Getting the CI to pass consistently on Windows (hopefully) #4624
Conversation
|
All the demos for this PR have been deployed at https://huggingface.co/spaces/gradio-pr-deploys/pr-4624-all-demos |
| while not self.started: | ||
| time.sleep(1e-3) | ||
| if time.time() - start > 5: |
There was a problem hiding this comment.
It takes <10 ms on my computer to start the server. So 5 seconds is a very generous allotment for slow computers.
There was a problem hiding this comment.
nit: Might be good to add "to start within 5 seconds." for more context. Will push this up since you're away
There was a problem hiding this comment.
Actually this message is not displayed to the user so won't make the change
|
|
||
| for port in server_ports: | ||
| try: | ||
| # The fastest way to check if a port is available is to try to bind to it with socket. |
There was a problem hiding this comment.
Sorry for the copious comments, but I wanted to make sure the new logic is clear to someone who is reading the code for the first time.
abidlabs
changed the title
| server_name = server_name or LOCALHOST_NAME | ||
| # if port is not specified, search for first available port | ||
| if server_port is None: | ||
| port = get_first_available_port( |
There was a problem hiding this comment.
I haven't removed this method get_first_available_port from the code since we use it in reload mode. I do think we need to overhaul reload mode at some point, and then we could remove this method too
|
All the demos for this PR have been deployed at https://huggingface.co/spaces/gradio-pr-deploys/pr-4624-all-demos |
We had an issue where the backend CI tests on Windows would sometimes time out, maybe 40% of the time. Upon investigation, the issue was that one of the tests would try to run a Gradio demo on a port that was not available:
This shouldn't have been possible, because we first check to find an available port
get_first_available_portmethod. However, it was possible that a port could become occupied between the time this method is run and the Gradio server launches. In that case, the server launch would silently fail and the script would hang forever.What this PR does is:
This PR is a little tricky to test besides re-running the CI, which I've done several times and it seems like everything passes consistently now.
-- also feel free to merge to this branch directory, or merge this PR in if ready, as I'll be OOO. Thanks!