Merge branch 'bugfix/BB-624-connection-retry' into tmp/octopus/w/7.70…

…/bugfix/BB-624-connection-retry
scality · Nov 6, 2024 · 35fb11e · 35fb11e
2 parents 57fc235 + af67f46
commit 35fb11e
Show file tree

Hide file tree

Showing 20 changed files with 169 additions and 61 deletions.
diff --git a/.dockerignore b/.dockerignore
@@ -0,0 +1,20 @@
+node_modules
+localData/*
+localMetadata/*
+.git
+.github
+.tox
+coverage
+.DS_Store
+
+.dockerignore
+docs/
+images/
+res/
+_config.yml
+.eslintrc
+.gitignore
+circle.yml
+DESIGN.md
+README.md
+Using.md
diff --git a/bin/queuePopulator.js b/bin/queuePopulator.js
@@ -52,6 +52,7 @@ const queuePopulator = new QueuePopulator(
     zkConfig, kafkaConfig, qpConfig, httpsConfig, mConfig, rConfig, extConfigs);
 
 async.waterfall([
+    done => queuePopulator.open(done),
     done => startProbeServer(qpConfig.probeServer, (err, probeServer) => {
         if (err) {
             log.error('error starting probe server', {
@@ -73,7 +74,6 @@ async.waterfall([
         }
         done();
     }),
-    done => queuePopulator.open(done),
     done => {
         const taskState = {
             batchInProgress: false,

diff --git a/extensions/notification/destination/KafkaNotificationDestination.js b/extensions/notification/destination/KafkaNotificationDestination.js
@@ -48,8 +48,9 @@ class KafkaNotificationDestination extends NotificationDestination {
                 this._log.info('error setting up kafka notif destination',
                     { error: err.message });
                 done(err);
+            } else {
+                done();
             }
-            done();
         });
     }
 

diff --git a/extensions/notification/destination/KafkaProducer.js b/extensions/notification/destination/KafkaProducer.js
@@ -72,12 +72,13 @@ class KafkaProducer extends EventEmitter {
             'request.timeout.ms': ACK_TIMEOUT,
         });
         this._ready = false;
-        this._producer.connect({}, error => {
+        this._producer.connect({ timeout: 60000 }, error => {
             if (error) {
                 this._log.info('error connecting to broker', {
                     error,
                     method: 'KafkaProducer.constructor',
                 });
+                this.emit('error', error);
             }
         });
         this._producer.on('ready', () => {

diff --git a/extensions/notification/queueProcessor/QueueProcessor.js b/extensions/notification/queueProcessor/QueueProcessor.js
@@ -116,17 +116,20 @@ class QueueProcessor extends EventEmitter {
      * @param {boolean} [options.disableConsumer] - true to disable
      *   startup of consumer (for testing: one has to call
      *   processQueueEntry() explicitly)
+     * @param {function} done callback
      * @return {undefined}
      */
-    start(options) {
+    start(options, done) {
         async.series([
             next => this._setupZookeeper(next),
             next => this._setupNotificationConfigManager(next),
             next => this._setupDestination(this.destinationConfig.type, next),
-            next => this._destination.init(() => {
+            // if connection to destination fails, process will stop & restart
+            next => this._destination.init(next),
+            next => {
                 if (options && options.disableConsumer) {
                     this.emit('ready');
-                    return undefined;
+                    return next();
                 }
                 const { groupId, concurrency, logConsumerMetricsIntervalS }
                     = this.notifConfig.queueProcessor;
@@ -142,22 +145,31 @@ class QueueProcessor extends EventEmitter {
                     queueProcessor: this.processKafkaEntry.bind(this),
                     logConsumerMetricsIntervalS,
                 });
-                this._consumer.on('error', () => { });
+                this._consumer.on('error', err => {
+                    this.logger.error('error starting notification consumer',
+                    { method: 'QueueProcessor.start', error: err.message });
+                    // crash if got error at startup
+                    if (!this.isReady()) {
+                        return next(err);
+                    }
+                    return undefined;
+                });
                 this._consumer.on('ready', () => {
                     this._consumer.subscribe();
                     this.logger.info('queue processor is ready to consume ' +
                         'notification entries');
                     this.emit('ready');
+                    return next();
                 });
-                return next();
-            }),
+                return undefined;
+            },
         ], err => {
             if (err) {
                 this.logger.info('error starting notification queue processor',
                     { error: err.message });
-                return undefined;
+                return done(err);
             }
-            return undefined;
+            return done();
         });
     }
 
