fix: disable gRPC retries by default

src/v1.test.ts

@@ -859,3 +859,38 @@ describe("createStructFromObject unit tests", () => {
     );
   });
 });
+
+describe("gRPC retry default (#200)", () => {
+  // Capture the channel options the client actually applies, via grpc-js's
+  // channelFactoryOverride hook (called with the final merged options).
+  function capturedChannelOptions(extra?: grpc.ClientOptions): grpc.ChannelOptions {
+    let captured: grpc.ChannelOptions = {};
+    const client = NewClient(
+      "tok",
+      "localhost:50051",
+      ClientSecurity.INSECURE_LOCALHOST_ALLOWED,
+      PreconnectServices.PERMISSIONS_SERVICE,
+      {
+        ...extra,
+        channelFactoryOverride: (
+          address: string,
+          creds: grpc.ChannelCredentials,
+          options: grpc.ChannelOptions,
+        ) => {
+          captured = options;
+          return new grpc.Channel(address, creds, options);
+        },
+      } as grpc.ClientOptions,
+    );
+    client.close();
+    return captured;
+  }
+
+  it("disables retries by default", () => {
+    expect(capturedChannelOptions()["grpc.enable_retries"]).toBe(0);
+  });
+
+  it("lets the caller re-enable retries", () => {
+    expect(capturedChannelOptions({ "grpc.enable_retries": 1 })["grpc.enable_retries"]).toBe(1);
+  });
+});

src/v1.ts

@@ -74,6 +74,10 @@ class ZedClient implements ProxyHandler<ZedDefaultClientInterface> {
     options: grpc.ClientOptions | undefined,
   ) {
     this.options = {
+      // Default to disabling gRPC retries to avoid unrecoverable failures
+      // through envoy/istio (see #112). Placed before ...options so callers can
+      // re-enable retries by passing "grpc.enable_retries" explicitly.
+      "grpc.enable_retries": 0,
       ...options,
       interceptors: [
         ...(options?.interceptors ?? []),
@@ -85,26 +89,30 @@ class ZedClient implements ProxyHandler<ZedDefaultClientInterface> {
     };
 
     if (preconnect & PreconnectServices.PERMISSIONS_SERVICE) {
-      this.acl = new PermissionsServiceClient(this.endpoint, this.creds, options);
+      this.acl = new PermissionsServiceClient(this.endpoint, this.creds, this.options);
     }
     if (preconnect & PreconnectServices.SCHEMA_SERVICE) {
-      this.ns = new SchemaServiceClient(this.endpoint, this.creds, options);
+      this.ns = new SchemaServiceClient(this.endpoint, this.creds, this.options);
     }
     if (preconnect & PreconnectServices.WATCH_SERVICE) {
-      this.watch = new WatchServiceClient(this.endpoint, this.creds, options);
+      this.watch = new WatchServiceClient(this.endpoint, this.creds, this.options);
     }
     if (preconnect & PreconnectServices.WATCH_PERMISSIONS_SERVICE) {
-      this.watchPermissions = new WatchPermissionsServiceClient(this.endpoint, this.creds, options);
+      this.watchPermissions = new WatchPermissionsServiceClient(
+        this.endpoint,
+        this.creds,
+        this.options,
+      );
     }
     if (preconnect & PreconnectServices.WATCH_PERMISSIONSETS_SERVICE) {
       this.watchPermissionSets = new WatchPermissionSetsServiceClient(
         this.endpoint,
         this.creds,
-        options,
+        this.options,
       );
     }
     if (preconnect & PreconnectServices.EXPERIMENTAL_SERVICE) {
-      this.experimental = new ExperimentalServiceClient(this.endpoint, this.creds, options);
+      this.experimental = new ExperimentalServiceClient(this.endpoint, this.creds, this.options);
     }
   }