fix: slurm host system access

base.yml

@@ -11,6 +11,14 @@
       name: [ vim-enhanced, htop, screen, bind-utils, nmap-ncat, net-tools ]
       state: present
 
+  - name: "screenrc native scrolling in tmux"
+    copy:
+      content: "termcapinfo xterm* ti@:te@\ntermcapinfo screen* ti@:te@\n"
+      dest: "{{item}}"
+    with_items:
+    - "~{{unpriv_user}}/.screenrc"
+    - "~root/.screenrc"
+
   - name: "install ssh-key"
     authorized_key:
       user: "{{cfg_unpriv_user}}"

cobald.yml

@@ -43,6 +43,7 @@
         # containers runtime name (not hostname supplied!) and netname is
         # the network name in host environment. We should run our own dns...
       docker_network: slurm
+      slurm_hostsystem_cluster_access: True
     when: '"slurm" in group_names'
     tags: slurm, cobald, influxdb, slurm-config
     # tags: cobald requires some slurm facts, so cobald tag is included here
@@ -57,4 +58,4 @@
       container_name: cobald
       # docker_network: slurm # overriden by vars/slurm.yml
     when: '"cobald" in group_names'
-    tags: cobald, influxdb
+    tags: cobald, influxdb, singularity

htcondor.yml

@@ -20,7 +20,7 @@
     tags: htcondor
 
   - name: "setup singularity"
-    import_tasks: "singularity.yml"
+    import_role: name="singularity"
     tags: singularity
 
   - name: "setup docker"

roles/cobald/files/28-sync-container-slurmd

@@ -0,0 +1,3 @@
+#!/bin/sh
+[ /slurm-singimage/slurmd.sif -nt /shared/slurmd.sif ] && \
+	cp /slurm-singimage/slurmd.sif /shared/slurmd.sif

roles/cobald/files/31-slurmd-configless

@@ -0,0 +1,3 @@
+#!/bin/sh
+slurmd --conf-server ${slurmctld} -D -N ${nodename} 2>/dev/null 1>/dev/null &
+tail --retry --pid $! -f ${SLURMD_LOG_PATH} ${SLURM_SCHED_LOG_PATH}

roles/cobald/files/cgroup.conf.noautomount

@@ -0,0 +1,11 @@
+###
+#
+# Slurm cgroup support configuration file
+#
+# See man slurm.conf and man cgroup.conf for further
+# information on cgroup configuration parameters
+#--
+CgroupAutomount=no
+
+ConstrainCores=no
+ConstrainRAMSpace=no

roles/cobald/files/cobald-config/config.yaml

@@ -0,0 +1,31 @@
+---
+pipeline:
+- __type__: cobald.controller.linear.LinearController
+  low_utilisation: 0.9
+  high_allocation: 0.9
+  rate: 0.10
+- !Limiter
+  minimum: 3
+- !TelegrafPipelineMonitor
+  poll: True
+- !TardisPoolFactory
+  configuration: /etc/cobald/tardis.yaml
+logging:
+  version: 1
+  root:
+    level: DEBUG
+    handlers: [console, file]
+  handlers:
+    console:
+      class: logging.StreamHandler
+      formatter: test
+      level: DEBUG
+      stream: ext://sys.stderr
+    file:
+      class: logging.handlers.RotatingFileHandler
+      formatter: test
+      level: WARNING
+      filename: /var/log/cobald/cobald-tardis.log
+  formatters:
+    test:
+      format: "  %(name)s %(message)s"

roles/cobald/files/cobald-config/tardis.yaml

@@ -0,0 +1,41 @@
+Plugins:
+  SqliteRegistry:
+    db_file: /tmp/drone_registry.db
+  TelegrafMonitoring:
+    host: ed-telegraf
+    port: 8094
+#BatchSystem:
+#  adapter: FakeBatchSystem
+#  allocation: 1.0
+#  utilisation: !PeriodicValue
+#    period: 60
+#    amplitude: 0.15
+#    offset: 0.80
+##    phase: 1.
+#    phase: 1.6
+#  machine_status: Available
+BatchSystem:
+  adapter: Slurm
+  max_age: 0.1
+  options:
+    partition: cobald
+Sites:
+- name: slurmtest
+  adapter: Slurm
+  quota: 20
+slurmtest:
+  # executor: ...
+  StatusUpdate: 0.1
+  MachineTypes:
+    - m1.a
+  MachineTypeConfiguration:
+    m1.a:
+      Walltime: 5
+      Partition: container
+      StartupCommand: /usr/local/bin/start-drone
+      # SubmitOptions: ...
+  MachineMetaData:
+    m1.a:
+      Cores: 3 # cores
+      Memory: 1 # GB
+      Disk: 4 # not passed

