Set a 60s timeout across Bazel remote cache and RBE gRPC calls. Transient connection blackholes between CI runners and RBE can wedge in-flight gRPC streams. Previously, `--remote_timeout=1200` left wedged cache and ByteStream RPCs stalled for 20 minutes, while `Execution` (`Execute` / `WaitExecution`) streams had no client-side deadline at all and stalled ~30 minutes until an intermediate proxy closed the stream. Drop the `--remote_timeout=1200` (and unused `--bes_timeout=600s`) overrides in `.bazelrc` to use Bazel's 60s default, and apply a matching channel-wide 60s gRPC deadline via `--remote_grpc_service_config` so wedged cache, ByteStream, and `Execution` RPCs time out and retry within a minute. PiperOrigin-RevId: 1013412302
diff --git a/.bazelrc b/.bazelrc index 3e964e4..c64b255 100644 --- a/.bazelrc +++ b/.bazelrc
@@ -86,7 +86,6 @@ # Remote build configuration for CI. build:remote-cache --remote_cache=https://storage.googleapis.com/absl-bazel-remote-cache -build:remote-cache --remote_timeout=1200 --bes_timeout=600s build:remote-cache --remote_download_minimal # cc_test puts Windows .pdb files in the pdb_file output group rather than # runfiles, so explicitly download symbolize_test.pdb when linking hits the @@ -100,7 +99,7 @@ test:remote-cache --build_tests_only build:remote-base --define=EXECUTOR=remote -build:remote-base --remote_timeout=1200 --bes_timeout=600s +build:remote-base --remote_grpc_service_config=ci/rbe_grpc_service_config.json build:remote-base --jobs=50 build:remote-base --remote_download_minimal # Retry up for up to 15 minutes under heavy load (exponential backoff).
diff --git a/ci/rbe_grpc_service_config.json b/ci/rbe_grpc_service_config.json new file mode 100644 index 0000000..58dc87f --- /dev/null +++ b/ci/rbe_grpc_service_config.json
@@ -0,0 +1,8 @@ +{ + "methodConfig": [ + { + "name": [{}], + "timeout": "60s" + } + ] +}