Como faço com que os scripts de inicialização do cloud-init corram sempre que as minhas botas de instância EC2?

Em /etc/init.d/cloud-init-user-scripts, edite esta linha:

/usr/bin/cloud-init-run-module once-per-instance user-scripts execute run-parts ${SCRIPT_DIR} >/dev/null && success || failure

A

 /usr/bin/cloud-init-run-module always user-scripts execute run-parts ${SCRIPT_DIR} >/dev/null && success || failure

Uma possibilidade, embora um pouco hackish, é apagar o arquivo de bloqueio que o cloud-init usa para determinar se o script do usuário já foi executado ou não. No meu caso (Amazon Linux AMI), este ficheiro de bloqueio está localizado em /var/lib/cloud/sem/ e chama-se user-scripts.i-7f3f1d11 (a parte hash no fim muda cada arranque). Portanto, o seguinte script de dados de usuário adicionado ao final do arquivo Include irá fazer o truque:

#!/bin/sh
rm /var/lib/cloud/sem/user-scripts.*

Não tenho a certeza se isto terá quaisquer efeitos adversos em qualquer outra coisa, mas tem trabalhou nas minhas experiências.

O Cloud-init suporta isto agora de forma nativa, veja as descrições de comandos do runcmd vs bootcmd na documentação (http://cloudinit.readthedocs.io/en/latest/topics/examples.html#run-commands-on-first-boot):

"runcmd":

#cloud-config

# run commands
# default: none
# runcmd contains a list of either lists or a string
# each item will be executed in order at rc.local like level with
# output to the console
# - runcmd only runs during the first boot
# - if the item is a list, the items will be properly executed as if
#   passed to execve(3) (with the first arg as the command).
# - if the item is a string, it will be simply written to the file and
#   will be interpreted by 'sh'
#
# Note, that the list has to be proper yaml, so you have to quote
# any characters yaml would eat (':' can be problematic)
runcmd:
 - [ ls, -l, / ]
 - [ sh, -xc, "echo $(date) ': hello world!'" ]
 - [ sh, -c, echo "=========hello world'=========" ]
 - ls -l /root
 - [ wget, "http://slashdot.org", -O, /tmp/index.html ]

"bootcmd":

#cloud-config

# boot commands
# default: none
# this is very similar to runcmd, but commands run very early
# in the boot process, only slightly after a 'boothook' would run.
# bootcmd should really only be used for things that could not be
# done later in the boot process.  bootcmd is very much like
# boothook, but possibly with more friendly.
# - bootcmd will run on every boot
# - the INSTANCE_ID variable will be set to the current instance id.
# - you can use 'cloud-init-per' command to help only run once
bootcmd:
 - echo 192.168.1.130 us.archive.ubuntu.com >> /etc/hosts
 - [ cloud-init-per, once, mymkfs, mkfs, /dev/vdb ]

Note também o exemplo de comando "cloud-init-per" no bootcmd. Da sua ajuda.

Usage: cloud-init-per frequency name cmd [ arg1 [ arg2 [ ... ] ]
   run cmd with arguments provided.

   This utility can make it easier to use boothooks or bootcmd
   on a per "once" or "always" basis.

   If frequency is:
      * once: run only once (do not re-run for new instance-id)
      * instance: run only the first boot for a given instance-id
      * always: run every boot

MyResource:
  Type: AWS::EC2::Instance
  Metadata:
    AWS::CloudFormation::Init:
      configSets:
        setup_process:
          - "prepare"
          - "run_for_instance"
      prepare:
        commands:
          01_apt_update:
            command: "apt-get update"
          02_clone_project:
            command: "mkdir -p /replication && rm -rf /replication/* && git clone https://github.com/awslabs/dynamodb-cross-region-library.git /replication/dynamodb-cross-region-library/"
          03_build_project:
            command: "mvn install -DskipTests=true"
            cwd: "/replication/dynamodb-cross-region-library"
          04_prepare_for_west:
            command: "mkdir -p /replication/replication-west && rm -rf /replication/replication-west/* && cp /replication/dynamodb-cross-region-library/target/dynamodb-cross-region-replication-1.2.1.jar /replication/replication-west/replication-runner.jar"
      run_for_instance:
        commands:
          01_run:
            command: !Sub "java -jar replication-runner.jar --sourceRegion us-east-1 --sourceTable ${TableName} --destinationRegion ap-southeast-1 --destinationTable ${TableName} --taskName -us-ap >/dev/null 2>&1 &"
            cwd: "/replication/replication-west"
  Properties:
    UserData:
      Fn::Base64:
        !Sub |
          #cloud-config
          cloud_final_modules:
           - [scripts-user, always]
          runcmd:
           - /usr/local/bin/cfn-init -v -c setup_process --stack ${AWS::StackName} --resource MyResource --region ${AWS::Region}
           - /usr/local/bin/cfn-signal -e $? --stack ${AWS::StackName} --resource MyResource --region ${AWS::Region}

Esta é a configuração para o processo de replicação transversal do DynamoDb.