-
-
Notifications
You must be signed in to change notification settings - Fork 719
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Issues starting workers #2506
Comments
It appears the issue is with distributed 1.25.3 as mentioned in #2504 |
Interesting. Thanks for reporting this @bpmweel . Unfortunately nothing jumps out at me as an obvious cause of this. From the issue that you refer to it sounds like this might have been a recent change. Do you have any interest in using |
I did a git bisect and it seems it originates already in this commit: 70c5129 |
cc @danpf |
Thanks for going through that process @bpmweel |
darn... your examples appear to work on my local machine. I'm looking into this! |
I've tried python 3.[5,6,7] pip, and conda and i'm unable to recreate this :/ Could you try upping this number to something like 3, 6, or 10? |
Sorry. I’ve been at a training last week. I will try to get to it on Monday
|
I tried upping the threadpoolexecuter count, but that did not fix anything. I tried wrapping the client creation in a init function and this does help: 324,330c324,332
< if PY3: # see github PR #2403 discussion for more info
< _executor = ThreadPoolExecutor(2)
< _resolver = netutil.ExecutorResolver(close_executor=False,
< executor=_executor)
< else:
< _resolver = None
< client = TCPClient(resolver=_resolver)
---
> def __init__(self):
> if not hasattr(BaseTCPConnector, 'client'):
> if PY3: # see github PR #2403 discussion for more info
> _executor = ThreadPoolExecutor(2)
> _resolver = netutil.ExecutorResolver(close_executor=False,
> executor=_executor)
> else:
> _resolver = None
> BaseTCPConnector.client = TCPClient(resolver=_resolver) I'm not savvy enough in python to know if this is very fundamentally different from creating a class attribute in the previous manner. |
@danpf any thoughts on the above? |
This would overwrite the class attribute each time we make a new object which probably isn't a good idea. This is what happens with suggested
This is in tornado 4.5.1 with the original (before my commit) method
I don't see theese errors in tornado 5 thoug @bpmweel do you have any dependency that requires tornado <5? could you try explicitly installing a higher version of tornado? I don't see any problems when I use tornado 5.0.0 you can check this by doing
@mrocklin |
Shouldn't the That being said, I don't have any particular tornado dependencies, so I will try version 5 and report back. |
oh you're right I missed that line when I copied it : p sorry! your fix seems to work for ipython, but not for cli distributed, but i have 0 idea why. the client should be resolved when the class is compiled/parsed so.... (this is tornado 4.5.1)
|
@bpmweel can you try updating to Tornado 5.0 or above and see if that resolves the problem? |
Today Tornado 5 is the minimum version supported in |
I'm trying to start a dask cluster on my local machine. I can use the LocalCluster just fine, but starting the scheduler and client from the command line I run into issues:
I have the feeling this has to do with having multiple interfaces and the nanny process. The following combination works on the command line:
However, now I cannot connect to the scheduler from ipython:
The text was updated successfully, but these errors were encountered: