tags

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

play.yml

@@ -65,15 +65,20 @@
     tags: docker
 
   - name: "get facts from existing cobald instance (i.e. hostname)"
-    import_role:
+    include_role:
       name: cobald
       tasks_from: facts
+      apply:
+        tags: slurm, cobald, slurm-config
+    tags: slurm, cobald, slurm-config
     vars:
       container_name: cobald
-    tags: [ slurm, cobald ]
 
   - name: "setup slurm test environment in docker containers"
-    import_role: name=slurm
+    include_role:
+      name: slurm
+      apply:
+        tags: slurm
     vars:
       slurm_user: slurm # or root
       slurm_user_accounts:
@@ -81,16 +86,30 @@
         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
+    tags: slurm, slurm-config
 
   - name: "install cobald"
-    import_role: name=cobald
+    include_role:
+      name: cobald
+      apply:
+        tags: cobald
     vars:
       cobald_slurm: True
       container_name: cobald
       # docker_network: slurm # overriden by vars/slurm.yml
     when: '"cobald" in group_names'
-    tags: cobald
+    tags: cobald, influxdb

roles/cobald/tasks/main.yml

@@ -118,9 +118,6 @@
     networks:
     - name: "{{cobald_docker_network}}"
     networks_cli_compatible: True
-#    env:
-#      slurmuser: "{{slurm_user}}"
-#    privileged: "{{ container_privileged | bool }}"
     state: started
     detach: True
     cleanup: True

roles/cobald/vars/cobald-slurm.yml

@@ -1,5 +1,5 @@
 cobald_image_tag: slurm
-cobald_docker_base_image: "slurm:slurmd"
+cobald_docker_base_image: "{{slurm.base_image}}"
 cobald_docker_default_command: False
 cobald_docker_network: "{{slurm.network}}"
 cobald_domainname: "{{slurm.domain}}"

roles/slurm/tasks/main.yml

@@ -67,9 +67,11 @@
     - slurm.conf
     - cgroup.conf
   vars:
-    alloc_nodes: "{{ [ slurm_prefix+'-submit1' ] + extra_nodes | default([])}}"
+    slurm_alloc_nodes_default:
+    - name: "{{slurm_prefix+'-submit1'}}"
+    alloc_nodes: "{{ slurm_alloc_nodes_default + extra_nodes | default([])}}"
   notify: reconfigure slurm
-  tags: [ slurm-config ]
+  tags: slurm-config
 
 - name: "create docker network to make service discovery work"
   docker_network:
@@ -108,7 +110,7 @@
   loop: "{{ slurm_nodes_all }}"
   loop_control:
     label: "{{slurm_prefix}}-{{ item.machine }}"
-  tags: [ slurm-config ]
+  tags: slurm-config
 
 - name: export facts about slurm cluster to be used by other modules
   set_fact:

roles/slurm/templates/slurm.conf.j2

@@ -165,6 +165,8 @@ SlurmSchedLogFile={{slurm_log_path_sched}}
 # COMPUTE NODES
 NodeName=slurm-exec[1-{{num_nodes}}] CPUs=2 CoresPerSocket=2 State=UNKNOWN
 {% for i in alloc_nodes -%}
-NodeName={{i}} State=UNKNOWN
+NodeName={{i.name}}
+	{%- if i.hostname is defined %} NodeHostname={{i.hostname}} {% endif %}
+	State=UNKNOWN
 {% endfor %}
-PartitionName=debug Nodes=slurm-exec[1-{{num_nodes}}] AllocNodes={{alloc_nodes | join(',')}} Default=YES MaxTime=INFINITE State=UP
+PartitionName=debug Nodes=slurm-exec[1-{{num_nodes}}] AllocNodes={{alloc_nodes |map(attribute='name') | join(',')}} Default=YES MaxTime=INFINITE State=UP