dask / dask/distributed

Simplify hosing of Dask Scheduler Dashboard behind the proxy

Open
#8,957 0 comments 1 reaction 0 assignees View on GitHub
needs triage
Dominant language
Python
Stars
1.7k
Forks
778
Avg merge
2h 50m
Merged PRs (30d)
3

Description

I'm trying to get Dask Scheduler running behind a custom reverse proxy.
The proxy does URL rewriting from `/session/{SID}/ports/{PID}/*` to just `/*`.

The dask scheduler dashboard is executed inside a Kubernetes pod.

In this setup, Dask Scheduler Dashboard doesn't really easily work out of the box.

Here is a couple of issues I have faced so far.

## 1. Absolute redirect to the root path

If you just try to open `/session/{SID}/ports/{PID}/`, the dashboard does a redirect to `/status` root path which omits the proxy subpath and breaks the target URL redirecting to `host.com/status`.

## 2. Bokeh assets are broken

The very next issue when you manually try to manually open the correct status URL e.g. `/session/{SID}/ports/{PID}/status` is all bokeh assets are broken because they are requested with absolute URL e.g. `/static/js/bokeh-gl.min.js?v=ver`:

Screenshot 2024-12-05 at 13 30 31

Interestingly, if you run it locally, the path will be relative e.g. `static/js/bokeh-gl.min.js?v=ver`.

It's possible to work this around by specifying this environment variable: `export BOKEH_RESOURCES="cdn"` that would effectively try to pull bokeh assets not from the current scheduler server but CDN.

## 3. Incorrect relative paths for sortable.js

The third problem that I have noticed is wrong relative paths for some assets on some pages. For example, when I open `/session/{SID}/ports/{PID}/info/main/workers.html` page, it tries to load `../../../statics/js/sortable.min.js` that assumes three levels of nesting:

Screenshot 2024-12-05 at 13 39 44
 
Apparently, there should have been two levels navigation up e.g. `../..` instead of three to reach the root because out full path looks like:

```text
info/main/workers.html
2 1 we are here^
```

Three level navigation e.g. `../../..` works okay when you host the Dask Dashboard under the hostname e.g. `host.com/info/main/workers.html` because you simply cannot leave the root dir, so browser/server correctly handles this case.
However, when you host your app under subpath e.g. `/session/{SID}/ports/{PID}/info/main/workers.html` it ends up incorrectly navigating to `/session/{SID}/ports/statics/js/sortable.min.js` instead of `/session/{SID}/ports/{PID}/statics/js/sortable.min.js`.

## Summary

Given all these issues with route building when the Dask Dashboard is hosted under the subpath, could we possibly make it less challenging and easier to get it work in this setup? 🙏

The first issue could be theoretically worked around by rewriting the response redirect header. The workaround for the second issue has been mentioned. As for the third one, there is as much as you can do because it's a part of the generated HTML and broken on the browser level. You could fix this by modifying the underlying content of the page but that's a lot do and support if the proxy is not Dask-specific.

### Other Context

I'm installing Dask like this:

```sh
pip install "dask[distributed]" bokeh
```

Then I'm running the dashboard in one of these way:

```python
from dask.distributed import Client

client = Client()
```

```sh
dask-scheduler --host 0.0.0.0 --dashboard-address 8787
```

The Dask version that I used for testing was `2023.11.0`.

There is no Dask or Bokeh specific env vars in my environment.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.