Where communities thrive

  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
Repo info
    Pedro Silva
    hello. As per my understanding 2.0.7 was the last release supporting Scala 2.11, though I have found a bug there with committablePartitionedManualOffsetSource subscriptions. I assume that a fix for a bug raised on 2.0.7 will only be available on Scala 2.12 with the next release 2.1.x ? thank you
    5 replies
    Hello everyone,
    I have a stream that consumes from multiple sources and merge all sources using Source.combine and MergePrioritized. Each source is committableSource. I have a logic for processing each message and then I use Committer.sink to commit the messages.
    When trying to commit I get the following error:
    java.lang.IllegalArgumentException: requirement failed: CommittableOffset [CommittableOffsetImpl(PartitionOffset(GroupTopicPartition(consumer_1,topic_1,1),29826492),)] committer for groupId [group_id_test] must be same as the other with this groupId.
    have anyone encounter this problem?
    3 replies

    I am facing issue with alpakka consumer with plain source, where my consumer stops consuming the messages without any errors printed on the akka-stream. I see the below message being printed:

    2021-03-11 16:50:57,349 INFO  [akka.actor.default-dispatcher-45] org.apache.kafka.clients.consumer.internals.AbstractCoordinator
    [Consumer clientId=consumer-CONSUMER-GRP-2, groupId=CONSUMER-GRP] Attempt to heartbeat failed since group is rebalancing
    2021-03-11 16:52:41,710 INFO  [akka.kafka.default-dispatcher-95] org.apache.kafka.clients.consumer.internals.ConsumerCoordinator
    [Consumer clientId=consumer-CONSUMER_GRP-1, groupId=CONSUMER-GRP] Revoke previously assigned partitions topic-5, topic-4
    2021-03-11 16:52:41,934 INFO  [akka.kafka.default-dispatcher-93] org.apache.kafka.clients.consumer.internals.AbstractCoordinator
    [Consumer clientId=consumer-CONSUMER-GRP-1, groupId=CONSUMER-GRP] Member consumer-CONSUMER-GRP-1-cdfb1446-3781-4605-9e85-739c2fe83439 sending LeaveGroup request to coordinator <IP>:9092 (id: 2147482644 rack: null) due to the consumer is being closed

    I use akka-streams with Graph dsl for my data flow and obtain parallelism. I have a akka cluster with my actors running with clustersingleton to have a timer to invoke actors remotely. I have multiple topics with the same consumer group name, where all the consumer groups get unassigned with the consumer id, and then the stream never consume any message. But on the same process I have a plain java consumer thread without graphDSL which gets re-assigned with the consumer client id.
    Can someone help me on identifying the issue?

    Jesse Yates
    Hiya! Any word on when the next branch will be cut? Some changes we would love to integrate into our system
    3 replies

    I am trying to use Consumer.committablePartitionedSource() and creating stream per partition as shown below

    public void setup() {
            control = Consumer.committablePartitionedSource(consumerSettings,
                    Subscriptions.topics("chat").withPartitionAssignmentHandler(new PartitionAssignmentListener()))
                    .mapAsyncUnordered(Integer.MAX_VALUE, pair -> setupSource(pair, committerSettings))
                    .toMat(Sink.ignore(), Consumer::createDrainingControl)
        private CompletionStage<Done> setupSource(Pair<TopicPartition, Source<ConsumerMessage.CommittableMessage<String, String>, NotUsed>> pair, CommitterSettings committerSettings) {
            LOGGER.info("SETTING UP PARTITION-{} SOURCE", pair.first().partition());
            return pair.second().mapAsync(1, msg -> CompletableFuture.supplyAsync(() -> consumeMessage(msg), actorSystem.dispatcher())
                    .thenApply(param -> msg.committableOffset()))
                    .withAttributes(ActorAttributes.supervisionStrategy(ex -> Supervision.restart()))
                    .runWith(Committer.sink(committerSettings), Materializer.matFromSystem(actorSystem));

    While setting us the source per partition I am using parallelism which I want to change based on no of partitions assigned to the node. That I can do that in the first assignment of partitions to the node. But as new nodes join the cluster assigned partitions are revoked and assigned. This time stream not emitting already existing partitions to reconfigure parallelism.

