fix hostname of cobald slurm node

made cobald be able to run slurm jobs, previously failing with
permission denied.

play.yml

@@ -81,7 +81,18 @@
         dir: /var/lib/cobald
       num_nodes: "{{slurm_num_nodes}}"
       extra_nodes:
-      - "{{cobald_container_hostname}}" # from cobald_facts, read or generated
+      - name: cobald
+        hostname: "{{cobald_container_hostname}}" # from cobald/facts.yml above
+        # hostname is used as NodeHostname, which is used slurms "networking
+        # code" (https://bugs.schedmd.com/show_bug.cgi?id=8615).
+        # It works either way around, but one of NodeName or NodeHostname has
+        # to match the container name (-n flag, not --hostname) since when
+        # submitting tasks to the slurm controller, it matches access
+        # permissions against a reverse lookup of the submitting ip address.
+        # Docker always and unconfigureably resolves the container ip in any
+        # network to containername.netname, where containername is the
+        # containers runtime name (not hostname supplied!) and netname is
+        # the network name in host environment. We should run our own dns...
       docker_network: slurm
     when: '"slurm" in group_names'
     tags: slurm