-
Notifications
You must be signed in to change notification settings - Fork 37
start_timeout and http_timeout ignored #20
Comments
I've noticed the same issue and it's pretty painful when trying to start a notebook on a separate worker node. I get a 500 error almost every time and reloading the hub via browser tends to be error prone (too many redirects, auth token expired, etc) |
Does the remote worker have the image locally available? |
@barrachri On mine, it does not when first launching. That's when I need a very large timeout as it takes a long time to download from my private repository from a separate datacenter. This problem compounds when considering I have multiple worker nodes each having to go through this initialization. Setting Would be good to also implement the I wonder if this is because we are going against JupyterHub 0.7.2 and it's not yet implemented until 0.8.0? |
I don't know about Can you try to spawn the notebook with pulling the image before? |
I got authentication error when trying to get the image on the fly, so in my case the image is already there. How I got this timeout error is by spawning a lot of notebooks at the same time (around 6-7). Some of them will take longer than 10 seconds to spawn, hence the timeout error. |
What seems to be happening for me is that if the single-user notebook isn't already on the worker node, it takes a long time for it to pull that image and start up. That's expected. So it seems like when SwarmSpawner tries to start the service, the worker has to download the image but the browser client will get an error 500 after 20 seconds once logged in:
Note that this is a run with the following configs set:
Those seem to at least make a difference in SwarmSpawner not dying after a few seconds of not getting the service up and running. But the browser client still ends up with:
Expected behavior is for SwarmSpawner to report back that the service is starting. |
I'm seeing the same occurrence as above. |
Actually, it seems like after I set up http_timeout = 300, I will eventually get a 'Server never showed up at XXX', just only after the extended duration of time that I had set. It seems like my Hub is unable to communicate with the service it creates, because once I create it, the service appears to be running healthy. |
I had similar issues that were somewhat improved by increasing |
Let me post my config and logs In a moment. |
Okay, so here are my logs and config file. Jupyterhub output:
Here is my Jupyterhub Config. I'm relatively new to jupyterhub, and quite new to jupyterhub on swarm mode /w docker, so I may be missing something that may seem simple to others. Configuration file for JupyterHub c = get_config() c.JupyterHub.ssl_cert = '/srv/jupyterhub/jupyter.hub.crt' c.JupyterHub.ip = '0.0.0.0' |
I have set up
c.SwarmSpawner.start_timeout=60
andc.SwarmSpawner.http_timeout=60
injupyterhub_config.py
, but it keeps timing out at 10 s. I could not see in the code that it overwrites the value fromSpawner
class. Is it not yet implemented?The text was updated successfully, but these errors were encountered: