Commit e4100990 authored by Stefano Alberto Russo's avatar Stefano Alberto Russo
Browse files

Added the standalone computing resource service with Podman, Docker and... 

Added the standalone computing resource service with Podman, Docker and Singularity support. Improved demo slurm cluster naming.
parent 76b3ad5a

docker-compose-dev.yml

+15 −6
Original line number Diff line number Diff line 
version: '3'

services:



  slurmclustermaster-main:

  slurmclustermaster:

    image: "rosetta/slurmclustermaster"

    container_name: slurmclustermaster-main

    hostname: slurmclustermaster-main

    container_name: slurmclustermaster

    hostname: slurmclustermaster

    environment:

      - SAFEMODE=False

    privileged: true
@@ -12,10 +12,10 @@ services: 
      - ./data/shared:/shared

      # - ./data/singularity_cache:/rosetta/.singularity/cache # Not working, check permissions...



  slurmclusterworker-one:

  slurmclusterworker:

    image: "rosetta/slurmclusterworker"

    container_name: slurmclusterworker-one

    hostname: slurmclusterworker-one

    container_name: slurmclusterworker

    hostname: slurmclusterworker

    environment:

      - SAFEMODE=False

    privileged: true
@@ -23,6 +23,15 @@ services: 
      - ./data/shared:/shared

      - /var/run/docker.sock:/var/run/docker.sock



  standaloneworker:

    image: "rosetta/standaloneworker"

    container_name: standaloneworker

    hostname: standaloneworker

    privileged: true

    volumes:

      - ./data/shared:/shared

      - /var/run/docker.sock:/var/run/docker.sock



  dregistry:

    container_name: dregistry

    hostname: dregistry

rosetta/build

+1 −0
Original line number Diff line number Diff line
@@ -34,6 +34,7 @@ if [[ "x$SERVICE" == "x" ]] ; then 
    $BUILD_COMMAND services/slurmcluster -t rosetta/slurmcluster    

    $BUILD_COMMAND services/slurmclustermaster -t rosetta/slurmclustermaster    

    $BUILD_COMMAND services/slurmclusterworker -t rosetta/slurmclusterworker    

    $BUILD_COMMAND services/standaloneworker -t rosetta/standaloneworker    

    $BUILD_COMMAND services/dregistry -t rosetta/dregistry

    $BUILD_COMMAND services/webapp -t rosetta/webapp

    $BUILD_COMMAND services/postgres -t rosetta/postgres

services/slurmbase/slurm.conf

+9 −10
Original line number Diff line number Diff line
@@ -2,7 +2,7 @@ 
# Put this file on all nodes of your cluster.

# See the slurm.conf man page for more information.

#

ControlMachine=slurmclustermaster-main

ControlMachine=slurmclustermaster

#ControlAddr=

#BackupController=

#BackupAddr=
@@ -155,16 +155,15 @@ SlurmdLogFile=/var/log/slurm-llnl/slurmd.log 
#SuspendRate=

#SuspendTime=

#

# Must add controller node explictly but don't place it into any partition

NodeName=slurmclustermaster-main CPUs=1 State=UNKNOWN

#NodeName=partitiona-instrument CPUs=1 State=UNKNOWN

#NodeName=partitionb-instrument CPUs=1 State=UNKNOWN

#NodeName=cris-instrument CPUs=1 State=UNKNOWN

# Must add controller node explicitly but don't place it into any partition

NodeName=slurmclustermaster CPUs=1 State=UNKNOWN

#

# COMPUTE NODES

NodeName=slurmclusterworker-one CPUs=1 State=UNKNOWN

#NodeName=slurmclusterworker-two CPUs=1 State=UNKNOWN

PartitionName=partition1 Nodes=slurmclusterworker-one MaxTime=INFINITE State=UP

#PartitionName=partition2 Nodes=slurmclusterworker-two MaxTime=INFINITE State=UP

NodeName=slurmclusterworker CPUs=1 State=UNKNOWN

#NodeName=slurmclusterworker-multi-one CPUs=1 State=UNKNOWN

#NodeName=slurmclusterworker-multi-two CPUs=1 State=UNKNOWN

PartitionName=partition1 Nodes=slurmclusterworker MaxTime=INFINITE State=UP

#PartitionName=partition2 Nodes=slurmclusterworker-multi-one,slurmclusterworker-multi-two MaxTime=INFINITE State=UP

services/standaloneworker/Dockerfile

0 → 100755
+36 −0
Original line number Diff line number Diff line 
FROM quay.io/podman/stable:v3.2.3



# This is necessary due to some base image permission errors.

RUN chown -R podman:podman /home/podman



# Change user

RUN usermod -l testuser podman

RUN usermod -d /home/testuser testuser

RUN ln -s /home/podman /home/testuser

RUN groupmod -n testuser podman



# Replace uid/gid mapping from podman to testuser user

COPY subuid /etc/subuid

COPY subgid /etc/subgid



#RUN dnf repolist 

#RUN dnf update --refresh

RUN dnf install -y docker singularity openssh-server

RUN ssh-keygen -A

RUN mkdir /home/testuser/.ssh

COPY keys/id_rsa.pub /home/testuser/.ssh/authorized_keys

RUN dnf install -y python wget





#----------------------

# Entrypoint

#----------------------



# Copy entrypoint

COPY entrypoint.sh /



# Give right permissions

RUN chmod 755 /entrypoint.sh



# Set entrypoint

ENTRYPOINT ["/entrypoint.sh"]
 
 No newline at end of file

services/standaloneworker/entrypoint.sh

0 → 100644
+23 −0
Original line number Diff line number Diff line 
#!/bin/bash



# Exit on any error. More complex thing could be done in future

# (see https://stackoverflow.com/questions/4381618/exit-a-script-on-error)

set -e



# Fix FUSE permissions

chmod 777 /dev/fuse



#---------------------

#  Entrypoint command

#---------------------



if [[ "x$@" == "x" ]] ; then

    echo -n "[INFO] Executing Docker entrypoint command: /usr/sbin/sshd -D"

    /usr/sbin/sshd -D

else

    ENTRYPOINT_COMMAND=$@

    echo -n "[INFO] Executing Docker entrypoint command: "

    echo $ENTRYPOINT_COMMAND

    exec "$ENTRYPOINT_COMMAND"

fi

#exec sudo -i -u testuser /bin/bash -c "$ENTRYPOINT_COMMAND"