    What are the options I have to control parallelism on each partitioned source on every rebalancing operation?

    Lior Shapsa
    Hi, Is there a way to read each partition from offset 0 to X and make sure the stream ends once all partitions reach the last offset?
    1 reply
    Sean Glover
    hi. i'm planning to release 2.1.0-RC1 by next Monday. hopefully a final release a week or so after that. thanks for everyone's patience.
    Maksym Besida
    is this an expected behavior for Consumer to reconnect forever if, for example, brokers were misconfigured?
    2 replies
    Nikhil Arora
    I have a use case where I want to commit the offset when I get the success reply from server. The reply from server is async and I don't know the order. How can I handle this ? This blog https://quarkus.io/blog/kafka-commit-strategies/#the-throttled-strategy talks about my use case. I am wondering if there is anything in alpakka kafka for this.
    3 replies
    Sean Glover
    Alpakka Kafka 2.1.0-RC1 has been released to sonatype https://discuss.lightbend.com/t/alpakka-kafka-2-1-0-rc1-released/8144
    Why api of partitioned sources accept only AutoSubscription?
    I need manual partition assignment and I wanted to use partitioned source
    to consume and commit partitions independently.
    Could anyone suggest good solution for this?
    3 replies
    I realized that could write separate consumer for each partition
    Tejas Somani

    I am trying to use Consumer.committableSource with DrainingControl & RestartSource with backoff for my use case based on this https://github.com/akka/alpakka-kafka/blob/master/tests/src/test/scala/docs/scaladsl/ConsumerExample.scala#L501

    val control = new AtomicReference[Consumer.Control](Consumer.NoopControl)
        val result = RestartSource
          .onFailuresWithBackoff(RestartSettings(minBackoff = 3.seconds, maxBackoff = 30.seconds, randomFactor = 0.2)) {
            () =>
                .committableSource(consumerSettings, Subscriptions.topics(topicName))
                .mapMaterializedValue(c => control.set(c))
        val drainingControl = DrainingControl.apply(control.get(), result)

    drainingControl.drainAndShutdown() keeps on running into RTE The correct Consumer.Control has not been assigned, yet.
    Need some help figuring this out. am i missing something?

    2 replies
    Matthew Smedberg
    How would I go about asking my consumer which partitions it is currently assigned (for purposes of an application healthcheck)? I've looked at KafkaConsumerActor and MetadataClient, but it looks like the response to a Metadata.GetPartitionsFor(topic) message/method call is all partitions of the topic, not just the ones that my node is assigned. (Also, it looks like the only consumer strategies that support KafkaConsumerActor require a ManualSubscription, where my use-case requires something like Consumer.committablePartitionedSource.)
    2 replies
    I know about feature with async boundaries akka/alpakka-kafka#1038 and as I understand this is source why in my app messages duplicated and handled in parallel when rebalance started and partition moved to another consumer (also I use manual offset committing which also may be a problem but in eager rebalance we have stop processing phase in which no consumers can process events). I know there are no good solution to guarantee that consumers don't start processing in parallel but what about that solution? see thread
    My goal is at least once semantics but exclude simultaneous parallel handling on different consumers
    7 replies
    Sean Glover
    Alpakka Kafka 2.1.0 final will be released by the end of the week
    Enno Runne
    The 2.1 release is out. Akka 2.6 is now required. https://doc.akka.io/docs/alpakka-kafka/2.1/

    Hello everyone. A little bit of context: I'm consuming an event from an SQS queue. After some further processing, the output of that stream is going to be a file uploaded to an S3 bucket. I'm facing the issue of removing the message from the queue once the file is uploaded.
    This is the sink uploading the file:

    val s3Sink: Sink[ByteString, Future[MultipartUploadResult]] =
        S3.multipartUpload("test", "test.json", ContentTypes.`application/json`)

