Bugfix: correctly shutdown even without replication

[carlomion]

Fixes #553
start.sh: avoid killing replication process if it is not active.

GitHub CI:

• Add test to check for correct shutdown of container
• Use tail | grep to check for streaming replication logs
• Remove unnecessary sleeps (the python functions used for the tests use a retry internally)

[what-the-diff]

PR Summary

• Improved Sleep Time Control for Docker Containers
  Changes in .github/workflows/ci.yml provide enhanced management of sleep time for several Docker containers. This optimization helps improve the efficiency of our workflows.

• Graceful Shutdown Enhancement in Replication Process
  Updates in 4.3/start.sh now allow for smoother halting of the replication process. In addition to this, Docker is now capable of receiving signals when shutdown is successful. This enhancement ensures the better reliability of our systems.

• Improved Build Performance Through Caching
  By implementing caching for Apt and Pip in 4.4/Dockerfile, we have managed to enhance build performance. This feature cuts down on build time, ensuring faster integration and deployment process.

• Enhanced Shutdown Process in 4.4
  The script 4.4/start.sh also saw improvements in its shutdown process. These changes parallel those in 4.3/start.sh with the aim of maintaining consistency across versions and improving the overall reliability of system shutdowns.

[leonardehrenfried]

This is also a good pull request. Let's merge #552 first, rebase on top of it and get this merged as well.

[leonardehrenfried]

Could you explain what is happening here?

[carlomion]

It's a feature introduces (relatively) recently by Docker buildkit: https://docs.docker.com/build/cache/#use-the-dedicated-run-cache.
It instructs the image build process to mount that directory as a volume inside the container, only for the duration of the build. This way subsequent invocation of docker build will find the cached file locally, without leaving them inside the final image, thus reducing its final size.

[leonardehrenfried]

I've tried a few times to get rid of these sleeps but always failed. I'm going to run the tests a few times to see if there is any difference now.

[carlomion]

According to the python script, it should automatically retry failed connection with an exponential backoff.
Hard-coding a forced sleep may prolong the test more than necessary, while the container may be already ready before the timeout. IMHO a polling HTTP call done periodically by the script sounds more snappy than the forced sleep.

[leonardehrenfried]

It still wasn't enough though. I will run it a few times and see.

[leonardehrenfried]

I ran the tests 5 times and they always worked without the sleep so therefore I approve.