A brief timeline from the logs in the defect (mostly what Al already found):
13:37:51 - controller-1 unlocked
13:41:16 - controller-1 finishes booting
13:44:06 - controller manifest finishes applying
13:44:14 - kubelet starts
13:44:47 - worker manifest finishes applying
13:51:14 - nova-compute container starts
13:51:28 - hypervisor up
The kubelet starts creating containers at 13:44:16. Creating all the containers takes a long time (about 7 minutes) because:
- It has to create 178 containers.
- There are dependencies between containers.
- These containers are all affined to two cores.
I took a brief look at the gaps between container creations and found one large gap - there was a 1m48s gap between dockerd launching the calico-node container before launching the next one. Maybe that is expected?
In any case, nothing jumps out here as a specific problem, other than a lot of containers being created, using only two cores.
A brief timeline from the logs in the defect (mostly what Al already found):
13:37:51 - controller-1 unlocked
13:41:16 - controller-1 finishes booting
13:44:06 - controller manifest finishes applying
13:44:14 - kubelet starts
13:44:47 - worker manifest finishes applying
13:51:14 - nova-compute container starts
13:51:28 - hypervisor up
The kubelet starts creating containers at 13:44:16. Creating all the containers takes a long time (about 7 minutes) because:
- It has to create 178 containers.
- There are dependencies between containers.
- These containers are all affined to two cores.
I took a brief look at the gaps between container creations and found one large gap - there was a 1m48s gap between dockerd launching the calico-node container before launching the next one. Maybe that is expected?
In any case, nothing jumps out here as a specific problem, other than a lot of containers being created, using only two cores.