    This is the flow deleting the message:

    val deleteFlow: Flow[Message, Unit, NotUsed] = {
          .map(_ => ())

    Currently I'm able to save the file in the bucket like this:

    _.asSourceWithContext({ m => m })
          .collect({ case Right(account) => account })
          // some more processing
          .to(s3Sink) // this is the sink previously show 

    And I could also delete the message:

     _.asSourceWithContext({ m => m })
          .collect({ case Right(account) => account })
          // some more processing
          .via(deleteMessageFlow.flow) // this is the flow previously shown

    I cannot figure out how to be able to connect both, ideally, the removal of the message should be done after the upload.


    Hello all!
    So, I'm trying to read a message from an sqs queue:

    val source: Source[Message, NotUsed] = {

    And then save it into an S3 bucket:

    ./// some processing

    This is the s3 sink:

     val s3Sink: Sink[ByteString, Future[MultipartUploadResult]] =
        S3.multipartUpload("bucket", "file.txt", ContentTypes.`text/xml(UTF-8)`)

    If I keep .withCloseOnEmptyReceive(true) in the source config it works. However, I want the source keep listening to the queue always. If I remove it, the future never complete and the file is never uploaded. Any tip on what approach could I use? Is it possible to restart the source?

    BTW, I tried using an actor, once the stream finishes I send a message to the actor to start the stream again. That comes with unexpected behavior.
    Any feedback is appreciated. Thanks!

    Do you know what might be the issues one could face when running a graph inside another graph? Something like:
                v => 
    2 replies

    Hello all,

    we use alpakka kafka in connection with redpanda.
    When we run an integration test with an external dockerized redpanda to test a consumer, we get the following error:


    o.a.kafka.common.utils.AppInfoParser - Error registering AppInfo mbean
    javax.management.InstanceAlreadyExistsException: kafka.producer:type=app-info,id=producer-1
            at java.management/com.sun.jmx.mbeanserver.Repository.addMBean(Repository.java:436)
            at java.management/com.sun.jmx.interceptor.DefaultMBeanServerInterceptor.registerWithRepository(DefaultMBeanServerInterceptor.java:1855)
            at java.management/com.sun.jmx.interceptor.DefaultMBeanServerInterceptor.registerDynamicMBean(DefaultMBeanServerInterceptor.java:955)
            at java.management/com.sun.jmx.interceptor.DefaultMBeanServerInterceptor.registerObject(DefaultMBeanServerInterceptor.java:890)
            at java.management/com.sun.jmx.interceptor.DefaultMBeanServerInterceptor.registerMBean(DefaultMBeanServerInterceptor.java:320)
      | => dat java.management/com.sun.jmx.mbeanserver.JmxMBeanServer.registerMBean(JmxMBeanServer.java:522)
        at org.apache.kafka.common.utils.AppInfoParser.registerAppInfo(AppInfoParser.java:64)
        at org.apache.kafka.clients.producer.KafkaProducer.<init>(KafkaProducer.java:426)
        at org.apache.kafka.clients.producer.KafkaProducer.<init>(KafkaProducer.java:287)
        at akka.kafka.ProducerSettings$.createKafkaProducer(ProducerSettings.scala:215)
        at akka.kafka.ProducerSettings.createKafkaProducer(ProducerSettings.scala:434)


