I’ve tried all workaround mentioned in this post. From all solutions switching to older image worked best for me. I am not a friend of ‘sleeps’ and hard coded timeouts.
Anyway I learned a lot from this post and appreciate any effort people put in to help each other.
Cheers
Ah, fair enough. I was under the impression you were not willing to settle for that solution. If you are happier with that than a time penalty on the newer OS, then great 
travis supports the way we’re currently doing awscli, node, docker and docker-compose using a vm and not a limited docker image. we’ve hit plenty of speedbumps with their service, but we’re not having to manage the managed service–it’s more converting things from $CIRCLE_BRANCH to $TRAVIS_BRANCH and not trying to hammer a bad vm image into shape.
i think in the future if we were using a single language or had more limited needs for docker services and networks, the circleci 2.0 workflows would be productive. but for right now, we really need a vm and that’s just not something circleci provides.