[call-v3] Put retries under an experiment (grpc#38255)

I've observed a few test flakes with the new retry code... put it under an experiment for now so that those flakes don't block submission of unrelated code. Closes grpc#38255 COPYBARA_INTEGRATE_REVIEW=grpc#38255 from ctiller:retry-experiment 1ca6a18 PiperOrigin-RevId: 704596499
XuanWang-Amos · Dec 10, 2024 · 25813a0 · 25813a0
1 parent cd57733
commit 25813a0
Show file tree

Hide file tree

Showing 19 changed files with 45 additions and 0 deletions.
diff --git a/bazel/experiments.bzl b/bazel/experiments.bzl
diff --git a/src/core/lib/experiments/experiments.cc b/src/core/lib/experiments/experiments.cc
diff --git a/src/core/lib/experiments/experiments.h b/src/core/lib/experiments/experiments.h
diff --git a/src/core/lib/experiments/experiments.yaml b/src/core/lib/experiments/experiments.yaml
@@ -177,6 +177,11 @@
   owner: [email protected]
   test_tags: []
   allow_in_fuzzing_config: false # experiment currently crashes if enabled
+- name: retry_in_callv3
+  description: Support retries with call-v3
+  expiry: 2025/06/06
+  owner: [email protected]
+  test_tags: [core_end2end_test]
 - name: rq_fast_reject
   description:
     Resource quota rejects requests immediately (before allocating the request

diff --git a/test/core/end2end/end2end_test_suites.cc b/test/core/end2end/end2end_test_suites.cc
@@ -536,6 +536,7 @@ class ChaoticGoodFixture : public CoreTestFixture {
         args.Set(GRPC_ARG_CHAOTIC_GOOD_DATA_CONNECTIONS, data_connections_)
             .Set(GRPC_ARG_CHAOTIC_GOOD_MAX_RECV_CHUNK_SIZE, chunk_size_)
             .Set(GRPC_ARG_CHAOTIC_GOOD_MAX_SEND_CHUNK_SIZE, chunk_size_)
+            .Set(GRPC_ARG_ENABLE_RETRIES, IsRetryInCallv3Enabled())
             .ToC()
             .get(),
         nullptr);

diff --git a/test/core/end2end/tests/retry.cc b/test/core/end2end/tests/retry.cc
@@ -34,6 +34,7 @@ namespace grpc_core {
 // - first attempt returns ABORTED
 // - second attempt returns OK
 CORE_END2END_TEST(RetryTest, Retry) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_cancel_after_first_attempt_starts.cc b/test/core/end2end/tests/retry_cancel_after_first_attempt_starts.cc
@@ -30,6 +30,7 @@ namespace {
 // Tests that we can unref a call after the first attempt starts but
 // before any ops complete.  This should not cause a memory leak.
 CORE_END2END_TEST(RetryTest, RetryCancelAfterFirstAttemptStarts) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   // This is a workaround for the flakiness that if the server ever enters
   // GracefulShutdown for whatever reason while the client has already been
   // shutdown, the test would not timeout and fail.

diff --git a/test/core/end2end/tests/retry_cancellation.cc b/test/core/end2end/tests/retry_cancellation.cc
@@ -94,10 +94,12 @@ void TestRetryCancellation(CoreEnd2endTest& test,
 }
 
 CORE_END2END_TEST(RetryTest, RetryCancellation) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   TestRetryCancellation(*this, std::make_unique<CancelCancellationMode>());
 }
 
 CORE_END2END_TEST(RetryTest, RetryDeadline) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   TestRetryCancellation(*this, std::make_unique<DeadlineCancellationMode>());
 }
 

diff --git a/test/core/end2end/tests/retry_non_retriable_status.cc b/test/core/end2end/tests/retry_non_retriable_status.cc
@@ -34,6 +34,7 @@ namespace {
 // - 1 retry allowed for ABORTED status
 // - first attempt gets INVALID_ARGUMENT, so no retry is done
 CORE_END2END_TEST(RetryTest, RetryNonRetriableStatus) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_non_retriable_status_before_trailers.cc b/test/core/end2end/tests/retry_non_retriable_status_before_trailers.cc
@@ -36,6 +36,7 @@ namespace {
 // - first attempt gets INVALID_ARGUMENT, so no retry is done
 CORE_END2END_TEST(RetryTest,
                   RetryNonRetriableStatusBeforeRecvTrailingMetadataStarted) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_recv_initial_metadata.cc b/test/core/end2end/tests/retry_recv_initial_metadata.cc
@@ -35,6 +35,7 @@ namespace {
 // - first attempt receives initial metadata before trailing metadata,
 //   so no retry is done even though status was ABORTED
 CORE_END2END_TEST(RetryTest, RetryRecvInitialMetadata) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_recv_message.cc b/test/core/end2end/tests/retry_recv_message.cc
@@ -35,6 +35,7 @@ namespace {
 // - first attempt receives a message and therefore does not retry even
 //   though the final status is ABORTED
 CORE_END2END_TEST(RetryTest, RetryRecvMessage) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_send_initial_metadata_refs.cc b/test/core/end2end/tests/retry_send_initial_metadata_refs.cc
@@ -37,6 +37,7 @@ namespace {
 // - first attempt returns ABORTED
 // - second attempt returns OK
 CORE_END2END_TEST(RetryTest, RetrySendInitialMetadataRefs) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_server_pushback_delay.cc b/test/core/end2end/tests/retry_server_pushback_delay.cc
@@ -35,6 +35,7 @@ namespace {
 // - first attempt gets ABORTED with a long delay
 // - second attempt succeeds
 CORE_END2END_TEST(RetryTest, RetryServerPushbackDelay) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_server_pushback_disabled.cc b/test/core/end2end/tests/retry_server_pushback_disabled.cc
@@ -35,6 +35,7 @@ namespace {
 // - first attempt gets ABORTED
 // - second attempt gets ABORTED but server push back disables retrying
 CORE_END2END_TEST(RetryTest, RetryServerPushbackDisabled) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_throttled.cc b/test/core/end2end/tests/retry_throttled.cc
@@ -34,6 +34,7 @@ namespace {
 // - 1 retry allowed for ABORTED status
 // - first attempt gets ABORTED but is over limit, so no retry is done
 CORE_END2END_TEST(RetryTest, RetryThrottled) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,
       "{\n"

diff --git a/test/core/end2end/tests/retry_too_many_attempts.cc b/test/core/end2end/tests/retry_too_many_attempts.cc
@@ -34,6 +34,7 @@ namespace {
 // - first attempt gets ABORTED
 // - second attempt gets ABORTED but does not retry
 CORE_END2END_TEST(RetryTest, RetryTooManyAttempts) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_unref_before_finish.cc b/test/core/end2end/tests/retry_unref_before_finish.cc
@@ -29,6 +29,7 @@ namespace {
 // Tests that we can unref a call whose status is cached but not yet
 // requested by the application.  This should not cause a memory leak.
 CORE_END2END_TEST(RetryTest, RetryUnrefBeforeFinish) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,

diff --git a/test/core/end2end/tests/retry_unref_before_recv.cc b/test/core/end2end/tests/retry_unref_before_recv.cc
@@ -32,6 +32,7 @@ namespace {
 // they complete.  This ensures that we don't drop callbacks or cause a
 // memory leak.
 CORE_END2END_TEST(RetryTest, UnrefBeforeRecv) {
+  if (!IsRetryInCallv3Enabled()) SKIP_IF_V3();
   InitServer(ChannelArgs());
   InitClient(ChannelArgs().Set(
       GRPC_ARG_SERVICE_CONFIG,