    16:14:35.012 [JobsQueueSpecSystem-akka.kafka.default-dispatcher-16] WARN  o.a.kafka.common.utils.AppInfoParser - Error registering AppInfo mbean
    javax.management.InstanceAlreadyExistsException: kafka.consumer:type=app-info,id=consumer-datapool-1
            at java.management/com.sun.jmx.mbeanserver.Repository.addMBean(Repository.java:436)
            at java.management/com.sun.jmx.interceptor.DefaultMBeanServerInterceptor.registerWithRepository(DefaultMBeanServerInterceptor.java:1855)
            at java.management/com.sun.jmx.interceptor.DefaultMBeanServerInterceptor.registerDynamicMBean(DefaultMBeanServerInterceptor.java:955)
            at java.management/com.sun.jmx.interceptor.DefaultMBeanServerInterceptor.registerObject(DefaultMBeanServerInterceptor.java:890)
            at java.management/com.sun.jmx.interceptor.DefaultMBeanServerInterceptor.registerMBean(DefaultMBeanServerInterceptor.java:320)
      | => dat java.management/com.sun.jmx.mbeanserver.JmxMBeanServer.registerMBean(JmxMBeanServer.java:522)
        at org.apache.kafka.common.utils.AppInfoParser.registerAppInfo(AppInfoParser.java:64)
        at org.apache.kafka.clients.consumer.KafkaConsumer.<init>(KafkaConsumer.java:814)
        at org.apache.kafka.clients.consumer.KafkaConsumer.<init>(KafkaConsumer.java:631)
        at akka.kafka.ConsumerSettings$.createKafkaConsumer(ConsumerSettings.scala:237)
        at akka.kafka.ConsumerSettings$.$anonfun$apply$3(ConsumerSettings.scala:111)
        at akka.kafka.internal.KafkaConsumerActor.akka$kafka$internal$KafkaConsumerActor$$applySettings(KafkaConsumerActor.scala:461)
        at akka.kafka.internal.KafkaConsumerActor.preStart(KafkaConsumerActor.scala:438)
        at akka.actor.Actor.aroundPreStart(Actor.scala:543)
        at akka.actor.Actor.aroundPreStart$(Actor.scala:543)
        at akka.kafka.internal.KafkaConsumerActor.aroundPreStart(KafkaConsumerActor.scala:212)
        at akka.actor.ActorCell.create(ActorCell.scala:637)
        at akka.actor.ActorCell.invokeAll$1(ActorCell.scala:509)
        at akka.actor.ActorCell.systemInvoke(ActorCell.scala:531)

    This is the consumer configuration:

    ConsumerSettings(system, new StringDeserializer, new StringDeserializer)
      ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG -> "true",
      ConsumerConfig.AUTO_OFFSET_RESET_CONFIG  -> "earliest"

    Redpanda v21.6.1
    Akka 2.6.0
    Alpakka Kafka 2.0.6
    Specs2 4.6.0

    Do you have any idea why the consumer and producer are not closed/shutdown after running the integration test ? Thanks!


    Hello everyone.
    So, something I still don't get when using streams is whether this makes sense:

    .map(_.sequence) // doesn't compile just an example

    So the type of the aFunctionThatCallsAnExternalService function is:
    aFunctionThatCallsAnExternalService: Future[Either[Failure, List[ImportantThing]]]
    And say this function returns a huge amount of ImportantThing
    What should I do there, should I put this list back into a stream again? Does it makes sense to use an
    stream and then having this huge list in which I run this (_.sequence) call?

    9 replies

    Hello everyone.
    I am quite new with alpakka-kafka and I have a question
    I need to skip one message from the kafka topic but before in I need to know the current offset
    For skipping I do :

            _ => Future.successful(Map(new TopicPartition(config.topic, config.partition) -> config.offset)))
          .flatMapMerge(1, _._2)
          .map { message =>

    How can I know the current offset before skipping?

