Hi
Could some one from the telemetry team jump in the PR request in order provide competent answers?


Here are current comments from Docker:

The telemetry-agent-supervisor.sh calls two commands, redirecting the stdout and stderr to files. Since these commands are not started in the background, presumably they are either short-lived processes or fork and exit on their own. With that being the case, why is it necessary for the telemetry-agent-supervisor.sh script to remain running?

The conditional for the value of the PERCONA_TELEMETRY_DISABLE variable explicitly checks if the value is "0". I see nothing that sets the value explicitly to "0". As such, when will that conditional ever be true, short of the user setting it explicitly to zero? The documentation indicates you should set it to one to disable it, but as I read it, it seems by default that conditional would not evaluate to true, so perhaps the docs should say that telemetry is disabled by default and you should set PERCONA_TELEMETRY_DISABLE to zero to enable it.

Running unsupervised background processes in the container is not generally allowed in Official Images since that then requires a third process to supervise them: handling and sending signals, reaping zombies, and deciding what to do if one or more of them exit (or don't exit when asked to). Unsupervised background processes are processes that don't have a responsible parent process. Even if their PID relationship puts them in a parent/child relationship (like because of an exec), the parent does not take responsibility for the child (tracking/restarting, forwarding signals, reaping, etc). Most of this is instead better handled at a higher level in functionality provided by dockerd or in an orchestration of multiple containers like docker-compose and kubernetes (though neither seem a great fit for a "sidecar" telemetry process).