docs: Add section in load balancing about Bokeh extensions #7832
Conversation
There was a problem hiding this comment.
Nice doc and very useful image. What software was used to create it?
Left some suggestions / comments inline, mainly adding some comma's and other interpunction, and suggestions for further clarification.
| Load balancing is the most complex approach to set up but is guaranteed to | ||
| improve concurrent usage of your application since different users are not | ||
| contending for access to the same process or even necessarily the same physical | ||
| compute and memory resources. At the same time it is more wasteful of resources |
There was a problem hiding this comment.
At the same time it can be more resource-intensive. Since each process is isolated, there is no sharing of cached data or global state between processes.
Note: are there potential solutions to isolated cache / global state? Like a cache at a higher level that is accessible by each process?
If so, provide some high-level mention or hint to these solutions?
There was a problem hiding this comment.
Removing multiple machines as part of the sentence is incorrect to me, mainly because a way to share resources if you were on the same machine would be to save them to disk.
I don't think there is a high-level solution. It will depend on the use case. One acceptable solution would be to accept it is isolated; another would be to have a server used to communicate with it or set your system up to have file access.
People who need to use load balancing will likely already know how to handle it; if not, there will be better resources for this than this page.
There was a problem hiding this comment.
Agree, my suggestion (without the multiple machine part) was an incomplete paste from the original.
I meant to suggest keeping the paragraph the same, but using "resource-intensive" instead of "wasteful of resources".
| compute and memory resources. At the same time it is more wasteful of resources | ||
| since it potentially occupies multiple machines and since each process is | ||
| isolated there is no sharing of cached data or global state. | ||
|
|
There was a problem hiding this comment.
Should there be a mention that Tornado is used by default, and that Django, Flask and FastAPI are also options? I am not sure if and how using one of these four (or more)? options relates to concurrency / load balancing, but should this at least be mentioned, with links to the Panel docs for use of Flask and FastAPI?
There was a problem hiding this comment.
I would not mention it, as it has nothing to do with load balancing.
|
Can you do the suggestions as code review ±? The image itself is the image from nginx. Have just changed it to our server, because the original url no longer exists. |
|
Yes I went for the comment option by mistake. Can do it as code review but not before this weekend. If a release is imminent then I will do it during the next release. |
|
Would be good to add a section on / mention of https://panel.holoviz.org/how_to/profiling/profile.html , so users know there are ways to profile an application for memory use etc. So they can use this to plan / monitor when implementing load balancing. Like the example here: https://discord.com/channels/1075331058024861767/1088149441988526241/1345112448474288210 . And link to https://panel.holoviz.org/how_to/profiling/admin.html |
|
Thank you for the review 👍 I have applied most of the changes.
Good idea, maybe have a |
There was a problem hiding this comment.
Agree, my sugestion was incomplete. I meant to suggest keeping the paragraph the same, but using "resource-intensive" instead of "wasteful of resources".
| Load balancing is the most complex approach to set up but is guaranteed to | ||
| improve concurrent usage of your application since different users are not | ||
| contending for access to the same process or even necessarily the same physical | ||
| compute and memory resources. At the same time it is more wasteful of resources |
There was a problem hiding this comment.
Agree, my suggestion (without the multiple machine part) was an incomplete paste from the original.
I meant to suggest keeping the paragraph the same, but using "resource-intensive" instead of "wasteful of resources".
| Load balancing is the most complex approach to set up but is guaranteed to | ||
| improve concurrent usage of your application since different users are not | ||
| contending for access to the same process or even necessarily the same physical | ||
| compute and memory resources. At the same time it is more wasteful of resources |
There was a problem hiding this comment.
| compute and memory resources. At the same time it is more wasteful of resources | |
| compute and memory resources. At the same time it is more resource-intensive |
There was a problem hiding this comment.
I ran the files locally pretty successfully with docker, the only change I needed was to remove localhost/ from docker run -dt -p 8000:80 localhost/my-container. The app was then available at http://localhost:8000/proxy/app.
However, I don't manage to understand what this is really fixing. Is it that, in a load-balancing context, serving additional Bokeh extensions like ipywidgets_bokeh can only be done reliably with BOKEH_RESOURCES=server? Or are there other specific settings make in the Docker or NGINX files that are worth highlighting?
If you want to serve a custom Bokeh model, in the example it uses |
|
This pull request has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs. |
Fixes #4074 by adding an example of how to use it with custom extensions.