    @olgakabirova Did you figure it out? Just switch the order of the take and the map? Also, you seem to only keep the offset in your map, consider passing downstream more than that.
    @olgakabirova You could also consider using a map followed by a filter, where due to some predicate of your choice the filter would happen to only prevent the first message from passing downstream.
    @olgakabirova Yet another alternative could be https://doc.akka.io/docs/akka/current/stream/operators/Source-or-Flow/drop.html depending on what you want to do.
    what are the open source tool is available to monitor/go through kafka messages ?
    Ignasi Marimon-Clos
    🚀 We're pleased to announce the PATCH release of Alpakka Kafka: https://github.com/akka/alpakka-kafka/releases/tag/v2.1.1
    what are the open source tool is available to monitor/go through kafka messages ?
    Lenses.io I think
    But indeed is not free. You can also montior kafka with Grafana, and prometheus.
    And these two have a community edition.
    what are the open source tools available to monitor/go through kafka messages ?

    Hi-I've got an already existing application that is using Alpakka Kafka and is reading from a kafka cluster using a committable source(with offsets committed back to kafka). Now, I'm being asked to read from a second cluster.

    I don't see how it's handled in the Alpakka Kafka 2.1.0 source code, but I don't see anything in the docs against mixing sources from different brokers.

    Is it possible to merge committable sources from different clusters?


    I think I answered my own question: The source cluster seems to be retained as part of the Committer and a KafkaAsyncConsumerCommitterRef.

    sorry for the chatter

    Hi, I am new to the channel. We are using Alpakka Kafka in production and monitor this with Cinnamon en Elastic APM integration. It gives us E2E traceability which is very useful for financial messages ( payments ) handling. However sometimes the the transaction duration has a spike > 60 sec. This is not what really happens as we see the transaction handled in +/- 1 sec. Apparently somewhere the reporting is incorrect. Anybody aware of a kind of 60 sec timeout on reporting metrics ? As the abnormal spikes are each time above 60 sec.
    Alessandro D'Armiento

    Hello, I am using Alpakka-HDFS to save a stream of data in an Orc file on HDFS.
    To do so I first wait for there to be enough messages upstream, then I batch all of them, serialize them in a ORC byte array and then flush the single blob as a file on HDFS.
    This works.

    Now, we decided to drop HDFS in favor of Alluxio, which long story short exposes an Hadoop FileSystem interface but is backed by an object-store.
    After this update, I don't want to use anymore the built-in rename mechanism which (as it makes sense with HDFS) write in a temp directory and then renames the file to have it in the output directory.
    Is it possible?

    Levi Ramsey
    Perhaps it makes sense to use Alpakka-S3 to deal with Alluxio? That should be less likely to assume filesystem semantics for an object store.
    Felipe Oliveira
    hi everyone, I need to consume messages that are older than a few seconds. my current idea is to do offsetsForTimes() + seek(). with alpakka, would i have manage my own offsets? would you suggest any other idea? thank you very much!
    Chetan Kumar
    Hi, I want to understand what happens when a subStream fails in CommittablePartitionedSource, will a new substream automatically start with the failed topic partition? Or will it trigger a rebalance?

    Hi guys, I need some advice on how akka/kafka play with websockets.
    There are akka-http server, kafka topic and websocket clients.
    WS clients from UI connect to the server in order to get the messages from the same topic. They all should get the same messages.
    I implemented this using broadcast hub:

    def kafkaBroadcastHubSource(topic: String): Source[String, NotUsed] = {
        val consumerSettings =
          ConsumerSettings(actorSystem, new IntegerDeserializer, new StringDeserializer)
            .withProperty(ConsumerConfig.ENABLE_AUTO_COMMIT_CONFIG, "true")
            .withProperty(ConsumerConfig.AUTO_COMMIT_INTERVAL_MS_CONFIG, "5000")
        val liveSource = Consumer.plainSource(consumerSettings, Subscriptions.topics(topic))

    Then used this source in a route:

     lazy val source = kafkaBroadcastHubSource(topic)
      val routes: Route = get {
        path("ws") {
          extractWebSocketUpgrade { upgrade =>
            val msgSource = source
              .map(msg => TextMessage(msg))
            complete {
              upgrade.handleMessagesWithSinkSource(Sink.ignore, msgSource)

    As far as I know broabcast hub is aligned by the slowest consumer. Do I understand correctly that if the hub buffer is full because the messages keep arriving and the consumption is slow, then backpressure will be triggered and no data can be processed anymore?
    Will the direct kafka consumer (which is a producer to the boradcast hub) be affected as well and no data will be fetched from the topic?
    What else should I add to make this architecture production ready?
    Thanks in advance!

