Provide a way to complete a ClientRequestContext in Preprocessor

[jrhee17]

Motivation:

This issue was found while working on GrpcServicesPreprocessor refactoring.

Currently, there is a bug where exceptions/response failures in Preprocessor do not complete ctx.whenInitialized.
e.g.

WebClient.of((delegate, ctx, req) -> {
    // ctx.whenInitialized will not be completed
    throw new RuntimeException("error");
}).get("/");

While normally this may not be an issue, gRPC relies on completion of this future to signal errors:
ref:

armeria/grpc/src/main/java/com/linecorp/armeria/internal/client/grpc/ArmeriaClientCall.java

Line 180 in 1a31f3e

ctx.whenInitialized().handle((unused1, unused2) -> {

I propose two modifications to resolve this issue:

• When a Preprocessor throws an exception from the PreClient#execute calling thread, it should complete the ctx

WebClient.of((delegate, ctx, req) -> {
    throw new RuntimeException("error");
}).get("/");

• When ctx.cancel is called (even before ctx.init) is called, it should complete the ctx

WebClient.of((delegate, ctx, req) -> {
    ctx.cancel(e);
    return HttpResponse.of(400);
}).get("/");

Modifications:

• Introduced an initialized flag to guard against concurrent initialization attempts
  • finishInitialization is also modified to guard against concurrent calls
  • A new initAndFail method is introduced which acquires an event loop, initializes the cancellation scheduler, and completes the ctx
  • ctx.cancel will trigger initAndFail if the ctx is not initialized yet
  • If PreClient#execute throws an exception, initAndFail is called
  • For derived contexts, if an endpoint does not exist, it means ClientUtil#initContextAnd* needs to be called. Hence, initialization is set only if an endpoint exists.
  • responseCancellationScheduler.finishNow is called in DefaultClientRequestContext#failEarly to record cancellationCause consistently.
  • XdsPreprocessor and RouterFilter now calls ctx.cancel if a request is short-circuited before ctx.init is called.

Result:

• Requests failed at the XdsPreprocessor-level are propagated to the user correctly when using gRPC.

[github-actions]

🔍 Build Scan® (commit: cbdeab3)

Job name	Status	Build Scan®
build-ubicloud-standard-16-jdk-8	✅	https://ge.armeria.dev/s/vcu2sa7zydlsk
build-ubicloud-standard-16-jdk-21-snapshot-blockhound	✅	https://ge.armeria.dev/s/i3h25xhfpxjrw
build-ubicloud-standard-16-jdk-17-min-java-17-coverage	✅	https://ge.armeria.dev/s/eciv3smwz5rfs
build-ubicloud-standard-16-jdk-17-min-java-11	✅	https://ge.armeria.dev/s/jvpeswnjogyew
build-ubicloud-standard-16-jdk-17-leak	✅	https://ge.armeria.dev/s/nj7xfionummok
build-ubicloud-standard-16-jdk-11	✅	https://ge.armeria.dev/s/z2kjj7addoiko
build-macos-latest-jdk-21	✅	https://ge.armeria.dev/s/bo2apg2zoadew

[minwoox]

Looks great. 👍
Left small suggestions. 😉

[minwoox]

Probably, we don't need to call initializeResponseCancellationScheduler in acquireEventLoop?

[minwoox]

What do you think about moving the call to completeLogIfIncomplete from line 111 to here? It seems like it would prevent the case where an http response is returned in the pre decorator.

[minwoox]

Shouldn't we also change this line?
futureConverter, errorResponseFactory, req, false);

armeria/core/src/main/java/com/linecorp/armeria/internal/client/TailPreClient.java

Line 82 in 89e1a84

futureConverter, errorResponseFactory, req, true);

Additionally, this isn't related to this PR but shouldn't tryCompleteLog be true in these lines? cc @ikhoon

armeria/core/src/main/java/com/linecorp/armeria/client/retry/RetryingClient.java

Lines 325 to 328 in 51f0790

	(context, cause) -> HttpResponse.ofFailure(cause), ctxReq, false);
	} else {
	response = executeWithFallback(unwrap(), derivedCtx,
	(context, cause) -> HttpResponse.ofFailure(cause), ctxReq, false);

[jrhee17]

Shouldn't we also change this line? futureConverter, errorResponseFactory, req, false);

Understood that the intention is that completeLogIfIncomplete is already registered when preclients are invoked, and hence another callback doesn't need to be added

[codecov]

Codecov Report

❌ Patch coverage is 50.84746% with 29 lines in your changes missing coverage. Please review.
✅ Project coverage is 74.16%. Comparing base (8150425) to head (cbdeab3).
⚠️ Report is 231 commits behind head on main.

Files with missing lines	Patch %	Lines
...corp/armeria/xds/client/endpoint/RouterFilter.java	0.00%	13 Missing ⚠️
...a/internal/client/DefaultClientRequestContext.java	68.57%	5 Missing and 6 partials ⚠️
.../linecorp/armeria/client/retry/RetryingClient.java	0.00%	0 Missing and 1 partial ⚠️
...necorp/armeria/client/retry/RetryingRpcClient.java	0.00%	0 Missing and 1 partial ⚠️
...m/linecorp/armeria/internal/client/ClientUtil.java	83.33%	0 Missing and 1 partial ⚠️
...ria/internal/common/NoopCancellationScheduler.java	0.00%	1 Missing ⚠️
...p/armeria/xds/client/endpoint/XdsPreprocessor.java	0.00%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff              @@
##               main    #6466      +/-   ##
============================================
- Coverage     74.46%   74.16%   -0.30%     
- Complexity    22234    23023     +789     
============================================
  Files          1963     2061      +98     
  Lines         82437    86190    +3753     
  Branches      10764    11305     +541     
============================================
+ Hits          61385    63926    +2541     
- Misses        15918    16868     +950     
- Partials       5134     5396     +262     

☔ 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.

[ikhoon]

👍 👍