Bokeh service does not start when using processes=False

The bokeh service does not start on Debian when using processes=False. Strangely, it does start on macOS.

Minimal reproducing example:

from distributed import LocalCluster, Client

cluster = LocalCluster(processes=False)
client = Client(cluster)
print(client.scheduler_info()))
client.close()

On macOS with distributed 1.22.0, the output is:

{
   "type": "Scheduler",
   "id": "Scheduler-275f860b-d0cd-46a8-865a-c75b8ebd6fc8",
   "address": "inproc://10.40.36.245/56255/1",
   "services": {
      "bokeh": 8787
   },
   "workers": {
      "inproc://10.40.36.245/56255/2": {
         "name": "inproc://10.40.36.245/56255/2",
         "memory_limit": 17179869184,
         "host": "10.40.36.245",
         "resources": {},
         "ncores": 8,
         "services": {
            "bokeh": 61024
         },
         "local_directory": "/Users/laurent/[...]/dask-worker-space/worker-bh9f2yjh",
         "pid": 56255,
         "cpu": 0,
         "memory": 100495360,
         "time": 1.53373248E9,
         "read_bytes": 0,
         "write_bytes": 0,
         "num_fds": 68
      }
   }
}

While running the same code on Debian (in a Docker container with no other processes running) results in the below output. Notice that on Debian, the "services" value is an empty dict! With processes=True, the bokeh service does start.

{
   "type": "Scheduler",
   "id": "Scheduler-fd83c67d-515a-499f-9c1c-fd9641d54d51",
   "address": "inproc://172.18.0.2/104/1",
   "services": {},
   "workers": {
      "inproc://172.18.0.2/104/2": {
         "name": "inproc://172.18.0.2/104/2",
         "memory_limit": 12575559680,
         "host": "172.18.0.2",
         "resources": {},
         "ncores": 6,
         "services": {
            "bokeh": 41839
         },
         "local_directory": "/[...]/dask-worker-space/worker-0yv06g56",
         "pid": 104,
         "cpu": 0,
         "memory": 112869376,
         "time": 1.53373286E9,
         "read_bytes": 0,
         "write_bytes": 0,
         "num_fds": 24
      }
   }
}

[weiweiWYW]

add the ‘ip’ params, e.g. client = Client(processes=False, threads_per_worker=4,ip=‘127.0.0.1’,n_workers=1, memory_limit=‘2GB’)

[dgakka]

Same problem here on a Windows 10, with dask, dask.distributed and bokeh installed: Setting processes=False on Client or LocalCluster, creates this ‘inproc’ unreachable endpoint.

I tried setting the ip as suggested by @weiweiWYW but there is no apparent ip parameter in the Dask.distributed Client API. Simply passing the address parameter to the Client,

client = Client(address="127.0.0.1:8787")

Raises a NotImplementedError at asyncio/events.py