Improve `docker service create` command to fit in CI/CD world

Posted in category ci on 2017-01-10

Table of Contents

This is a follow-up post on the Improve docker service update command to fit in CI/CD world I wrote the other day. Premise is exactly as before but the target is now docker service create command. We will try to ensure that CI/CD setup can get feedback from executing docker service create command.

Description of issues and assumptions taken in the solution should be identical to those described in Improve docker service update command to fit in CI/CD world, so I will not repeat myself here.


Following is a very simple wrapping bash script that is doing a little magic to help wait for the feedback from an update process of a service in docker cluster:

After creating service, the script fetches service status by running docker service ps <service_name> which gives information about currently running containers. There should be no old instances of a service registered, otherwise docker service create will fail saying that service has already been created. But the number of old instances of a service can suddenly go up in cases when installed service is not stable and crashes making swarm to re-create it again and again.

#!/usr/bin/env bash

if [[ "$#" -ne 4 ]]; then
  echo "Expected arguments are missing: SERVICE_NAME STABILITY_DELAY_S ERROR_SLIP CREATE_CMD"
  exit 1

stability_delay=$(echo $2 | sed 's/[^0-9]//g')
error_slip=$(echo $3 | sed 's/[^0-9]//g')

function get_new_ids() {
  echo $(docker service ps $name | awk '(NR > 1) {print $1}' | wc -l)

echo ""
echo "     Create command:  $create_cmd"
echo "    Stability delay:  $stability_delay sec"
echo "         Error slip:  $error_slip"
echo -n "   Creating service:  "
eval $create_cmd
if [ "$exit_code" -ne 0 ]; then
  echo "ERROR: cannot create service"
  exit $exit_code
echo "done"

echo -n "     Stability test:  "
replicas=$(docker service inspect $name | grep -i replicas | sed 's/[^0-9]//g')
max_n=$(($replicas + $replicas * $error_slip))
if [ $n -gt $max_n ]; then
  echo "failed"
  exit 3
sleep $stability_delay
if [ $n -gt $max_n ]; then
  echo "failed"
  exit 3
echo "done"
echo ""

echo "SUCCESS"

The script is also quite talkative and will produce a lot of information which is typically what you want when running it as part of your CI/CD setup to ensure that troubleshooting process is not a nightmare.

Usage example

./ redis-slave 60 1 \
  "docker service create --name redis-slave redis:3.2.6-alpine"

It will output something like this:

     Create command:  docker service create --name redis-slave redis:3.2.6-alpine
    Stability delay:  120 sec
         Error slip:  1
   Creating service:  8oz8ldtxoju18v4ds1g6ith6i
     Stability test:  done


In case of a failure script is producing non-zero result code which is what is typically getting detected by CI/CD setup.