@@ -263,6 +275,15 @@ class QueueProcessor extends EventEmitter {
             return done();
         }
     }
+
+    /**
+     * Checks if queue processor is ready to consume
+     *
+     * @returns {boolean} is queue processor ready
+     */
+    isReady() {
+        return this._consumer && this._consumer.isReady();
+    }
 }
 
 module.exports = QueueProcessor;
diff --git a/extensions/notification/queueProcessor/task.js b/extensions/notification/queueProcessor/task.js
@@ -24,7 +24,11 @@ try {
         'could not be found in destinations defined');
     const queueProcessor = new QueueProcessor(
         zkConfig, kafkaConfig, notifConfig, destinationConfig, destination);
-    queueProcessor.start();
+    queueProcessor.start(undefined, err => {
+        if (err) {
+            process.exit(1);
+        }
+    });
 } catch (err) {
     log.error('error starting notification queue processor task', {
         method: 'notification.task.queueProcessor.start',

diff --git a/extensions/replication/failedCRR/FailedCRRConsumer.js b/extensions/replication/failedCRR/FailedCRRConsumer.js
@@ -35,6 +35,7 @@ class FailedCRRConsumer {
      * @return {undefined}
      */
     start(cb) {
+        let consumerReady = false;
         const consumer = new BackbeatConsumer({
             kafka: {
                 hosts: this._kafkaConfig.hosts,
@@ -46,8 +47,17 @@ class FailedCRRConsumer {
             queueProcessor: this.processKafkaEntry.bind(this),
             fetchMaxBytes: CONSUMER_FETCH_MAX_BYTES,
         });
-        consumer.on('error', () => {});
+        consumer.on('error', err => {
+            if (!consumerReady) {
+                this.logger.fatal('could not setup a backbeat consumer', {
+                    method: 'FailedCRRConsumer.start',
+                    error: err,
+                });
+                process.exit(1);
+            }
+        });
         consumer.on('ready', () => {
+            consumerReady = true;
             consumer.subscribe();
             this.logger.info('retry consumer is ready to consume entries');
         });

diff --git a/extensions/replication/failedCRR/FailedCRRProducer.js b/extensions/replication/failedCRR/FailedCRRProducer.js
@@ -29,7 +29,7 @@ class FailedCRRProducer {
             kafka: { hosts: this._kafkaConfig.hosts },
             topic: this._topic,
         });
-        this._producer.once('error', () => {});
+        this._producer.once('error', cb);
         this._producer.once('ready', () => {
             this._producer.removeAllListeners('error');
             this._producer.on('error', err =>

diff --git a/extensions/replication/queueProcessor/QueueProcessor.js b/extensions/replication/queueProcessor/QueueProcessor.js
@@ -401,6 +401,7 @@ class QueueProcessor extends EventEmitter {
      */
     start(options) {
         this._setupProducer(err => {
+            let consumerReady = false;
             if (err) {
                 this.logger.info('error setting up kafka producer',
                     { error: err.message });
@@ -424,8 +425,15 @@ class QueueProcessor extends EventEmitter {
                 logConsumerMetricsIntervalS: this.repConfig.queueProcessor.logConsumerMetricsIntervalS,
                 circuitBreaker: this.circuitBreakerConfig,
             });
-            this._consumer.on('error', () => { });
+            this._consumer.on('error', () => {
+                if (!consumerReady) {
+                    this.logger.fatal('queue processor failed to start a ' +
+                                   'backbeat consumer');
+                    process.exit(1);
+                }
+            });
             this._consumer.on('ready', () => {
+                consumerReady = true;
                 this._consumer.subscribe();
                 this.logger.info('queue processor is ready to consume ' +
                     `replication entries from ${this.topic}`);
@@ -571,13 +579,15 @@ class QueueProcessor extends EventEmitter {
 
         // consumer stats lag is on a different update cycle so we need to
         // update the metrics when requested
-        const lagStats = this._consumer.consumerStats.lag;
-        Object.keys(lagStats).forEach(partition => {
-            metricsHandler.lag({
-                partition,
-                serviceName: this.serviceName,
-            }, lagStats[partition]);
-        });
+        if (this._consumer) {
+            const lagStats = this._consumer.consumerStats.lag;
+            Object.keys(lagStats).forEach(partition => {
+                metricsHandler.lag({
+                    partition,
+                    serviceName: this.serviceName,
+                }, lagStats[partition]);
+            });
+        }
         const metrics = await promClient.register.metrics();
 
         res.writeHead(200, {

diff --git a/extensions/replication/queueProcessor/task.js b/extensions/replication/queueProcessor/task.js
@@ -56,6 +56,21 @@ function getProbeConfig(queueProcessorConfig, site) {
 }
 
