Abhimanyu Singh

@PragTob https://github.com/jruby/jruby/blob/master/lib/ruby/stdlib/coverage.rb for `JRuby`. We can probably borrow these implementations to use in `2.5`. My point is that we have incorrect data compared to when using Ruby-provided coverage and stubbed...

@PragTob I meant borrow CRuby and JRuby implementations. For example, I did this in [rspec-tracer](https://github.com/avmnu-sng/rspec-tracer/blob/main/lib/rspec_tracer/coverage_reporter.rb#L142).

https://github.com/ruby/ruby/blob/master/ext/coverage/lib/coverage.rb The line_stub is Ruby code that I borrowed without any modifications.

@theomelo We can merge this PR if it is finalized from your end.

@jcmoraisjr Thanks. It will fix the issue.

Since doing `dumb-init`, we can rewrite the signal `TERM (15)` to `USR1 (10)`. With this, `HAProxy` now performs the graceful shutdown. Additionally: - The `haproxy-ingress-controller` should handle `USR1`. https://github.com/jcmoraisjr/haproxy-ingress/blob/cf9d05a07e313e17173ec7439ca0e40fa9f04fcf/pkg/main.go#L52 -...

Yeah, `--single-child` should solve the problem. I'll try to test it out as well. However, we should update the HA process termination, by sending `USR1` and not `KILL`.

So, when we invoke `kill` HA, the ingress controller ensures all the ongoing connections have finished? If that's the case, I've missed this logic when going through the code.

I see the gap now. I used the graceful shutdown as "all the HA requests were completed before quitting HA." We can use `ha-shutdown-timeout` as this timeout because when the...

We must also consider the server and backend queues and the kernel socket queues. We use backend queues and did a quick test with `sleep 60`, but still, I see...