roles/cobald/files/slurm-slurmd.def

@@ -0,0 +1,31 @@
+Bootstrap: docker-daemon
+From: slurm:slurmd
+
+%files
+	31-slurmd-configless /etc/docker-init.d/31-slurm-configless
+	/container/volumes/munge/munge.key /etc/munge/munge.key
+	cgroup.conf.noautomount /etc/slurm/cgroup.conf
+
+%post
+	rm /etc/docker-init.d/30-slurmd
+	chmod 755 /etc/docker-init.d/31-slurm-configless
+
+%startscript
+	if [ -z "${1}" -o -z "${2}" ] ; then
+		echo "undefined variables slurmctld or nodename"
+		exit 1
+	fi
+	export slurmctld="${1}"
+	export nodename="${2}"
+	echo ${slurmctld} ${nodename} ${SLURMD_LOG_PATH} ${SLURM_SCHED_LOG_PATH}
+	exec /usr/local/sbin/entrypoint.sh /usr/local/sbin/docker-init
+
+%runscript
+	if [ -z "${1}" -o -z "${2}" ] ; then
+		echo "undefined variables slurmctld or nodename"
+		exit 1
+	fi
+	export slurmctld="${1}"
+	export nodename="${2}"
+	echo ${slurmctld} ${nodename} ${SLURMD_LOG_PATH} ${SLURM_SCHED_LOG_PATH}
+	exec /usr/local/sbin/entrypoint.sh /usr/local/sbin/docker-init

roles/cobald/files/start-drone

@@ -0,0 +1,59 @@
+#!/bin/sh
+#SBATCH -D /shared
+export
+echo $@
+nodename=$(hostname | awk '{ print "drone" substr($1,match($1, "([[:digit:]]+)")) }')
+
+SHUTDOWN_DONE=0
+
+function handler_quit(){
+	[ $SHUTDOWN_DONE -ne 0 ] && return
+	set -x
+	echo "drain container"
+	scontrol update NodeName=${nodename} State=DRAIN Reason="cobald node quit"
+	shutdown_jobs=$(squeue -w ${nodename} --noheader -O jobid)
+	[ -n "${shutdown_jobs}" ] && scancel ${shutdown_jobs}
+	#scancel -w ${nodename}
+	i=$(( $(scontrol show config | grep KillWait | \
+			sed 's/^KillWait.*= \([0-9]*\) sec/\1/') - 2 ))
+	while [ -n "$(squeue -w ${nodename} --noheader -O jobid)" -o ${i} -lt 1 ]
+	do
+		i=$(( ${i} - 1 ))
+		sleep 1
+	done
+	scancel -s KILL -w ${nodename}  # hard kill all remaining jobs
+	echo "shutdown container"
+	scontrol update NodeName=${nodename} State=DOWN Reason=shutdown
+	singularity instance stop slurm-drone
+	scontrol update NodeName=${nodename} State=FUTURE
+	umount /inner-cgroup/freezer
+	umount /inner-cgroup
+	SHUTDOWN_DONE=1
+	exit 0
+}
+
+# set -x
+
+trap handler_quit EXIT
+
+echo "mounting cgroups"
+mkdir /inner-cgroup
+mount -t tmpfs none /inner-cgroup
+mkdir /inner-cgroup/freezer/
+mount --bind /sys/fs/cgroup/freezer/slurm/ /inner-cgroup/freezer/
+mount -o remount,ro /inner-cgroup
+
+echo "starting ${nodename}"
+scontrol update NodeName=${nodename} State=RESUME # revoke last DRAIN
+scontrol update NodeName=${nodename} State=FUTURE
+singularity instance start \
+	-B /inner-cgroup/:/sys/fs/cgroup/ \
+	--writable-tmpfs /shared/slurmd.sif slurm-drone \
+	slurm-ctl ${nodename}
+# scontrol update NodeName=${nodename} NodeHostname=${SLURM_JOB_ID}
+scontrol update NodeName=${nodename} NodeHostname=${TardisDroneUuid}
+if [ $? -eq 0 ] ; then
+	echo "container started, sleeping $(( 60 * ${SLURM_Walltime} - 2 ))"
+	sleep $(( 60 * ${SLURM_Walltime} - 2 ))
+fi
+handler_quit

