new startup for cobald containers

inv.yml

@@ -19,6 +19,7 @@ all:
     slurm:
       hosts:
         ed-c7-1:
+        ed-c7-2:
     cobald:
       hosts:
         ed-c7-1:

play.yml

@@ -53,10 +53,10 @@
       - htcondor-containered
       - htcondor
 
-- hosts: slurm
+- hosts: slurm, cobald
   vars:
     container_privileged: True
-    num_nodes: 3
+    slurm_num_nodes: 10
   roles:
   - name: "setup docker"
     role: docker
@@ -65,15 +65,12 @@
     role: slurm
     vars:
       slurm_user: slurm # or root
+      num_nodes: "{{slurm_num_nodes}}"
+    when: '"slurm" in group_names'
     tags: slurm
-
-- hosts: cobald
-  roles:
-  - name: "setup docker"
-    role: docker
-    tags: docker
   - name: "install cobald"
     role: cobald
     vars:
       docker_network: slurm
+    when: '"cobald" in group_names'
     tags: cobald

roles/slurm/files/docker-init

@@ -0,0 +1,18 @@
+#!/bin/bash
+function trp_term(){
+        echo pkill -P $pids
+        for j in $pids ; do
+                pkill -P $j
+                kill -SIGTERM $j
+        done
+}
+trap trp_term SIGINT SIGTERM
+pids=""
+for i in /etc/docker-init.d/* ; do
+        [ ! -f $i ] && break
+        $i &
+        pids="$pids $!"
+done
+wait $pids
+
+# TODO: call start scripts like "foo.sh start" and "foo.sh stop" to avoid pkill

roles/slurm/files/slurmctld.Dockerfile

@@ -1,4 +1,4 @@
-FROM docker.io/library/centos:7 as base
+FROM docker.io/library/centos:7
 
 RUN yum install -y epel-release && \
 	yum install -y slurm && \
@@ -7,6 +7,7 @@ RUN yum install -y epel-release && \
 RUN	yum install -y less iproute bind-utils nmap-ncat net-tools && \
 	yum clean all && rm -rf /var/cache/yum
 
+# FIXME
 COPY entrypoint.sh /usr/local/sbin/entrypoint.sh
 
 RUN chown root:root /usr/local/sbin/entrypoint.sh && \
@@ -23,10 +24,15 @@ RUN useradd -d /var/lib/slurm -m --no-log-init --system $slurmuser &&\
 RUN	yum install -y slurm-slurmctld && \
 	yum clean all && rm -rf /var/cache/yum
 
+COPY docker-init /usr/local/sbin/docker-init
+RUN mkdir /etc/docker-init.d  && chmod 755 /usr/local/sbin/docker-init
+# FIXME
+COPY start-scripts/10-munge /etc/docker-init.d/
+COPY start-scripts/20-slurmctld /etc/docker-init.d/
+RUN chmod 755 /etc/docker-init.d/*
+
 ENV SLURMCTLD_LOG_PATH="/var/log/slurm/slurmctld.log"
 ENV SLURMD_LOG_PATH="/var/log/slurm/slurmd.log"
 ENV SLURM_SCHED_LOG_PATH="/var/log/slurm/slurmsched.log"
 
-CMD bash -c 'cat <({ su -s /bin/sh -c "munged -F" munge & \
+CMD /usr/local/sbin/docker-init
-	su -s /bin/sh -c "slurmctld -D" ${slurmuser} 2>/dev/null 1>/dev/null & \
-	tail --retry --pid $! -f ${SLURMCTLD_LOG_PATH} ${SLURM_SCHED_LOG_PATH} & })'

roles/slurm/files/slurmd.Dockerfile

@@ -7,6 +7,7 @@ RUN yum install -y epel-release && \
 RUN	yum install -y less iproute bind-utils nmap-ncat net-tools && \
 	yum clean all && rm -rf /var/cache/yum
 
+# FIXME
 COPY entrypoint.sh /usr/local/sbin/entrypoint.sh
 
 RUN chown root:root /usr/local/sbin/entrypoint.sh && \
@@ -23,10 +24,15 @@ RUN useradd -d /var/lib/slurm -m --no-log-init --system $slurmuser &&\
 RUN yum install -y slurm-slurmd && \
 	yum clean all && rm -rf /var/cache/yum
 
+COPY docker-init /usr/local/sbin/docker-init
+RUN mkdir /etc/docker-init.d  && chmod 755 /usr/local/sbin/docker-init
+# FIXME
+COPY start-scripts/10-munge /etc/docker-init.d/
+COPY start-scripts/30-slurmd /etc/docker-init.d/
+RUN chmod 755 /etc/docker-init.d/*
+
 ENV SLURMCTLD_LOG_PATH="/var/log/slurm/slurmctld.log"
 ENV SLURMD_LOG_PATH="/var/log/slurm/slurmd.log"
 ENV SLURM_SCHED_LOG_PATH="/var/log/slurm/slurmsched.log"
 
