Core: treat gRPC ResourceExhausted 'message larger than max' as non-r…

CHANGELOG.md

@@ -1,3 +1,6 @@
+## 2.4.12 ##
+* Core: treat gRPC ResourceExhausted "message larger than max" as non-retryable
+
 ## 2.4.11 ##
 * Table: fixed StackOverflow problem for the immediately released sessions
 

core/src/main/java/tech/ydb/core/grpc/GrpcStatuses.java

@@ -26,7 +26,7 @@ public static tech.ydb.core.Status toStatus(io.grpc.Status status, String endpoi
             return Status.SUCCESS;
         }
         Issue message = Issue.of(getMessage(status, endpoint), Issue.Severity.ERROR);
-        StatusCode code = getStatusCode(status.getCode());
+        StatusCode code = getStatusCode(status);
         Throwable cause = status.getCause();
 
         if (cause == null) {
@@ -49,16 +49,29 @@ private static String getMessage(io.grpc.Status status, String endpoint) {
         return message;
     }
 
-    private static StatusCode getStatusCode(io.grpc.Status.Code code) {
-        switch (code) {
+    private static StatusCode getStatusCode(io.grpc.Status status) {
+        switch (status.getCode()) {
             case UNAVAILABLE: return StatusCode.TRANSPORT_UNAVAILABLE;
             case UNAUTHENTICATED: return StatusCode.CLIENT_UNAUTHENTICATED;
             case CANCELLED: return StatusCode.CLIENT_CANCELLED;
             case UNIMPLEMENTED: return StatusCode.CLIENT_CALL_UNIMPLEMENTED;
             case DEADLINE_EXCEEDED: return StatusCode.CLIENT_DEADLINE_EXCEEDED;
-            case RESOURCE_EXHAUSTED: return StatusCode.CLIENT_RESOURCE_EXHAUSTED;
+            case RESOURCE_EXHAUSTED:
+                // A message that exceeds the gRPC size limit is a deterministic client-side error:
+                // retrying it with the same payload always fails identically. Map it to a non-retryable
+                // status so the caller fails fast instead of spinning through the retry budget.
+                // Other ResourceExhausted errors (e.g. quota or rate limiting) stay retryable.
+                if (isMessageLargerThanMax(status.getDescription())) {
+                    return StatusCode.BAD_REQUEST;
+                }
+                return StatusCode.CLIENT_RESOURCE_EXHAUSTED;
             default:
                 return StatusCode.CLIENT_GRPC_ERROR;
         }
     }
+
+    private static boolean isMessageLargerThanMax(String description) {
+        return description != null && (description.contains("trying to send message larger than max")
+                || description.contains("received message larger than max"));
+    }
 }

core/src/test/java/tech/ydb/core/grpc/GrpcStatusesTest.java

@@ -117,6 +117,34 @@ public void statusResourceExhausted() {
         assertEquals(Status.of(StatusCode.CLIENT_RESOURCE_EXHAUSTED).withIssues(issue), status);
     }
 
+    @Test
+    public void statusResourceExhaustedMessageTooLargeSend() {
+        Status status = GrpcStatuses.toStatus(
+                io.grpc.Status.RESOURCE_EXHAUSTED.withDescription("trying to send message larger than max"), "test");
+        Issue issue = Issue.of(
+                "gRPC error: (RESOURCE_EXHAUSTED) on test, trying to send message larger than max",
+                Issue.Severity.ERROR);
+        assertEquals(Status.of(StatusCode.BAD_REQUEST).withIssues(issue), status);
+    }
+
+    @Test
+    public void statusResourceExhaustedMessageTooLargeReceive() {
+        Status status = GrpcStatuses.toStatus(
+                io.grpc.Status.RESOURCE_EXHAUSTED.withDescription("received message larger than max"), "test");
+        Issue issue = Issue.of(
+                "gRPC error: (RESOURCE_EXHAUSTED) on test, received message larger than max",
+                Issue.Severity.ERROR);
+        assertEquals(Status.of(StatusCode.BAD_REQUEST).withIssues(issue), status);
+    }
+
+    @Test
+    public void statusResourceExhaustedOtherDescription() {
+        Status status = GrpcStatuses.toStatus(
+                io.grpc.Status.RESOURCE_EXHAUSTED.withDescription("quota exceeded"), "test");
+        Issue issue = Issue.of("gRPC error: (RESOURCE_EXHAUSTED) on test, quota exceeded", Issue.Severity.ERROR);
+        assertEquals(Status.of(StatusCode.CLIENT_RESOURCE_EXHAUSTED).withIssues(issue), status);
+    }
+
     @Test
     public void statusThrowable() {
         Throwable th = new RuntimeException("Hello");