html_url,issue_url,id,node_id,user,created_at,updated_at,author_association,body,reactions,performed_via_github_app,issue
https://github.com/pydata/xarray/issues/3781#issuecomment-776160305,https://api.github.com/repos/pydata/xarray/issues/3781,776160305,MDEyOklzc3VlQ29tbWVudDc3NjE2MDMwNQ==,885575,2021-02-09T18:51:13Z,2021-02-09T18:51:13Z,NONE,"@lvankampenhout, I ran into your problem. OP's seems like it's actually in `to_netcdf()`, but I think yours (ours) is in Dask's lazy loading and therefore unrelated. 

In short, `ds` will have some Dask arrays whose contents don't actually get loaded until you call `to_netcdf()`. By default, Dask loads in parallel, and the default Dask parallel scheduler chokes when you do your own parallelism on top. In my case, I was able to get around it by doing 

```python
ds.load(scheduler='sync')
```

at some point. If it's outside `do_work()`, I think you can skip the `scheduler='sync'` part, but inside `do_work()`, it's required. This bypasses the parallelism in Dask, which is probably what you want if you're doing your own parallelism.","{""total_count"": 2, ""+1"": 2, ""-1"": 0, ""laugh"": 0, ""hooray"": 0, ""confused"": 0, ""heart"": 0, ""rocket"": 0, ""eyes"": 0}",,567678992