roles/cobald/tasks/main.yml

@@ -10,11 +10,16 @@
     slurm_image_prefix: cobald
     image_name: "{{cobald_image_tag}}"
     dockerfile: "{{ lookup('template', 'cobald.Dockerfile') }}"
-    files:
-    - dest: cobald-entrypoint.sh
-      content: "{{ lookup('file', 'cobald-entrypoint.sh') }}"
-    - dest: init-cobaldmodules.sh
-      content: "{{ lookup('file', 'init-cobaldmodules.sh') }}"
+    files_list:
+    - cobald-entrypoint.sh
+    - init-cobaldmodules.sh
+    - start-drone
+    - 28-sync-container-slurmd
+    files: "
+      {%- set files = [] -%} {%- for i in files_list -%}
+        {%- set files = files.append(
+          { 'dest': i, 'content': lookup('file', i) }) -%}
+        {%- endfor %}{{ files }}"
   when: cobald_slurm | default(False)
 
 - name: build generic cobald docker image
@@ -28,13 +33,14 @@
     owner: "{{unpriv_user}}"
     group: docker
 
-- name: copy cobald config (does nothing yet)
+- name: copy cobald config
   copy:
     src: cobald-config/
-    dest: /container/volumes/cobald
+    dest: "~{{unpriv_user}}/cobald/"
+    force: False
     owner: "{{unpriv_user}}"
     group: docker
-  when: False
+    mode: "0644"
 
 - name: ensure network for cobald container exists
   docker_network:
@@ -64,7 +70,7 @@
     repo: https://github.com/thoto/cobald
     dest: "~{{unpriv_user}}/cobald-src"
     version: bugfix/mixed_construction_methods
-    update: no
+    update: no  # FIXME
   become: yes
   become_user: "{{unpriv_user}}"
   register: cobald_git_pull
@@ -74,6 +80,7 @@
     repo: https://github.com/MatterMiners/tardis
     dest: "~{{unpriv_user}}/tardis-src"
     version: master
+    update: no  # FIXME
   become: yes
   become_user: "{{unpriv_user}}"
   register: tardis_git_pull
@@ -109,6 +116,13 @@
   include_tasks: facts.yml
   when: cobald_container_hostname is not defined
 
+- name: build singularity container
+  include_tasks:
+    file: singularity.yml
+    apply:
+      tags: singularity
+  tags: singularity
+
 - name: run cobald container
   docker_container:
     name: "{{ container_name | default('cobald') }}"
@@ -129,6 +143,7 @@
     cobald_mounts:
     - "~{{unpriv_user}}/cobald:/etc/cobald"
     # - /container/volumes/cobald:/etc/cobald:ro
+    - "/container/docker-images/sing-slurmd/build/:/slurm-singimage/:ro"
     - "~{{unpriv_user}}/cobald/modules:/usr/local/src/cobaldmodules"
     - "~{{unpriv_user}}/cobald-src:/usr/local/src/cobald:ro"
     - "~{{unpriv_user}}/tardis-src:/usr/local/src/tardis:ro"

roles/cobald/tasks/singularity.yml

@@ -0,0 +1,48 @@
+- name: setup singularity
+  import_role: name="singularity"
+  tags: singularity
+
+- name: make singularity image build directory
+  file:
+    state: directory
+    path: "{{item}}"
+    owner: "{{unpriv_user}}"
+    group: "docker"
+    mode: "0755"
+  loop:
+    - /container/docker-images/sing-slurmd
+    - /container/docker-images/sing-slurmd/cache
+    - /container/docker-images/sing-slurmd/build
+
+- name: copy slurm singularity container files
+  copy:
+    src: "{{item}}"
+    dest: "/container/docker-images/sing-slurmd/{{item}}"
+    owner: "{{unpriv_user}}"
+    group: "docker"
+  loop:
+    - slurm-slurmd.def
+    - 31-slurmd-configless
+    - cgroup.conf.noautomount
+  register: cobald_copy_sing_files
+
+- name: remove old container
+  file:
+    path: /container/docker-images/sing-slurmd/build/slurmd.sif
+    state: absent
+  when: cobald_copy_sing_files.changed
+
+- name: build container
+  shell:
+    chdir: /container/docker-images/sing-slurmd/
+    cmd: SINGULARITY_TMPDIR=/container/docker-images/sing-slurmd/cache
+      singularity build --disable-cache
+      /container/docker-images/sing-slurmd/build/slurmd.sif
+      /container/docker-images/sing-slurmd/slurm-slurmd.def
+    creates: /container/docker-images/sing-slurmd/build/slurmd.sif
+  register: cobald_sing_build
+
+- debug: msg="{{[cobald_sing_build.stdout, cobald_sing_build.stderr]}}"
+  tags: [ never, debug ]
+
+# TODO: trigger copy in cobald container when slurmd.sif rebuilt

