Some busy indicators not firing #1381
Comments
|
I wonder if we need to get the I/O event loop and the session event loop onto separate threads, like we did in R. |
|
I followed the call to conn.write() with a debugger down to here: When the first branch is taken, the message goes right away. But in the repro case, that's only true for the first progress message; the second message goes to the second branch, and all the remaining ones go to the third branch, and none of those get sent until computation is complete. |
During a synchronous phase of the event loop, we can build up network messages that need to get to the client. Because writes are naturally async, we do some weird stuff to semi-execute them which usually but doesn't always succeed. In the case where the write doesn't succeed, they're held at least until the next time the current task yields; before this commit, that could be a long time indeed, if the app is written synchronously (as about 100% of Shiny apps are). This commit yields to the event loop after invalidating the session, which gives the writes a chance to complete. This should take care of the observed problem of progress messages not making it (issue #1381). It also yields after each successful @effect, in case insert_ui, custom messages, etc. were sent during the @effect, and also so outputs' automatic progress messages can propagate to the client.
|
I filed a PR that will fix this. Still doesn't feel awesome that our I/O is tangled up with the app logic like this, I have a feeling this will come back with |
With the examples/busy_indicators/app.py app, on Windows, I'm pretty reliably able to get only one out of the four plots to show progress.
https://jam.dev/c/fce934dd-d6de-4514-9ae5-184bbc74684d
I can see from the Python console that Shiny is sending the right messages, but some of them are getting batched up, similar to #1373, except in this case uvloop is not installed. Also, if I use
--reloadit doesn't happen.The text was updated successfully, but these errors were encountered: