How to install a service and lauch it during a cloud-init phase via AWS cloudformation user data bash script?

Question

I needed to add a shutdown hooks on my EC2 instances to do some resource clean up stuff.

Moreover, I would also be able to start and stop manually my instance for testing purpose and I wanted the startup and shutdown hooks to be triggered the same way as on the initial bootstrap.

I then decided to install a script as a service on AWS EC2 Ubuntu 16.04 LTS instance via a Cloudformation bash script.

Here is the first naive version of the script:

UserData:
  "Fn::Base64":
    !Sub
      - |
          #!/usr/bin/env bash

          BOOTSTRAP_SCRIPT_NAME=bootstrap
          BOOTSTRAP_SCRIPT_PATH=/etc/init.d/${BOOTSTRAP_SCRIPT_NAME}

          cat > /etc/init.d/boostrap <<EOF
          ### BEGIN INIT INFO
          # Provides:             ${BOOTSTRAP_SCRIPT_NAME}
          # Required-Start:       \\\$local_fs \\\$remote_fs \\\$network \\\$syslog \\\$named
          # Required-Stop:        \\\$local_fs \\\$remote_fs \\\$network \\\$syslog \\\$named
          # Default-Start:        2 3 4 5
          # Default-Stop:         0 1 6
          # Short-Description:    Bootstrap an instance
          # Description:          Bootstrap an instance
          ### END INIT INFO

          function start() {

              echo "STARTUP on $(date)"

          }

          function stop() {

              echo "SHUTDOWN on $(date)"

          }

          case "\$1" {
            start)
             start | tee -a /var/log/${BOOTSTRAP_SCRIPT_NAME}.log
             ;;
            stop)
             stop | tee -a /var/log/${BOOTSTRAP_SCRIPT_NAME}.log
              ;;
          }
          EOF

          chmod +x ${BOOTSTRAP_SCRIPT_PATH}

          update-rc.d -f ${BOOTSTRAP_SCRIPT_NAME} remove
          update-rc.d ${BOOTSTRAP_SCRIPT_NAME} defaults

With this version, the bootstrap script is never started.

I quickly understood that the bootstrap script was installed during the cloud-init phase and by the way during the linux sysv init phase and would not take part of the current init phase ... (If this is wrong tell me ;-))

I then decided to start it manually such as apache2 in cloudformation bash examples. I added the following line at the end of the script.

${BOOTSTRAP_SCRIPT_PATH} start

I tested it again, and saw the "STARTUP on XXX" log in the bootstrap.log file after this fix. But when I tried to stop the instance in the consol, no "SHUTDOWN on XXX" logs appeared in the bootstrap.log file ...

I log into the instance and try to start/stop the script manually ... all the startup and shutdown logs appeared 8-O. I then supposed that as the boostrap script was not identified as an init script the stop callback would not be called on instance stop or terminate ... (If this is wrong tell me ;-))

I then start and stop several times the instance from the AWS console and both STARTUP and SHUTDOWN messages still appeared in the logs. This confirmed my hypothesis. The logs are only missing during the first init and shutdown cycle.

So I did something weird and ugly ... I replace the last line start command with this one :

reboot -n 

The script now works as I need but I think there should be a cleaner way to enable my script for init or a least for the shutdown phase during cloud-init without rebooting ...

Is anyone has a best solution or more details on the issue ?

PS : I tried init u and telinit u instead of reboot with no success

Answer 1

The reason for this seems to be that the bootstrap is not started as a service the first time. It is run as a normal script. Instead of ${BOOTSTRAP_SCRIPT_PATH} start, try adding the following line to your user-data:

sudo service ${BOOTSTRAP_SCRIPT_NAME} start