Worker pids limit

[candleindark]

This PR provides a workaround solution to set the PID limit of the worker service container. This workaround is needed because currently Podman Compose, as of version 1.0.6, doesn't support setting of the PID limit by the pid_limit key or the deploy.reservations.pids key.

This solution provides docker-compose.dev.no-worker.yml to bring up all services in Datalad-Registry except the worker service and worker-up.sh to bring up the worker service with option to choose the PID limit using the --pids-limit option.

Note:

1. A request has been made to the podman-compose project to add support of setting the PID limit of a service container. (Add support of setting PID limit containers/podman-compose#806)
2. Additionally, this workaround will not be needed as soon as Podman is update to version 4.4 or later, https://docs.podman.io/en/v4.4/markdown/podman-update.1.html#pids-limit-limit.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Comparison is base (85987c9) 98.73% compared to head (74673c2) 98.73%.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #276   +/-   ##
=======================================
  Coverage   98.73%   98.73%           
=======================================
  Files          47       47           
  Lines        2215     2215           
=======================================
  Hits         2187     2187           
  Misses         28       28           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[yarikoptic]

Thank you for tracking it all down. Good that you shared the ps output which finally let us see the actual reason for the need for so many PIDs - they were not picked up. We had similar issue awhile back in buildbot setup where we also had clients start many of processes in docker container. Pasting here some references from my reply on matrix

Here is the commit where I added a simple init process to fix it datalad/buildbot@3dc7cb1
Here more on that zombie ripper https://github.com/phusion/baseimage-docker/tree/master#whats-inside-the-image

So I think we should not bother raising PID limit, but rather add this init process into the setup

[yarikoptic]

ok, if easy to rebase image to be on top of that https://github.com/phusion/baseimage-docker/tree/master -- go ahead.

Otherwise -- we could just copy that https://github.com/phusion/baseimage-docker/blob/master/image/bin/my_init here, and use it instead of plain "bash -c" to start celery worker. Something like

diff --git a/docker-compose.dev.yml b/docker-compose.dev.yml
index 97a9974..f40bc43 100644
--- a/docker-compose.dev.yml
+++ b/docker-compose.dev.yml
@@ -37,10 +37,11 @@ services:
       db:
         condition: service_healthy
     command: [
-      "bash", "-c",
-      "celery -A datalad_registry.make_celery:celery_app worker --loglevel INFO --pool prefork"
+      "my_init", "--skip-startup-files", "--skip-runit", "--",
+      "celery", "-A", "datalad_registry.make_celery:celery_app", "worker", "--loglevel", "INFO", "--pool", "prefork"
     ]
     volumes:
+      - ./bin/my_init:/usr/local/sbin/my_init
       - ${WORKER_PATH_AT_HOST}/data/cache:/data/cache
     environment:
       <<: *env

[candleindark]

This approach has been abandon. We will ensure the Docker image for the project has a correct init process instead which is accomplished by #282.