"Stream is closed" error when creating dask client in Python 2 about dask-yarn HOT 7 CLOSED

My guess is that this is due to mismatched versions between the scheduler and client. Can you verify that your client process has the same version of distributed that is in the environment that you're giving to the YarnCluster?

from dask-yarn.

I just checked the distributed versions on client and environment sides and they appear to be the same

pip freeze | grep distributed
distributed==1.23.3

As a matter of fact I made the anaconda environment tar.gz file few days ago so it's fresh. I also checked the tornado and skein versions and they are also the same.

from dask-yarn.

Update. I tried the same code under Python 3 and It works great. The scenario I tried initially was as follows: I installed Anaconda2 (python 2), created environment for YARN from this anaconda, called the client from anaconda's python 2 and got the error I described above. Repeating the same steps but with Anaconda 3 doens't produce any errors. So the problem is with Python 2.
As it's indicated in the scheduler log the dask on yarn uses Python 3 (environment/lib/python3.6) even though the environment was created and packed into tar.gz file under Python 2.

from dask-yarn.

Ah, that's it then. This has nothing to do with dask-yarn, this is general across all of Dask: you can't use Python 3 for the scheduler when you're using Python 2 for the client. Versions of all packages (including Python) should be the same across the scheduler, client, and workers.

As it's indicated in the scheduler log the dask on yarn uses Python 3 (environment/lib/python3.6) even though the environment was created and packed into tar.gz file under Python 2.

An environment created with conda and conda-pack is completely self contained, I don't see how an environment that packaged Python 2 could be running Python 3 - I suspect the environment you packaged is not the same as the one you're running locally. Is there a possibility that you may have created or packaged the environment incorrectly?

from dask-yarn.

Well, I think the environment was packed correctly in some sense. I just discovered that anaconda2 creates environment with Python 3 being the default version even though Python 2 seems to be more logical default version created from anaconda 2.
At the end of the day mystery is solved and the issue can be closed.

from dask-yarn.

Glad to hear you figured things out.

from dask-yarn.

Is it possible to install dask-yarn on python2?
SergeySmith said he did it.

from dask-yarn.