Fix race condition when initializing HealthCheckedEndpointGroup

[hyunw9]

Motivation

Fixes #6018

A race condition can occur when adding a listener immediately after creating a HealthCheckedEndpointGroup.
Because the endpoint group initialization is asynchronous, the listener may receive endpoints in an incomplete or intermediate state.

Modification

Explicitly await the completion of HealthCheckedEndpointGroup initialization using whenReady().join() before adding a listener.
This ensures the listener will always receive a fully initialized endpoint list, eliminating any race conditions.

Result
• Race condition is fixed.
• Listener always receives a fully initialized EndpointGroup, ensuring reliable and predictable behavior.

[github-actions]

🔍 Build Scan® (commit: d64cad0)

Job name	Status	Build Scan®
build-ubicloud-standard-16-jdk-8	✅	https://ge.armeria.dev/s/eswgy6sqihtwc
build-ubicloud-standard-16-jdk-21-snapshot-blockhound	❌ (failure)	https://ge.armeria.dev/s/qvdq6jdb4sia4
build-ubicloud-standard-16-jdk-17-min-java-17-coverage	✅	https://ge.armeria.dev/s/6tmhsnu3pv4v2
build-ubicloud-standard-16-jdk-17-min-java-11	✅	https://ge.armeria.dev/s/wkvfzvsubtqlu
build-ubicloud-standard-16-jdk-17-leak	✅	https://ge.armeria.dev/s/n5if5kuqameqe
build-ubicloud-standard-16-jdk-11	✅	https://ge.armeria.dev/s/kx4ctlzbmxbzq
build-macos-latest-jdk-21	✅	https://ge.armeria.dev/s/5qps2wqzwk5aw

[jrhee17]

It seems like the test in question has failed again: https://github.com/line/armeria/actions/runs/14173631131/job/39703141791?pr=6188

[hyunw9]

I noticed that the test failure in HealthCheckedEndpointGroupTest#cacheReflectsAttributeChanges seemed to stem from an unexpected ordering issue during initialization.

In this test, registering the endpoint is essentially part of the setup phase. However, the internal initialization of HealthCheckedEndpointGroup is asynchronous, and addListener(..., true) can synchronously invoke listener.accept(latest) before the system is ready, which introduces a race condition.

Initially, I thought calling whenReady().join() would fully synchronize the group’s initialization. However, after digging into the code, I realized that although join() ensures the group is initialized, it doesn’t control the timing of the listener callback.

Internally, addListener() acquires a lock to update the listener list, but the actual call to listener.accept(latest) (triggered by notifyLatestValue = true) occurs outside the lock. That means a race condition can still occur between listener.accept(...) and the external setup logic that expects a stable state.

To address this, I changed the approach:
1. I registered the listener using addListener(endpointListener, false) to avoid immediate invocation.
2. Then, after the group was fully initialized via whenReady().join(), I manually called listener.accept(...) to reflect the initial state in a safe and deterministic manner.

This ensures that the state propagation is explicitly controlled and prevents any premature callback from breaking the test setup.

[minwoox]

Thanks!

[minwoox]

Can't we do this?

Suggested change

	endpointGroup.addListener(endpointsListener, false);
	endpointsListener.accept(endpointGroup.endpoints());
	endpointGroup.addListener(endpointsListener, true);

[hyunw9]

it works ! I think Fix duplicate endpoint handling in HealthCheckedEndpointGroup
this change helps the new group include endpoints from the previous group temporarily, which avoids any brief disappearance during the transition.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 74.57%. Comparing base (8150425) to head (d64cad0).
Report is 61 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff              @@
##               main    #6188      +/-   ##
============================================
+ Coverage     74.46%   74.57%   +0.11%     
- Complexity    22234    22460     +226     
============================================
  Files          1963     1972       +9     
  Lines         82437    82992     +555     
  Branches      10764    10798      +34     
============================================
+ Hits          61385    61894     +509     
- Misses        15918    15951      +33     
- Partials       5134     5147      +13     

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[jrhee17]

Question) Shouldn't the HealthCheckedEndpointGroup still be closed to close the ScheduledFuture at L539?

[ikhoon]

It seems like you pushed the changes of #6191 to this PR.

[hyunw9]

Reapplied correct changes after reverting accidental commits (rollback to 018a7b1).
Thanks for the notice!