add toggle for including error messages in reports

We always send traces that includes an error node if the trace has an
error. In the case that sending errors is disabled, we replace the
message and remove the location.

docs/source/api/apollo-server.md

@@ -344,3 +344,7 @@ addMockFunctionsToSchema({
    itself. Set this to false to disable. You can manually invoke 'stop()' and
    'sendReport()' on other signals if you'd like. Note that 'sendReport()'
    does not run synchronously so it cannot work usefully in an 'exit' handler.
+
+*  `sendErrorTraces`: boolean
+
+   To remove errors from traces, set this to false. Defaults to true

packages/apollo-engine-reporting/src/agent.ts

@@ -60,9 +60,9 @@ export interface EngineReportingOptions {
   maxAttempts?: number;
   // Minimum backoff for retries. Defaults to 100ms.
   minimumRetryDelayMs?: number;
-  // By default, errors sending reports to Engine servers will be logged
-  // to standard error. Specify this function to process errors in a different
-  // way.
+  // By default, errors that occur when sending trace reports to Engine servers
+  // will be logged to standard error. Specify this function to process errors
+  // in a different way.
   reportErrorFunction?: (err: Error) => void;
   // A case-sensitive list of names of variables whose values should not be sent
   // to Apollo servers, or 'true' to leave out all variables. In the former
@@ -81,6 +81,8 @@ export interface EngineReportingOptions {
   handleSignals?: boolean;
   // Sends the trace report immediately. This options is useful for stateless environments
   sendReportsImmediately?: boolean;
+  // To remove errors from traces, set this to false. Defaults to true
+  sendErrorTraces?: boolean;
 
   // XXX Provide a way to set client_name, client_version, client_address,
   // service, and service_version fields. They are currently not revealed in the

packages/apollo-engine-reporting/src/extension.ts

@@ -43,7 +43,10 @@ export class EngineReportingExtension<TContext = any>
     options: EngineReportingOptions,
     addTrace: (signature: string, operationName: string, trace: Trace) => void,
   ) {
-    this.options = options;
+    this.options = {
+      sendErrorTraces: true,
+      ...options,
+    };
     this.addTrace = addTrace;
     const root = new Trace.Node();
     this.trace.root = root;
@@ -225,15 +228,22 @@ export class EngineReportingExtension<TContext = any>
             node = specificNode;
           }
         }
-        node!.error!.push(
-          new Trace.Error({
-            message: error.message,
-            location: (error.locations || []).map(
-              ({ line, column }) => new Trace.Location({ line, column }),
-            ),
-            json: JSON.stringify(error),
-          }),
-        );
+
+        // With noErrorTraces, the Engine proxy strips all error information
+        // from the traces and sends error counts inside of the aggregated stats
+        // reports. To get similar behavior inside of the apollo server metrics
+        // reporting, we always send a trace and mask out the PII
+        const errorInfo = this.options.sendErrorTraces
+          ? {
+              message: error.message,
+              location: (error.locations || []).map(
+                ({ line, column }) => new Trace.Location({ line, column }),
+              ),
+              json: JSON.stringify(error),
+            }
+          : { message: 'Masked by Apollo Engine Reporting' };
+
+        node!.error!.push(new Trace.Error(errorInfo));
       });
     }
   }