If you are on beta.27, update. If beta.27 already left your install restarting in a loop, updating cannot reach you — recreate the container (docker compose up -d --force-recreate, or docker rm -f and start it again) and you will land here.
Same fix as v2.17.11, which has the full reasoning.
Fixed
The init that beta.27 put in front of the entrypoint was the first time a Docksentry image had ever changed its ENTRYPOINT, and the self-update could not survive it. Config.Entrypoint echoes the image's own entrypoint when nobody overrode anything, so "did the user choose this" is only answerable against the image the container was built from — it was compared against the new image, which turns every image-side change into a phantom override. And the swap dropped everything after the image in its command line, so the --entrypoint flag survived while the script it pointed at did not.
A bare python3 reads stdin, gets EOF, exits 0, and the restart policy starts it again. Measured at 7 restarts in 8 seconds; @NotRetarded reported over a thousand.
A genuine entrypoint override still survives with its arguments intact.
Changed
The init is not back yet, so the ssh process leak is still here. The update into this release still runs the old code, so the entrypoint has to stay unchanged for that update to survive. It returns once people are on a version that can take the change. tcp://, context:// and single-host installs were never affected.