What I find most appaling about this, is that Phusion invented their own problem (PID1 reaping) that doesn't even exist in reality - and they keep banging that drum to create as much FUD as possible around something that is completely wrong and against the actual intent of Docker (the company, the community, the tool, the best practice, the ecosystem).
I've been doing stuff with Docker for a while and I have run in to issues here and there, but never the PID reaping problem. In fact, I actually strive now to make the application PID 1 and only run 1 process. This is specifically so that the container will exit and the host or other scheduling logic can deal with the event.
And the problem appears in such edge cases that it is doubtfully worthy of so much attention. When a pid-1 process spawns daemons, that in turn leave orphaned zombies - this pid-1 will get assigned with them. (described in 'man 7 pid_namespaces')
Supposedly this non-init pid1 is not reaping orphaned zombies, which might get the namespace to pid_max (33k default) and run out of pids to use.
3 comments
[ 2.9 ms ] story [ 20.1 ms ] threadhttps://github.com/docker/docker/issues/11529
And the problem appears in such edge cases that it is doubtfully worthy of so much attention. When a pid-1 process spawns daemons, that in turn leave orphaned zombies - this pid-1 will get assigned with them. (described in 'man 7 pid_namespaces')
Supposedly this non-init pid1 is not reaping orphaned zombies, which might get the namespace to pid_max (33k default) and run out of pids to use.