 async.waterfall([
+    done => {
+        metricsProducer.setupProducer(err => {
+            if (err) {
+                log.error('error starting metrics producer for queue processor', {
+                    error: err,
+                    method: 'MetricsProducer::setupProducer',
+                });
+            }
+            done(err);
+        });
+    },
+    done => {
+        queueProcessor.on('ready', done);
+        queueProcessor.start();
+    },
     done => startProbeServer(
         getProbeConfig(repConfig.queueProcessor, site),
         (err, probeServer) => {
@@ -80,21 +95,6 @@ async.waterfall([
             done();
         }
     ),
-    done => {
-        metricsProducer.setupProducer(err => {
-            if (err) {
-                log.error('error starting metrics producer for queue processor', {
-                    error: err,
-                    method: 'MetricsProducer::setupProducer',
-                });
-            }
-            done(err);
-        });
-    },
-    done => {
-        queueProcessor.on('ready', done);
-        queueProcessor.start();
-    },
 ], err => {
     if (err) {
         log.error('error during queue processor initialization', {

diff --git a/extensions/replication/replay/ReplayProducer.js b/extensions/replication/replay/ReplayProducer.js
@@ -29,7 +29,7 @@ class ReplayProducer {
             kafka: { hosts: this._kafkaConfig.hosts },
             topic: this._topic,
         });
-        this._producer.once('error', () => {});
+        this._producer.once('error', cb);
         this._producer.once('ready', () => {
             this._producer.removeAllListeners('error');
             this._producer.on('error', err =>

diff --git a/extensions/replication/replayProcessor/task.js b/extensions/replication/replayProcessor/task.js
@@ -63,6 +63,21 @@ function getProbeConfig(replayProcessorConfig, site, topicName) {
 }
 
 async.waterfall([
+    done => {
+        metricsProducer.setupProducer(err => {
+            if (err) {
+                log.error('error starting metrics producer for queue processor', {
+                    error: err,
+                    method: 'MetricsProducer::setupProducer',
+                });
+            }
+            done(err);
+        });
+    },
+    done => {
+        queueProcessor.on('ready', done);
+        queueProcessor.start();
+    },
     done => startProbeServer(
         getProbeConfig(repConfig.replayProcessor, site, topic),
         (err, probeServer) => {
@@ -87,21 +102,6 @@ async.waterfall([
             done();
         }
     ),
-    done => {
-        metricsProducer.setupProducer(err => {
-            if (err) {
-                log.error('error starting metrics producer for queue processor', {
-                    error: err,
-                    method: 'MetricsProducer::setupProducer',
-                });
-            }
-            done(err);
-        });
-    },
-    done => {
-        queueProcessor.on('ready', done);
-        queueProcessor.start();
-    },
 ], err => {
     if (err) {
         log.error('error during queue processor initialization', {

diff --git a/extensions/replication/replicationStatusProcessor/ReplicationStatusProcessor.js b/extensions/replication/replicationStatusProcessor/ReplicationStatusProcessor.js
@@ -315,6 +315,7 @@ class ReplicationStatusProcessor {
      * @return {undefined}
      */
     start(options, cb) {
+        let consumerReady = false;
         this._FailedCRRProducer = new FailedCRRProducer(this.kafkaConfig);
         this._replayTopicNames.forEach(t => {
             this._ReplayProducers[t] = new ReplayProducer(this.kafkaConfig, t);
@@ -332,8 +333,14 @@ class ReplicationStatusProcessor {
             bootstrap: options && options.bootstrap,
             logConsumerMetricsIntervalS: this.repConfig.replicationStatusProcessor.logConsumerMetricsIntervalS,
         });
-        this._consumer.on('error', () => { });
+        this._consumer.on('error', () => {
+            if (!consumerReady) {
+                this.logger.fatal('error starting a backbeat consumer');
+                process.exit(1);
+            }
+        });
         this._consumer.on('ready', () => {
+            consumerReady = true;
             this.logger.info('replication status processor is ready to ' +
                 'consume replication status entries');
             this._consumer.subscribe();

diff --git a/extensions/replication/replicationStatusProcessor/task.js b/extensions/replication/replicationStatusProcessor/task.js
@@ -25,6 +25,7 @@ werelogs.configure({
 });
 
 async.waterfall([
+    done => replicationStatusProcessor.start(undefined, done),
     done => startProbeServer(
         repConfig.replicationStatusProcessor.probeServer,
         (err, probeServer) => {
@@ -49,7 +50,6 @@ async.waterfall([
             done();
         }
     ),
-    done => replicationStatusProcessor.start(undefined, done),
 ], err => {
     if (err) {
         log.error('error during queue processor initialization', {