-CMD bash -c 'cat <({ su -s /bin/sh -c "munged -F" munge & \
+CMD /usr/local/sbin/docker-init
-	slurmd -D 2>/dev/null 1>/dev/null & \
-	tail --retry --pid $! -f ${SLURMD_LOG_PATH} ${SLURM_SCHED_LOG_PATH} & })'

roles/slurm/files/start-scripts/10-munge

@@ -0,0 +1,2 @@
+#!/bin/sh
+exec su -s /bin/sh -c "munged -F" munge

roles/slurm/files/start-scripts/20-slurmctld

@@ -0,0 +1,4 @@
+#!/bin/sh
+su -s /bin/sh -c "slurmctld -D" ${slurmuser} 2>/dev/null 1>/dev/null &
+tail --retry --pid $! -f ${SLURMCTLD_LOG_PATH} ${SLURM_SCHED_LOG_PATH}
+

roles/slurm/files/start-scripts/30-slurmd

@@ -0,0 +1,4 @@
+#!/bin/sh
+slurmd -D 2>/dev/null 1>/dev/null &
+tail --retry --pid $! -f ${SLURMD_LOG_PATH} ${SLURM_SCHED_LOG_PATH}
+

roles/slurm/tasks/dockerimage.yml

@@ -1,31 +1,50 @@
 - file:
-    path: "/container/docker-images/{{item}}"
+    path: "/container/docker-images/slurm"
     state: directory
     owner: "{{unpriv_user}}"
     group: docker
 
-- copy:
+- copy: # FIXME: template
-    src: "{{item}}.Dockerfile"
+    src: "{{image.name}}.Dockerfile"
-    dest: "/container/docker-images/{{item}}/Dockerfile"
+    dest: "/container/docker-images/slurm/{{image.name}}.Dockerfile"
     owner: "{{unpriv_user}}"
     group: docker
   register: slurm_cp_dockerfile
 
-- copy:
+- name: copy entrypoint and docker-init
-    src: "entrypoint.sh"
+  copy: # FIXME: swap out
-    dest: "/container/docker-images/{{item}}/entrypoint.sh"
+    src: "{{item}}"
+    dest: "/container/docker-images/slurm/{{item}}"
     owner: root
     group: root
     mode: u=rwx,g=rx,o=rx
+  loop:
+  - entrypoint.sh
+  - docker-init
   register: slurm_cp_entrypt
 
+- name: copy startup scripts
+  copy:
+    src: "start-scripts/"
+    dest: "/container/docker-images/slurm/start-scripts/"
+    owner: root
+    group: root
+    mode: u=rwx,g=rx,o=rx
+  register: slurm_cp_stscrs
+
+- set_fact:
+    slurm_image_files_changed: "{{ (slurm_image_files_changed | default(False))
+      or slurm_cp_entrypt.changed or slurm_cp_stscrs.changed }}"
+
 - docker_image:
-    name: "slurm-{{item}}"
+    name: "slurm"
+    tag: "{{image.name}}"
 #    pull: False
     build:
       pull: False
-      path: "/container/docker-images/{{item}}"
+      path: "/container/docker-images/slurm/"
-#      target: "{{item}}" # unsupported on old docker-py versions as in el7 
+      dockerfile: "{{image.name}}.Dockerfile"
+#      target: "{{image.name}}" # unsupported on old docker-py version as in el7
     source: build
-    force_source: "{{slurm_cp_dockerfile.changed or slurm_cp_entrypt.changed}}"
+    force_source: "{{slurm_cp_dockerfile.changed or slurm_image_files_changed}}"
 

roles/slurm/tasks/main.yml

@@ -5,8 +5,17 @@
 
 - include_tasks: dockerimage.yml
   loop:
-    - slurmctld
+  - name: slurmctld
-    - slurmd
+    start:
+    - 10-munge
+    - 20-slurmctld
+  - name: slurmd
+    start:
+    - 10-munge
+    - 30-slurmd
+  loop_control:
+    loop_var: image
+    label: "{{ image.name }}"
 
 - name: generate munge key
   shell:
@@ -63,13 +72,14 @@
 - set_fact:
     slurm_nodes: # default nodes: controller and submit machine
     - machine: ctl
-      image: slurm-slurmctld
+      image: slurm:slurmctld
+      exposed_ports: [ "6817:6817/tcp" ]
     - machine: submit1
-      image: slurm-slurmd
+      image: slurm:slurmd
       extra_mounts:
       - "/home/{{unpriv_user}}/job3/:/mnt/:rw"
   tags: [ slurm-config ]
-    
+
 # TODO: reserve some address using docker_network_info and assign as aux
 # address to enable slurmctld to get a static address in order to be
 # reachable from slurm running on docker host to enable submitting jobs.
@@ -80,6 +90,7 @@
     hostname: "slurm-{{item.machine}}"
     domainname: "slurm.local"
     volumes: "{{default_mounts + ( item.extra_mounts | default([]) ) }}"
+    ports: "{{ item.exposed_ports | default([]) }}"
     networks:
     - name: "slurm"
     env:
@@ -90,15 +101,16 @@
     cleanup: True
     privileged: "{{ container_privileged | bool }}"
     networks_cli_compatible: True
+    interactive: True
   vars:
     default_mounts:
     - /container/volumes/slurm/:/etc/slurm/:rw
     - /container/volumes/munge/munge.key:/etc/munge/munge.key:rw
     - slurm-shared:/shared/:rw
     slurm_nodes_all: | # add execute nodes
-      {% for i in range(1, 4) -%}
+      {% for i in range(1, num_nodes+1) -%}
         {% set _ = slurm_nodes.extend([
-          {'machine':'exec%s'|format(i), 'image': 'slurm-slurmd'}]) -%}
+          {'machine':'exec%s'|format(i), 'image': 'slurm:slurmd'}]) -%}
       {%- endfor %}
       {{ slurm_nodes }}
   loop: "{{slurm_nodes_all}}"