search

OpenSourceBrain / OSBv2

An updated version of the Open Source Brain platform
https://www.v2.opensourcebrain.org/
Other
12 stars 6 forks source link

Separate jupyterhub pods from server pods #451

Closed filippomc closed 2 years ago

filippomc commented 2 years ago

Jupyterhub pods are user pods that are assigned to users upon opening a osbv2 workspace or nwb explorer / netpyne session. Separating the user pods from the server pods is good for security and to avoid these pods taking over server resources.

filippomc commented 2 years ago

In order to make a proper deployment and separation we need the following configurations to be fullfilled:

  1. A node pool for user pods:
    • add metadata hub.jupyter.org/node-purpose: user to force Jupyterhub pods go in that node (affinity already specified for jupyterhub pods)
    • A no-schedule taint hub.jupyter.org/dedicated: user (toleration already configured for jupyter pods)
    • scaling enabled
  2. A node pool for "service", stable pods. Actual requirements are 4 CPUs and abou7 10GB memory, no scaling required
  3. Configure workflows for the same tolerations and affinity of jupyterhub pods. Workflows are required to go in the same node of jupyterhub pods in order to be able to write in the same volume.
  4. Configure Node containers with Docker. Something is preventing workflows to work with the default containeros, even changing executor. This issue needs further future investigation (tested alternative executors on other clusters and work)