Add tini to Dockerfile and use it as entrypoint

[dbussink]

This change adds tini as the init process for the Teslamate Docker container. The reason for this is that currently the container runs with zombie processes.

This means the current process with PID=1 in the container doesn't reap children properly and given the output I get while fixing this, child processes are dying accidentally.

There's also an option to work around this with docker-compose since compose file version 3.7, which allows passing --init for the docker container. I'm currently using that to work around the zombie processes, but adding tini fixes this independent of how people run this container.

https://github.com/krallin/tini#why-tini has more explanation as to why it's useful to use it as PID=1 instead of the Erlang VM in this specific case.

Before using a different init, this is what I see locally here:

root      522809  0.0  0.0 109980  5148 ?        Sl   09:54   0:00  \_ containerd-shim -namespace moby -workdir /var/lib/containerd/io.containerd.runtime.v1.linux/moby/x -address 
nobody    522874  112  0.1 1266124 128176 ?      Ssl  09:54   0:07      \_ /opt/app/erts-10.7/bin/beam.smp -- -root /opt/app -progname erl -- -home /opt/app -- -noshell -s elixir start_cli -mode embedded -setcookie X
nobody    523098  0.0  0.0      0     0 ?        Zs   09:54   0:00          \_ [epmd] <defunct>
nobody    523099  0.0  0.0    840    36 ?        S    09:54   0:00          \_ /opt/app/erts-10.7/bin/epmd -daemon
nobody    523103  2.5  0.0    768     4 ?        Ss   09:54   0:00          \_ erl_child_setup 1048576
nobody    523141  0.0  0.0      0     0 ?        Z    09:54   0:00          \_ [inet_gethost] <defunct>

The zombie processes are visible here. After using init: true in my docker-compose locally (I have a new enough version that supports 3.7), it looks like:

root      524088  0.0  0.0 108700  5208 ?        Sl   09:57   0:00  \_ containerd-shim -namespace moby -workdir /var/lib/containerd/io.containerd.runtime.v1.linux/moby/x -address 
nobody    524112  0.1  0.0   1104     4 ?        Ss   09:57   0:00      \_ /sbin/docker-init -- /bin/sh /entrypoint.sh bin/teslamate start
nobody    524204 14.1  0.2 1295668 171140 ?      Sl   09:57   0:09          \_ /opt/app/erts-10.7/bin/beam.smp -- -root /opt/app -progname erl -- -home /opt/app -- -noshell -s elixir start_cli -mode embedded -setcookie x
nobody    524392  0.1  0.0    768     4 ?        Ss   09:58   0:00          |   \_ erl_child_setup 1048576
nobody    524459  0.0  0.0    800     4 ?        Ss   09:58   0:00          |       \_ inet_gethost 4
nobody    524460  0.0  0.0    800    32 ?        S    09:58   0:00          |           \_ inet_gethost 4
nobody    524388  0.0  0.0    840    36 ?        S    09:58   0:00          \_ /opt/app/erts-10.7/bin/epmd -daemon

No zombie processes anymore 😄.

With change, it would show tini instead of docker-init, but the end result is the same and that way it's independent of the version of Docker that someone has, whether they pass --init or have to do something else in case of for example Kubernetes.

[adriankumpf]

Very interesting. I wasn't aware of that. Thanks!

[dbussink]

Very interesting. I wasn't aware of that. Thanks!

https://www.elastic.io/nodejs-as-pid-1-under-docker-images/ & https://blog.phusion.nl/2015/01/20/docker-and-the-pid-1-zombie-reaping-problem/ talk more about some if the issues that happen if the single process in the Docker container doesn't handle signals in a way that's compatible with what is expected of a process running as PID 1.

I've even had it cause real bugs in cases, not just zombie processes when things were killed incorrectly 😄.

Thanks for merging this 🙇.

[dbussink]

@adriankumpf Can confirm that with 1.19.0 & 1.19.1 this indeed also fixes the zombie issue 🎉:

nobody   1296212  0.2  0.0    776     4 ?        Ss   16:22   0:00      \_ /sbin/tini -- /bin/sh /entrypoint.sh bin/teslamate start
nobody   1296307 17.4  0.2 1267448 138896 ?      Sl   16:22   0:08          \_ /opt/app/erts-10.7.1/bin/beam.smp -- -root /opt/app -progname erl -- -home /opt/app -- -noshell -s elixir start_cli -mode embedded -setcookie x -sname teslamate -config /opt/app/tmp/teslamate-1.19.1-20200420182215-dc0a.runtime -boot /opt/app/releases/1.19.1/start -boot_var RELEASE_LIB /opt/app/lib -- -extra --no-halt
nobody   1296466  0.1  0.0    768     4 ?        Ss   16:22   0:00          |   \_ erl_child_setup 1048576
nobody   1296559  0.0  0.0    800     4 ?        Ss   16:22   0:00          |       \_ inet_gethost 4
nobody   1296560  0.0  0.0    800    32 ?        S    16:22   0:00          |           \_ inet_gethost 4
nobody   1296462  0.0  0.0    840    32 ?        S    16:22   0:00          \_ /opt/app/erts-10.7.1/bin/epmd -daemon

[adriankumpf]

Nice!