roles/cobald/templates/cobald.Dockerfile

@@ -48,6 +48,10 @@ COPY init-cobaldmodules.sh /usr/local/lib/entrypoints.d/50-init-cobaldmodules.sh
 
 RUN chmod 755 /usr/local/lib/entrypoints.d/50-init-cobaldmodules.sh
 
+COPY start-drone /usr/local/bin/start-drone
+COPY 28-sync-container-slurmd /etc/docker-init.d/28-sync-container-slurmd
+RUN chmod 755 /usr/local/bin/start-drone /etc/docker-init.d/28-sync-container-slurmd
+
 RUN echo -e "#!/bin/sh\npython3 -m cobald.daemon /etc/cobald/config.yaml" >> /etc/docker-init.d/70-cobald && chmod 755 /etc/docker-init.d/70-cobald
 
 {% if cobald_docker_default_command | default(True) -%}

roles/singularity/tasks/main.yml

@@ -8,7 +8,7 @@
     value: "15000"
     sysctl_file: /etc/sysctl.d/90-max_net_namespaces.conf
 
-- name: "enable user thoto for fakeroot access"
+- name: "enable user {{unpriv_user}} for fakeroot access"
   lineinfile:
     line: "{{unpriv_user}}:4294836224:65536"
     dest: "{{item}}"

roles/slurm/files/slurmd.Dockerfile

@@ -9,3 +9,6 @@ RUN chmod 755 /etc/docker-init.d/30-slurmd
 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"
+
+RUN yum install -y singularity && \
+	yum clean all && rm -rf /var/cache/yum

roles/slurm/tasks/host-config.yml

@@ -0,0 +1,35 @@
+# TODO: this does not work quite right since slurm-ctl does not reach the host
+# system. sinfo, scontrol etc. work but srun does not!
+
+- name: "get addresses from docker network"
+  docker_network_info:
+    name: "{{ docker_network }}"
+  register: slurm_network_data
+
+- name: link host slurm config
+  file:
+    path: "/etc/slurm/slurm.conf"
+    src: "/container/volumes/slurm/slurm.conf"
+    force: True
+    state: link
+    backup: True
+
+- name: create slurm user
+  user:
+    name: slurm
+    system: True
+
+- name: place entry of slurm-ctl in host /etc/hosts
+  lineinfile:
+    line: "{{slurm_network_data.network.Containers | dict2items 
+      | json_query('[?value.Name==`slurm-ctl`].value.IPv4Address') | first
+      | ipaddr('address') }}\tslurm-ctl"
+    regexp: "^(\\S*)(\\s*)slurm-ctl$"
+    path: /etc/hosts
+    backup: True
+
+- name: start munge locally
+  service:
+    name: munge
+    enabled: True
+    state: started

roles/slurm/tasks/main.yml

@@ -58,6 +58,13 @@
     path: /container/volumes/slurm/
     state: directory
 
+- name: "create docker network to make service discovery work"
+  docker_network:
+    name: "{{ docker_network }}"
+    state: present
+  register: slurm_network_data
+  tags: slurm-config
+
 - name: upload slurm config
   template:
     force: true
@@ -71,24 +78,25 @@
     slurm_exec_node_mem: 5000  # RealMemory=5964
     slurm_alloc_nodes_default:
     - name: "{{slurm_prefix+'-submit1'}}"
+    - name: "{{ inventory_hostname }}"
+      addr: "{{ slurm_network_data.network.IPAM.Config[0].Gateway }}"
     alloc_nodes: "{{ slurm_alloc_nodes_default + extra_nodes | default([])}}"
+    partitions:
+    - name: cobald
+      nodeprefix: drone
+      num_nodes: 10
+      node_cores: 3
+      node_mem: 4900
+      port: 16818
+      initstate: FUTURE
   notify: reconfigure slurm
   tags: slurm-config
 
