userPlaceholder pods stuck Pending in zero-to-jupyterhub v0.8.2

I am trying to configure zero-to-jupyterhub v0.8.2 on DigitalOcean to use placeholder pods to have some idle capacity available to users. (Currently autoscaling takes 5-10 minutes to create a new node.) However, when I enable this feature (config below), the placeholder pods are always in the pending state and any new user pods get stuck in the pending state; the autoscaler is not triggered.

I have tried the following configurations:

scheduling:
  podPriority:
    enabled: true
    globalDefault: true
    defaultPriority: 10
    userPlaceholderPriority: 0
  userPlaceholder:
    enabled: true
    replicas: 3
  userScheduler:
    enabled: true

and

scheduling:
  podPriority:
    enabled: true
    globalDefault: false
    defaultPriority: 0
    userPlaceholderPriority: -10
  userPlaceholder:
    enabled: true
    replicas: 3
  userScheduler:
    enabled: true

removing the following seems to have fixed it

userScheduler:
enabled: true

I guess you may be using a quite modern version of k8s, like 1.16+? I think it would work if you upgrade to 0.9.0-beta.4

Thanks, I am indeed using k8s v1.16 and upgrading to 0.9.0-beta.4 seems to have solved my issue. Do you happen to known when 0.9.0 will be released?

:wink: I’ve told soon for a very long time. I’ve been a bit afraid of the burst of responsibility I would feel to ensure people do okay in the official transition without being certain I have time available to help out.

@mattwelborn I’ll make a push towards it tomorrow, now is a good time.