I am having large team using Jupyterhub deployed in Kubernetes because proxy and hub can become bottleneck in case of huge request load. I am trying to increase its replica from deployment which is hardcoded 1 in deployment.
Does Jupyterhub support this? because in testing i am facing inconsistency. Sometimes notebook is spawning and sometime it is giving internal error and many other types of error.
We are using helm chart version 2.0.0 because our kubernetes version is 1.20.
Support HA (High Availability) · Issue #1932 · jupyterhub/jupyterhub · GitHub is where this is discussed. The short answer is:
- The Hub cannot have multiple replicas due to its database implementation, and
- the proxy can have multiple replicas if you use jupyterhub-traefik-proxy (or configurable-http-proxy with redis storage backend)
However, we still haven’t made jupyterhub-traefik-proxy work in z2jh, so if you’re using the helm chart, both must be single replicas for now.
What sort of load do you have that’s causing a problem?
Is there any example to use traefik( with etcd) with jupyterhub . I can’t find much on this. No proper article or blog
We have followed this and able to run jupyterhub with traefik and etcd in my system. However, when deploying it in kubernetes cluster with multiple nodes. We are facing 2 issues
- Jupyterhub is not creating routes in etcd.
- Not able to identify how will traefik forward request to jupyterhub. We can find no properties (in etcd, jupyterhub and traefik) to pass the jupyterhub IP and port.