run slurmctld as user

Notice: also trying to run slurmd on execute nodes as user makes no
sense because it breaks sbatch. Furthermore there is another necessary
to run mpi jobs (just tried MpiDefault=none). I don't consider running
slurmd as root a good idea, but there seems to be no other choice at the
moment.

play.yml

@@ -51,4 +51,6 @@
     tags: docker
   - name: "setup slurm test environment in docker containers"
     role: slurm
+    vars:
+      slurm_user: slurm # or root
     tags: slurm

roles/slurm/defaults/main.yml

@@ -1 +1,5 @@
 container_privileged: False
+slurm_user: slurm
+slurm_log_path_ctld: /var/log/slurm/slurmctld.log
+slurm_log_path_d: /var/log/slurm/slurmd.log
+slurm_log_path_sched: /var/log/slurm/slurmsched.log

roles/slurm/files/entrypoint.sh

@@ -1,8 +1,9 @@
 #!/usr/bin/env bash
 set -e
 
-chown munge:munge /etc/munge/munge.key
-
-chmod 600 /etc/munge/munge.key
+if [ -f "/etc/munge/munge.key" ] ; then
+	chown munge:munge /etc/munge/munge.key
+	chmod 600 /etc/munge/munge.key
+fi
 
 exec "$@"

roles/slurm/files/slurm.Dockerfile

@@ -2,25 +2,42 @@ FROM docker.io/library/centos:7 as base
 
 RUN yum install -y epel-release && \
 	yum install -y slurm && \
-	yum clean all
+	yum clean all && rm -rf /var/cache/yum
 
 RUN	yum install -y less iproute bind-utils nmap-ncat net-tools && \
-	yum clean all
+	yum clean all && rm -rf /var/cache/yum
 
 COPY entrypoint.sh /usr/local/sbin/entrypoint.sh
 
-RUN chown root:root /usr/local/sbin/entrypoint.sh && chmod 755 /usr/local/sbin/entrypoint.sh
+RUN chown root:root /usr/local/sbin/entrypoint.sh && \
+	chmod 755 /usr/local/sbin/entrypoint.sh
 
 ENTRYPOINT [ "/usr/local/sbin/entrypoint.sh" ]
 
+ARG slurmuser=slurm
+ENV slurmuser=${slurmuser}
+
+RUN useradd -d /var/lib/slurm -m --no-log-init --system $slurmuser &&\
+	slurm-setuser -u $slurmuser -g $slurmuser -y
+
+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"
+
 FROM base as slurmd
 
 RUN	yum install -y slurm-slurmd && \
-	yum clean all
+	yum clean all && rm -rf /var/cache/yum
+
+CMD bash -c 'cat <({ su -s /bin/sh -c "munged -F" munge & \
+	slurmd -D 2>/dev/null 1>/dev/null & \
+	tail --retry --pid $! -f ${SLURMD_LOG_PATH} ${SLURM_SCHED_LOG_PATH} & })'
 
 FROM base as slurmctld
 
 RUN	yum install -y slurm-slurmctld && \
-	yum clean all
+	yum clean all && rm -rf /var/cache/yum
 
-# CMD bash -c 'cat <({ condor_master -f & tail --retry --pid $! -f /var/log/condor/MasterLog & })'
+CMD bash -c 'cat <({ su -s /bin/sh -c "munged -F" munge & \
+	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/slurmctld.Dockerfile

@@ -1,20 +1,32 @@
 FROM docker.io/library/centos:7 as base
 
 RUN yum install -y epel-release && \
-	yum install -y slurm slurm-slurmctld && \
-	yum clean all
+	yum install -y slurm && \
+	yum clean all && rm -rf /var/cache/yum
 
 RUN	yum install -y less iproute bind-utils nmap-ncat net-tools && \
-	yum clean all
-
-RUN	yum install -y slurm-slurmctld && \
-	yum clean all
+	yum clean all && rm -rf /var/cache/yum
 
 COPY entrypoint.sh /usr/local/sbin/entrypoint.sh
 
-RUN chown root:root /usr/local/sbin/entrypoint.sh && chmod 755 /usr/local/sbin/entrypoint.sh
+RUN chown root:root /usr/local/sbin/entrypoint.sh && \
+	chmod 755 /usr/local/sbin/entrypoint.sh
 
 ENTRYPOINT [ "/usr/local/sbin/entrypoint.sh" ]
 
-CMD bash -c 'cat <({ su -s /bin/sh -c "munged -F" munge & slurmctld -D & })'
-# ... & tail --retry --pid $! -f /var/log/condor/MasterLog & })'
+ARG slurmuser=slurm
+ENV slurmuser=${slurmuser}
+
+RUN useradd -d /var/lib/slurm -m --no-log-init --system $slurmuser &&\
+	slurm-setuser -u $slurmuser -g $slurmuser -y
+
+RUN	yum install -y slurm-slurmctld && \
+	yum clean all && rm -rf /var/cache/yum
+
+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 & \
+	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