    6 replies
    Chetan Kumar
    Is there a better way to handle errors inConsumer. committableSource sources, other than restarting the whole consumer.
    How to pass custom context to supervision strategy ? Need the context/payload for logging in supervision strategy.
    Jacob Bogers

    Hi, I am new to akka streams and trying to understand how akka stream .buffer() operator works

    There is a source with buffer:

    val src = Source(1 to 259)
            .buffer(size = 256, OverflowStrategy.dropHead)

    If I connect it to a regular sink:

    val sink = Sink.foreach[Int] { x =>
            println("Element: " + x)

    then seems like the elements don't get into the buffer, so all elements are printed:

    Element: 1
    Element: 2
    Element: 3
    Element: 4

    If I put this source into websocket directive and use it in the test:

    val route = extractWebSocketUpgrade { _ =>
            val msgSource = source
              .buffer(256, OverflowStrategy.dropHead)
              .map { x =>
                println("Element: " + x)
            handleWebSocketMessagesForProtocol(Flow.fromSinkAndSourceCoupled(Sink.ignore, msgSource), jwt)
     val wsClient = WSProbe()
      WS("ws", wsClient.flow) ~> route ~> check {
        isWebSocketUpgrade shouldBe true

    then each element from the source firstly gets into the buffer and then when buffer is full (contains all 256 elements) ,last 3 elements go to the buffer and the 3 oldest are removed (by dropHead strategy):

    Element: 3
    Element: 4
    Element: 5
    Element: 6

    So, the question is - what is the expected behaviour in the first and the second cases?
    Should the elements be enqueued into the buffer in the first case?
    I initially thought that buffer comes into play only when a backpressure is triggered.
    Then, why is it triggered in the second case?

    Levi Ramsey

    Remember that backpressure is basically lack of demand downstream (demand flows from the sink up to the source, generally). A buffer is always signalling demand (with the sole exception of when it's full and the OverflowStrategy is backpressure) to the upstream. Sink.foreach signals demand after processing every element, so it can only backpressure if the processing takes a long time. For the websocket case, demand is ultimately driven by the computer on the other end of the websocket, so it will backpressure more.

    Operator fusion in Akka Streams may also interact here (I forget whether buffer introduces an implied async boundary). But you should definitely be able (assuming you have a dispatcher with multiple threads) to observe the backpressure and dropping behavior with:

    Source(1 to 259).buffer(size = 256, OverflowStrategy.dropHead).async.runWith(Sink.foreach[Int] { x => Thread.sleep(1000); println("Element: " + x) })

    You might also observe the dropping behavior without having the .async there (I haven't experimented).

    Siyavash Habashi
    Hi there, is it planned to release a new akka-stream-kafka version soonish with support for akka 2.6.16? (https://github.com/akka/akka/releases/tag/v2.6.16)
    5 replies
    @leviramsey Thank you, that makes sense. I tried to run your example with .buffer().async and dropping was being observed indeed. It didn't work without async though. I guess it is because all stream pipeline is executed in a single thread
    Dominik Kunz
    Hi, I have a question about the DiscoverySupport using akka-dns method. I did not find a way to specify a port name for the lookup. So, is there a way to set a port name for the lookup. Because using only the service name, only an IP lookup is done which of course deos not include the port of the service.
    Hi I have a quick question concerning alpakka support for the underlying Idempotent Producer of kafka.
    I can see that Transaction is handled at alpakka level, but i do not see if i can use the idempotent producer with alpakka i.e. idempotence.enable=true ? Does it require either to be transactional or not and no in between with the idempotent producer enabled but at the kafka level ?
    5 replies