Unicorn does not come up (error 502) after hard restart of Docker server

Steps to reproduce

1. Run GitLab using the guide
2. Power cycle the server running Docker

Actual result

GitLab will never come up fully, showing error 502.

The docker container logs will have this:

2017-07-26 23:20:38,558 INFO spawned: 'unicorn' with pid 612
2017-07-26 23:20:39,160 INFO exited: unicorn (exit status 1; not expected)
...
2017-07-26 23:20:46,864 INFO spawned: 'unicorn' with pid 647
2017-07-26 23:20:47,312 INFO exited: unicorn (exit status 1; not expected)
2017-07-26 23:20:48,313 INFO gave up: unicorn entered FATAL state, too many start retries too quickly

unicorn_stderr.log will have this:

...
/home/git/gitlab/vendor/bundle/ruby/2.3.0/gems/unicorn-5.1.0/lib/unicorn/http_server.rb:195:in `pid=': Already running on PID:601 (or pid=/home/git/gitlab/tmp/pids/unicorn.pid is stale) (ArgumentError)
        from /home/git/gitlab/vendor/bundle/ruby/2.3.0/gems/unicorn-5.1.0/lib/unicorn/http_server.rb:127:in `start'
        from /home/git/gitlab/vendor/bundle/ruby/2.3.0/gems/unicorn-5.1.0/bin/unicorn_rails:209:in `<top (required)>'
        from /home/git/gitlab/vendor/bundle/ruby/2.3.0/bin/unicorn_rails:22:in `load'
        from /home/git/gitlab/vendor/bundle/ruby/2.3.0/bin/unicorn_rails:22:in `<main>'

Workaround

The only way to bring up GitLab will be to docker exec into the container, manually delete the stale pid file and restart the container:

docker exec -it gitlab rm /home/git/gitlab/tmp/pids/unicorn.pid && docker restart gitlab

Expected result

GitLab comes up without manual intervention.

Start with the Docker container startup behavior and the Unicorn PID path at /home/git/gitlab/tmp/pids/unicorn.pid; reproduce the hard-restart sequence and inspect unicorn_stderr.log. Trace how the container handles the stale PID after restart, and consider the issue done when GitLab starts without manually deleting the file.