Am Donnerstag, dem 07.05.2026 um 19:51 +0200 schrieb Alexander
Kukushkin:
> My guess would be that the postmaster in your container is running
> with
> PID=1.
> The unexpected process was probably started by docker entrypoint
> before it
> did exec postgres [...].
> As a result, this subprocess became a direct child of the postmaster
> and we
> got the problem.
>
> The most common solution is to avoid running postmaster with PID=1.
> There are two options available:
> 1. dumb-init
> 2. start container using "docker run --init [...]"
My /docker-entrypoint-initdb.d is empty and I did not customize the
default entrypoint or command to run (there was a ready to use data dir
/ volume, so it just skipped initialization on start), so I am not sure
what else should have been started, but I know, just guesswork here.
Between starting the pod initially:
16:52:28.317
and the time the untracked process did fail:
18:39:13.947
nearly 2 hours passed, ah too bad I will never know what PID 120 was.
Yes, the postgres image postgres:18-trixie does run the postgres
process with PID=1:
postgres 1 0.0 0.2 4443124 139360 ? Ss May06 0:27 postgres
Using kubernetes as runtime, there is no "--init" option and e.g. tini
is not in the image - if is is a problem running postgres with pid=1,
could e.g. tini be added to the image (if so, do you know where to
address this?) and used as default entrypoint (I know I can build one
myself and extend that image to provide tini, but I would prefer just
to use the upstream image as-is)?
kind regards
Torsten
