Async docker launch

Wrapped the docker launch in an async function. This should fix problems with the API hanging on docker startup, but also implies that the API does properly handle and report exceptions happening during the docker launch.

north/app/routes/instances.py

@@ -17,6 +17,7 @@
 #
 
 from typing import List
+import asyncio
 from fastapi import APIRouter, HTTPException, Request, Depends
 from collections import deque
 from datetime import datetime, timedelta
@@ -112,18 +113,24 @@ async def post_instances(request: Request, instance: InstanceModel, token=Depend
         path = current_containers[0].labels['path']
         return InstanceResponseModel(path=path)
 
-    docker_client.containers.run(
-        image="jupyter/datascience-notebook",
-        command=(
-            f'start-notebook.sh --NotebookApp.base_url={path}'
-            ' --NotebookApp.token="" --NotebookApp.password=""'
-        ),
-        ports={"8888": int(f'1000{channel}')},
-        detach=True,
-        name=container_name,
-        user="1000:1000",
-        group_add=["1000"],
-        labels={"path": path}
-    )
+    # We use an async function to run the container so that the API does not
+    # get blocked even if docker needs to do some heavier container startup
+    # routines (e.g. download the container image).
+    async def run_container():
+        docker_client.containers.run(
+            image="jupyter/datascience-notebook",
+            command=(
+                f'start-notebook.sh --NotebookApp.base_url={path}'
+                ' --NotebookApp.token="" --NotebookApp.password=""'
+            ),
+            ports={"8888": int(f'1000{channel}')},
+            detach=True,
+            name=container_name,
+            user="1000:1000",
+            group_add=["1000"],
+            labels={"path": path}
+        )
+    loop = asyncio.get_event_loop()
+    loop.create_task(run_container())
 
     return InstanceResponseModel(path=path)