Merge pull request #1713 from yuvipanda/nph

Keep a placeholder node around for UToronto Hub

config/clusters/utoronto/prod.values.yaml

@@ -1,4 +1,17 @@
 jupyterhub:
+  scheduling:
+    userPlaceholder:
+      # Keep at least one spare node around
+      replicas: 1
+      resources:
+        requests:
+          # Each node on the UToronto cluster has 59350076Ki of RAM
+          # You can find this out by looking at the output of `kubectl get node <node-name> -o yaml`
+          # Look under `allocatable`, not `capacity`. Unfortunately then you have to fiddle with it to
+          # find the right number that's big enough that no user pods will schedule here, but small enough
+          # that pods in `kube-system` will still schedule.
+          # So even though this is under `userPlaceholder`, it really is operating as a `nodePlaceholder`
+          memory: 57350076Ki
   hub:
     db:
       pvc: