Open gkossakowski opened 4 years ago
@gkossakowski If you have a notebook reproducing this (using public Maven / Ivy repositories), don't hesitate to post it here, that would help fixing that.
This is the notebook I was working on: https://github.com/gkossakowski/covid-19/blob/36c9e8e5cfc75d3286e6f993595fe0aa854ed52c/Rainier.ipynb
Are you seeing this only in a transient way? Your notebook ran fine locally for me (and it seems your last run went fine too, as the cell with a .par
is done running there too).
The frequent display updates when running the .par
cell worry me a bit. If the jupyter client in the browser doesn't accept messages fast enough, the queues buffering them can fill up, which should then block clients trying to publish down-the-line, so that kind of call can block. Could this be what you're seeing?
If that's the case, maybe disabling progress can help, by importing com.stripe.rainier.notebook._
like
import com.stripe.rainier.notebook.{progress => _, _}
I'm running Rainier's model fitting on multiple threads via Scala's parallel collections. I noticed the kernel got stuck burning a CPU indefinitely and lunched VisualVM to see what's going on. In particular, I was interested in whether Rainier went into an infinite loop but it looks like it's a problem with Almond.
Here's what sampler is showing me:
This suggests some race condition in accessing Almond's output reporter.
The thread dump is: