Class GrpcRetry

java.lang.Object

ai.pipestream.module.pipelineprobe.pipelinecrawl.GrpcRetry

public final class GrpcRetry
extends Object

Bounded retry wrapper around a blocking gRPC stub call. Used by every control-plane call out of the pipeline-crawl orchestrator so that a stale Stork sub-channel, a momentarily-unavailable target, or a transient RESOURCE_EXHAUSTED doesn't wedge the entire orchestration.

Per-attempt deadline must live inside the supplier. The caller is expected to call stub.withDeadlineAfter(...) inside the supplier so that each retry gets a fresh deadline budget. If the deadline were applied outside, every retry would inherit the same already-spent deadline and fail immediately.

Retryable status codes (per docs/architecture/14-backpressure-and-admission.md):

• UNAVAILABLE — Stork couldn't find a healthy instance, or the channel is in TRANSIENT_FAILURE and hasn't re-resolved yet.
• DEADLINE_EXCEEDED — previous attempt timed out; the next call forces channel re-resolution.
• RESOURCE_EXHAUSTED — the gap-window backstop: target just-flipped-yellow but Consul/Stork doesn't know yet.

All other status codes (INVALID_ARGUMENT, ALREADY_EXISTS, NOT_FOUND, INTERNAL, etc.) are re-thrown immediately — they're not transport failures and retrying just wastes time.

Caller must run on a virtual thread; this class uses Thread.sleep(long) between retries, which on a VT is a cheap park but on a platform thread is a wasted thread.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
static final record	GrpcRetry.RetryPolicy	Retry budget for a single op.

Method Summary

Modifier and Type	Method	Description
static <T> T	call(String opName, Supplier<T> stubCall, GrpcRetry.RetryPolicy policy)	Invoke stubCall with bounded retry per policy.

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Method Details

call

public static <T> T call(String opName,
 Supplier<T> stubCall,
 GrpcRetry.RetryPolicy policy)

Invoke stubCall with bounded retry per policy. The supplier MUST apply its own per-attempt deadline via stub.withDeadlineAfter(...).

Type Parameters:

T - the response type returned by the stub call

Parameters:

opName - short human-readable name of the operation, used in retry log messages so operators can see what's flapping

stubCall - the blocking-stub call, applied fresh on each attempt

policy - retry budget and backoff schedule

Returns:

the successful response

Throws:

io.grpc.StatusRuntimeException - the last failure if all attempts were exhausted, or the first non-retryable failure if one occurred