but overall the numbers shouldn't be too far off your older sample.
Image I'm using:
aws-dev
variant using core kit v2.3.5. Built using the following commands:What I expected to happen: I expect containerd to start up in ~1.5s after running
/bin/systemctl try-restart containerd.service
.What actually happened: It takes roughly ~6s for containerd to start.
How to reproduce the problem:
systemd-analyze blame