Replace stream= with comm= in handlers - #3860
Conversation
|
Thanks for doing this @jjerphan . What do you think about things like: distributed/distributed/comm/tcp.py Line 95 in 8480b4f distributed/distributed/comm/tcp.py Line 36 in 8480b4f I found those by doing:
|
| stream.synchronize() | ||
| cu_stream = numba.cuda.driver.drvapi.cu_stream(comm) | ||
| comm = numba.cuda.driver.Stream(ctx, cu_stream, None) | ||
| comm.synchronize() |
There was a problem hiding this comment.
This is actually a different kind of stream (CUDA streams rather than Dask streams) and we should probably leave it as is.
| if not comm.closed(): | ||
| logger.warning("Closing dangling stream in %s" % (r,)) | ||
| stream.close() | ||
| comm.close() |
There was a problem hiding this comment.
This is actually a different kind of stream (Torando IOStream rather than Dask stream) and we should probably leave it as-is.
|
Hum, I am trying to understand the failing test on the Travis CI build: it does not fail for me (it is skipped when running it locally). |
|
Yeah, that isn't your fault. I don't know what happened, but that test has become flaky recently, and only on Python 3.6. I'm going to go ahead and skip it for now. It's a very peripheral feature. See #3888 |
|
My apologies for the testing uncleanliness . Thank you for your patience with it. |
|
I'm merging this in. Thank you for your work here @jjerphan ! Also, I notice that this is your first code contribution to this repository. Welcome! |
It really is fine: don't worry. :) |
|
In fact, I once studied a problem between in the |
Fixes #3824.