-- name: "create docker network to make service discovery work"
-  docker_network:
-    name: "{{ docker_network }}"
-    state: present
-
 - name: "create docker volume for shared access between nodes"
   docker_volume:
     name: slurm-shared
     state: present
 
-# 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.
-
 - name: run slurm docker containers
   docker_container:
     name: "{{ slurm_prefix }}-{{ item.machine }}"
@@ -98,6 +106,7 @@
     ports: "{{ item.exposed_ports | default([]) }}"
     networks:
     - name: "{{ docker_network }}"
+      aliases: "{{ item.aliases | default(omit) }}"
     env:
       slurmuser: "{{ slurm_user }}"
     image: "{{ item.image }}"
@@ -114,6 +123,10 @@
     label: "{{slurm_prefix}}-{{ item.machine }}"
   tags: slurm-config
 
+- name: configure host system to integrate into slurm cluster
+  import_tasks: host-config.yml
+  when: slurm_hostsystem_cluster_access | default(False)
+
 - name: export facts about slurm cluster to be used by other modules
   set_fact:
     slurm:

roles/slurm/templates/slurm.conf.j2

@@ -9,6 +9,8 @@ ControlMachine=slurm-ctl
 AuthType=auth/munge
 #CheckpointType=checkpoint/none
 CryptoType=crypto/munge
+CommunicationParameters=NoAddrCache
+SlurmctldParameters=enable_configless
 #DisableRootJobs=NO
 #EnforcePartLimits=NO
 #Epilog=
@@ -103,8 +105,10 @@ Waittime=0
 #MaxMemPerCPU=0
 #SchedulerTimeSlice=30
 SchedulerType=sched/backfill
-SelectType=select/linear
+# SelectType=select/linear
+SelectType=select/cons_res
 #SelectTypeParameters=
+SelectTypeParameters=CR_CORE
 #
 #
 # JOB PRIORITY
@@ -166,12 +170,24 @@ SlurmSchedLogFile={{slurm_log_path_sched}}
 NodeName=slurm-exec[1-{{ num_nodes }}] CPUs={{ slurm_exec_node_cores }} {{''
 	}} RealMemory={{ slurm_exec_node_mem }} {{''
 	}} CoresPerSocket={{ slurm_exec_node_cores }} State=UNKNOWN
-# RealMemory=5964
+{% for p in partitions | default([]) %}
+NodeName={{ p.nodeprefix }}[1-{{ p.num_nodes }}] CPUs={{ p.node_cores }} {{''
+	}} RealMemory={{ p.node_mem }} {{''
+	}} CoresPerSocket={{ p.node_cores }} {{''
+	}} {%- if p.port is defined %} Port={{ p.port}} {% endif %}{{''
+	}} State={{ p.initstate | default('UNKNOWN') }}
+{% endfor %}
 {% for i in alloc_nodes -%}
 NodeName={{i.name}}
 	{%- if i.hostname is defined %} NodeHostname={{i.hostname}} {% endif %}
+	{%- if i.addr is defined %} NodeAddr={{i.addr}} {% endif %}
 	State=UNKNOWN
 {% endfor %}
-PartitionName=debug Nodes=slurm-exec[1-{{num_nodes}}] {{ ''
+PartitionName=container Nodes=slurm-exec[1-{{num_nodes}}] {{ ''
 	}} AllocNodes={{alloc_nodes |map(attribute='name') | join(',')}} {{ ''
 	}} Default=YES MaxTime=INFINITE State=UP
+{% for p in partitions | default([]) %}
+PartitionName={{ p.name }} Nodes={{ p.nodeprefix }}[1-{{ p.num_nodes }}] {{ ''
+	}} AllocNodes={{alloc_nodes |map(attribute='name') | join(',')}} {{ ''
+	}} MaxTime=INFINITE State=UP
+{% endfor %}

roles/slurm/vars/main.yml

@@ -10,7 +10,8 @@ slurm_nodes_exec: | # extend range to execute nodes list
   {% set slurm_nodes_exec = [] %}
   {% for i in range(1, num_nodes+1) -%}
     {% set _ = slurm_nodes_exec.extend([
-      {'machine':'exec%s'|format(i), 'image': 'slurm:slurmd'}]) -%}
+      {'machine':'exec%s'|format(i), 'image': 'slurm:slurmd',
+       'aliases':['drone%s'|format(i)]}]) -%}
   {%- endfor %}
   {{ slurm_nodes_exec }}
 slurm_default_mounts: