reverseproxy: Fix health_port being ignored in health checks

[akindemirci]

Fix #7524

Assistance Disclosure

"I used Claude Sonnet 4.6 for identifying and fixing the bug, then I tested the compiled binary."

[CLAassistant]

All committers have signed the CLA.

[mholt]

Thanks. Can your LLM add a test case?

Annnd... ideally, explain when/why it broke?

[akindemirci]

I made my LLM add a test case.

I really couldn't understand the algorithm behind the code. It's best to paste what LLM figured out:

Bug

Commit 6c38ae73 introduced health_upstream support. To populate DialInfo.Upstream,
it replaced the old code (which built DialInfo directly from the already port-adjusted
addr) with fillDialInfo called on the original upstream. This caused fillDialInfo
to re-parse upstream.Dial from scratch — discarding the health port that had already
been applied to addr. The transport then used dialInfo.Address (original port) to
dial, ignoring the URL host (correct port).

Fix

Added an else if branch in doActiveHealthCheckForAllHosts: when health_port is set,
build dialInfoUpstream using addr.JoinHostPort(0) — which already has the health port
baked in — so fillDialInfo produces the correct dialInfo.Address.

} else if upstream.activeHealthCheckPort != 0 {
    // health_port overrides the port; addr has already been updated
    // with the health port, so use its address for dialing
    dialInfoUpstream = &Upstream{
        Dial: addr.JoinHostPort(0),
    }
}

[francislavoie]

Thanks!