fix: grpc not retrying UNAVAILABLE/UNKNOWN, e2e test stability (#1457)

Signed-off-by: Todd Baert <[email protected]>

Author: toddbaert

libs/providers/flagd/src/e2e/step-definitions/providerSteps.ts

@@ -1,4 +1,4 @@
-import { OpenFeature, ProviderStatus } from '@openfeature/server-sdk';
+import { OpenFeature, ProviderEvents, ProviderStatus } from '@openfeature/server-sdk';
 import { FlagdComposeContainer } from '../tests/flagdComposeContainer';
 import type { State, Steps } from './state';
 import { FlagdProvider } from '../../lib/flagd-provider';
@@ -26,16 +26,19 @@ export const providerSteps: Steps =
     });
 
     afterEach(async () => {
-      await fetch('http://' + container.getLaunchpadUrl() + '/stop');
       await OpenFeature.close();
-      return Promise.resolve();
+      await fetch('http://' + container.getLaunchpadUrl() + '/stop');
     }, 50000);
 
     given(/a (.*) flagd provider/, async (providerType: string) => {
       const flagdOptions: FlagdProviderOptions = {
         resolverType: state.resolverType,
-        deadlineMs: 2000,
-        retryGracePeriod: 2,
+        retryGracePeriod: 2, // retryGracePeriod is related to test expectations; this must be 2
+        // these 3 options are optimized for test speed and stability
+        deadlineMs: 4000,
+        retryBackoffMaxMs: 1000,
+        retryBackoffMs: 1000,
+
         ...state.config,
         ...state.options,
       };

libs/providers/flagd/src/lib/service/common/grpc-util.ts

@@ -5,6 +5,11 @@ import type { Config } from '../../configuration';
 import type { Logger } from '@openfeature/server-sdk';
 import { ProviderFatalError } from '@openfeature/server-sdk';
 
+/**
+ * Get the string name of a gRPC status code.
+ */
+const statusName = (code: status): string => status[code];
+
 export const closeStreamIfDefined = (stream: ClientReadableStream<unknown> | undefined) => {
   /**
    * cancel() is necessary to prevent calls from hanging the process, so we need to we need to remove all the
@@ -90,7 +95,7 @@ export const buildRetryPolicy = (serviceName: string, retryBackoffMs?: number, r
           initialBackoff: `${Math.round(initialBackoff / 1000).toFixed(2)}s`,
           maxBackoff: `${Math.round(maxBackoff / 1000).toFixed(2)}s`,
           backoffMultiplier: 2,
-          retryableStatusCodes: [status.UNAVAILABLE, status.UNKNOWN],
+          retryableStatusCodes: [statusName(status.UNAVAILABLE), statusName(status.UNKNOWN)],
         },
       },
     ],