gaarf-js@2.11.3 Fix: GoogleAdsRpcApiClient: removed gRPC error codes …

…parsing, all errors not from the API are treated as transient with enabled retry

Change-Id: I853748bc53f1b28e3d317a2ce7d9671c1c554911

js/package.json

@@ -1,6 +1,6 @@
 {
   "name": "google-ads-api-report-fetcher",
-  "version": "2.11.2",
+  "version": "2.11.3",
   "description": "Google Ads API Report Fetcher (gaarf)",
   "main": "./dist/index.js",
   "types": "./src/index.ts",

js/src/lib/ads-api-client.ts

@@ -240,16 +240,15 @@ export class GoogleAdsRpcApiClient
     } else {
       // it could be an error from gRPC
       // we expect an Error instance with interface of ServiceError from @grpc/grpc-js library
-      // see status codes: https://grpc.github.io/grpc/core/md_doc_statuscodes.html
-      if (
-        (<any>error).code === 14 /* UNAVAILABLE */ ||
-        (<any>error).details === "The service is currently unavailable" ||
-        (<any>error).code === 8 /* RESOURCE_EXHAUSTED */ ||
-        (<any>error).code === 4 /* DEADLINE_EXCEEDED */ ||
-        (<any>error).code === 13 /* INTERNAL */
-      ) {
-        (<any>error).retryable = true;
-      }
+      // We used to handle only a subset of error by error code
+      // (see status codes: https://grpc.github.io/grpc/core/md_doc_statuscodes.html)
+      // particularly 14 (unavailble), 13 (internal server error),
+      // 8(RESOURCE_EXHAUSTED), 4 (DEADLINE_EXCEEDED)
+      // But there was always something new that we didn't expect, so
+      // in the end it seems much safer to treat any error not from the API
+      // server as transient and allow retrying.
+      (<any>error).retryable = true;
+      return error;
     }
   }
 
@@ -292,13 +291,17 @@ export class GoogleAdsRpcApiClient
   }
 
   async *executeQueryStream(query: string, customerId: string) {
-    const customer = this.getCustomer(customerId);
     try {
+      const customer = this.getCustomer(customerId);
       // As we return an AsyncGenerator here we can't use executeWithRetry,
       // instead usages of the method should be wrapped with executeWithRetry
       // NOTE: we're iterating over the stream instead of returning it
       // for the sake of error handling
       const stream = customer.queryStream(query);
+      this.logger.debug(`Created AsyncGenerator from queryStream`, {
+        customerId,
+        query,
+      });
       for await (const row of stream) {
         yield row;
       }

js/src/lib/logger-factory.ts

@@ -106,6 +106,16 @@ export function createCloudLogger(): winston.Logger {
           type: "cloud_function",
         },
         useMessageField: false,
+        // setting redirectToStdout:true actually disables using Logging API,
+        // and simply dumps entries to stdout where the logger agent
+        // parses them and redirect to Logging.
+        // It's the only way to overcome sparodic errors
+        // during calling Logging API:
+        // "GoogleError: Total timeout of API google.logging.v2.LoggingServiceV2
+        // exceeded 60000 milliseconds before any response was received"
+        // See https://github.com/googleapis/nodejs-logging/issues/1185
+        // And even recommended in
+        // https://cloud.google.com/nodejs/docs/reference/logging-winston/latest#alternative-way-to-ingest-logs-in-google-cloud-managed-environments
         redirectToStdout: true,
         handleRejections: LOG_LEVEL === "debug",
       }),