detect when kafkajs is stuck with stale broker metadata (#186)

* detect when kafkajs is stuck with stale broker metadata and terminate application * lint
walmartlabs · Sep 28, 2020 · 5001258 · 5001258
1 parent ffb5ecb
commit 5001258
Show file tree

Hide file tree

Showing 2 changed files with 24 additions and 1 deletion.
diff --git a/packages/kafka/package.json b/packages/kafka/package.json
@@ -1,6 +1,6 @@
 {
     "name": "@walmartlabs/cookie-cutter-kafka",
-    "version": "1.3.0-beta.3",
+    "version": "1.3.0-beta.4",
     "license": "Apache-2.0",
     "main": "dist/index.js",
     "types": "dist/index.d.ts",

diff --git a/packages/kafka/src/KafkaConsumer.ts b/packages/kafka/src/KafkaConsumer.ts
@@ -79,6 +79,7 @@ export class KafkaConsumer implements IRequireInitialization, IDisposable {
     private offsetCommitIntervalMs: number;
     private timer: NodeJS.Timer;
     private groupEpoch: number = 0;
+    private brokerMetadataErrors: number = 0;
 
     constructor(config: KafkaConsumerConfig) {
         this.config = config;
@@ -278,6 +279,7 @@ export class KafkaConsumer implements IRequireInitialization, IDisposable {
                 // tslint:disable-next-line:no-floating-promises
                 this.commitOffsetsIfNecessary();
             }, 0);
+            this.timer.unref();
         }
 
         await this.consumer.run({
@@ -407,15 +409,36 @@ export class KafkaConsumer implements IRequireInitialization, IDisposable {
                             this.metrics.gauge(KafkaMetrics.Lag, lag.toNumber(), tags);
                         }
                     }
+
+                    this.brokerMetadataErrors = 0;
                 } catch (e) {
                     this.logger.warn("Unable to retrieve watermarks", e);
+
+                    // this is a workaround for https://github.com/walmartlabs/cookie-cutter/issues/185
+                    // until a fix for kafkajs is available / the root cause of the problem is confirmed
+                    if (
+                        e &&
+                        e.toString().contains("server is not the leader for that topic-partition")
+                    ) {
+                        this.brokerMetadataErrors++;
+                    }
+
+                    if (this.brokerMetadataErrors > 10) {
+                        this.logger.error("detected stale broker metadata in kafkajs");
+                        this.brokerMetadataErrors = 0;
+
+                        // throwing an error here will cause a UnhandledPromiseException
+                        // and terminate the application
+                        throw new Error("stale broker metadata");
+                    }
                 }
             }
         } finally {
             this.timer = setTimeout(() => {
                 // tslint:disable-next-line:no-floating-promises
                 this.commitOffsetsIfNecessary();
             }, this.offsetCommitIntervalMs);
+            this.timer.unref();
         }
     }