@@ -1,17 +1,32 @@
 FROM docker.io/library/centos:7
 
 RUN yum install -y epel-release && \
-	yum install -y slurm slurm-slurmd && \
-	yum clean all
+	yum install -y slurm && \
+	yum clean all && rm -rf /var/cache/yum
 
 RUN	yum install -y less iproute bind-utils nmap-ncat net-tools && \
-	yum clean all
+	yum clean all && rm -rf /var/cache/yum
 
 COPY entrypoint.sh /usr/local/sbin/entrypoint.sh
 
-RUN chown root:root /usr/local/sbin/entrypoint.sh && chmod 755 /usr/local/sbin/entrypoint.sh
+RUN chown root:root /usr/local/sbin/entrypoint.sh && \
+	chmod 755 /usr/local/sbin/entrypoint.sh
 
 ENTRYPOINT [ "/usr/local/sbin/entrypoint.sh" ]
 
-# CMD bash -c 'cat <({ condor_master -f & tail --retry --pid $! -f /var/log/condor/MasterLog & })'
-CMD bash -c 'cat <({ su -s /bin/sh -c "munged -F" munge & slurmd -D & })'
+ARG slurmuser=slurm
+ENV slurmuser=${slurmuser}
+
+RUN useradd -d /var/lib/slurm -m --no-log-init --system $slurmuser &&\
+	slurm-setuser -u $slurmuser -g $slurmuser -y
+
+RUN yum install -y slurm-slurmd && \
+	yum clean all && rm -rf /var/cache/yum
+
+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 & \
+	slurmd -D 2>/dev/null 1>/dev/null & \
+	tail --retry --pid $! -f ${SLURMD_LOG_PATH} ${SLURM_SCHED_LOG_PATH} & })'

roles/slurm/tasks/main.yml

@@ -82,6 +82,8 @@
     volumes: "{{default_mounts + ( item.extra_mounts | default([]) ) }}"
     networks:
     - name: "slurm"
+    env:
+      slurmuser: "{{slurm_user}}"
     image: "{{item.image}}"
     state: started
     detach: True

roles/slurm/templates/slurm.conf.j2

@@ -32,6 +32,7 @@ CryptoType=crypto/munge
 #MaxStepCount=40000
 #MaxTasksPerNode=128
 MpiDefault=pmix
+# when running slurmd as user change to: MpiDefault=none
 #MpiParams=ports=#-#
 #PluginDir=
 #PlugStackConfig=
@@ -57,8 +58,9 @@ SlurmctldPort=6817
 SlurmdPidFile=/var/run/slurm/slurmd.pid
 SlurmdPort=6818
 SlurmdSpoolDir=/var/spool/slurm/d
-SlurmUser=root
-#SlurmdUser=root
+SlurmUser={{slurm_user}}
+SlurmdUser=root
+# SlurmdUser=slurm -> sbatch does not work
 #SrunEpilog=
 #SrunProlog=
 StateSaveLocation=/var/spool/slurm/ctld
@@ -130,7 +132,7 @@ AccountingStorageType=accounting_storage/none
 #AccountingStorageUser=
 AccountingStoreJobComment=YES
 ClusterName=cluster
-#DebugFlags=
+#DebugFlags=Steps,TraceJobs
 #JobCompHost=
 JobCompLoc=/tmp/jobcomp
 #JobCompPass=
@@ -141,10 +143,10 @@ JobCompType=jobcomp/filetxt
 JobAcctGatherFrequency=30
 JobAcctGatherType=jobacct_gather/none
 SlurmctldDebug=verbose
-#SlurmctldLogFile=
+SlurmctldLogFile={{slurm_log_path_ctld}}
 SlurmdDebug=verbose
-#SlurmdLogFile=
-#SlurmSchedLogFile=
+SlurmdLogFile={{slurm_log_path_d}}
+SlurmSchedLogFile={{slurm_log_path_sched}}
 #SlurmSchedLogLevel=
 #
 #
@@ -163,4 +165,4 @@ SlurmdDebug=verbose
 # COMPUTE NODES
 NodeName=slurm-exec[1-{{num_nodes}}] CPUs=2 CoresPerSocket=2 State=UNKNOWN
 NodeName=slurm-submit1 CPUs=1 State=UNKNOWN
-PartitionName=debug Nodes=slurm-exec[1-{{num_nodes}}] Default=YES MaxTime=INFINITE State=UP
+PartitionName=debug Nodes=slurm-exec[1-{{num_nodes}}] AllocNodes=slurm-submit1 Default=YES MaxTime=INFINITE State=UP