public async Task PlainSource_consumes_messages_from_KafkaProducer_with_topicPartitionOffset_assignment()
{
int elementsCount = 100;
int offset = 50;
var topic1 = CreateTopic(1);
var group1 = CreateGroup(1);
await GivenInitializedTopic(topic1);
await Produce(topic1, Enumerable.Range(1, elementsCount), ProducerSettings);
var consumerSettings = CreateConsumerSettings(group1);
var probe = CreateProbe(consumerSettings, topic1, Subscriptions.AssignmentWithOffset(new TopicPartitionOffset(topic1, 0, new Offset(offset))));
probe.Request(elementsCount);
foreach (var i in Enumerable.Range(offset, elementsCount - offset).Select(c => c.ToString()))
{
probe.ExpectNext(i, TimeSpan.FromSeconds(10));
}
probe.Cancel();
}
public async Task Committable_consumer_with_failed_downstream_stage_result_should_be_gapless()
{
var topic = CreateTopic(1);
var group = CreateGroup(1);
var topicPartition = new TopicPartition(topic, 0);
var consumerSettings = ConsumerSettings <Null, string> .Create(Sys, null, null)
.WithBootstrapServers(Fixture.KafkaServer)
.WithStopTimeout(TimeSpan.FromSeconds(1))
.WithProperty("auto.offset.reset", "earliest")
.WithGroupId(group);
var counter = 0;
await Source.From(Enumerable.Range(1, 11))
.Select(elem => new ProducerRecord <Null, string>(topicPartition, elem.ToString()))
.RunWith(KafkaProducer.PlainSink(ProducerSettings), Materializer);
var probe = KafkaConsumer.CommittableSource(consumerSettings, Subscriptions.AssignmentWithOffset(new TopicPartitionOffset(topicPartition, Offset.Unset)))
.Select(t =>
{
counter++;
// fail once, on the 7th message
if (counter == 7)
{
throw new Exception("BOOM!");
}
return(t);
})
.SelectAsync(1, async elem =>
{
await elem.CommitableOffset.Commit();
return(elem.Record.Value);
})
.ToMaterialized(this.SinkProbe <string>(), Keep.Right)
.Run(Materializer);
var messages = new List <string>();
probe.Request(11);
for (var i = 0; i < 6; i++)
{
messages.Add(probe.ExpectNext(TimeSpan.FromSeconds(5)));
}
// stream fails at index 7
var err = probe.ExpectEvent();
err.Should().BeOfType <TestSubscriber.OnError>();
var exception = ((TestSubscriber.OnError)err).Cause;
exception.Message.Should().Be("BOOM!");
// stream should be dead here
probe.ExpectNoMsg(TimeSpan.FromSeconds(5));
probe.Cancel();
// restart dead stream
probe = KafkaConsumer.CommittableSource(consumerSettings, Subscriptions.AssignmentWithOffset(new TopicPartitionOffset(topicPartition, Offset.Unset)))
.SelectAsync(1, async elem =>
{
await elem.CommitableOffset.Commit();
return(elem.Record.Value);
})
.ToMaterialized(this.SinkProbe <string>(), Keep.Right)
.Run(Materializer);
probe.Request(11);
for (var i = 0; i < 5; i++)
{
messages.Add(probe.ExpectNext(TimeSpan.FromSeconds(5)));
}
probe.Cancel();
// end result should be gapless
messages.Select(s => int.Parse(s)).Should().BeEquivalentTo(Enumerable.Range(1, 11));
}
private void StartConsuming()
{
#if DEBUG
Console.WriteLine("Start Consuming");
#endif
allEofsFound = new TaskCompletionSource <bool>();
//use eof to know when we are ready with a single topic partition
var consumerSettings = ConsumerSettings <string, byte[]>
.Create(settings.KafkaConfig, Deserializers.Utf8, Deserializers.ByteArray)
.WithBootstrapServers(settings.KafkaConfig.GetString("bootstrap.servers"))
.WithGroupId(settings.KafkaConfig.GetString("groupid.prefix") + Guid.NewGuid())
.WithProperty("enable.partition.eof", "true");
var adminClientBuilder = new AdminClientBuilder(consumerSettings.Properties);
Metadata metadata = null;
using (var adminClient = adminClientBuilder.Build())
{
metadata = adminClient.GetMetadata(TimeSpan.FromSeconds(10));
}
if (metadata == null)
{
throw new Exception("can not retrieve metadata from bootstrap servers");
}
var knownTopicPartitions = new HashSet <TopicPartition>();
foreach (var metadataTopic in metadata.Topics)
{
foreach (var metadataTopicPartition in metadataTopic.Partitions)
{
knownTopicPartitions.Add(new TopicPartition(metadataTopic.Topic, metadataTopicPartition.PartitionId));
}
}
var topicsNotInKafka = 0;
//create a consumer for all enabled partition topics
var tposForSubscription = new List <TopicPartitionOffset>();
foreach (var tp in enabledTopicPartitions)
{
if (knownTopicPartitions.Contains(tp))
{
tposForSubscription.Add(currentOffsets.TryGetValue(tp, out long offset)
? new TopicPartitionOffset(tp, offset + 1)
: new TopicPartitionOffset(tp, Offset.Beginning));
}
else
{
//if not present in kafka, then we know we are at the end
eofsFound.Add(tp);
topicsNotInKafka++;
}
}
if (tposForSubscription.Count == 0)
{
//empty start
allEofsFound.SetResult(true);
return;
}
var subscription = Subscriptions.AssignmentWithOffset(tposForSubscription.ToArray());
var source = KafkaConsumer.PlainSource(consumerSettings, subscription);
var writeBatches = new Dictionary <TopicPartition, WriteBatch>();
eofsFound = new HashSet <TopicPartition>();
//normally this won't take long. The topicpartition progress is also stored in rocksdb
//if we need to reread all events then this can take a while
var cts = new CancellationTokenSource();
var sourceTask = source
.Via(cts.Token.AsFlow <ConsumeResult <string, byte[]> >(true))
.RunForeach(msg =>
{
if (msg.IsPartitionEOF)
{
//save as this is during startup; no-one can touch this topic partition yet.
currentOffsets[msg.TopicPartition] = msg.Offset;
eofsFound.Add(msg.TopicPartition);
if (writeBatches.TryGetValue(msg.TopicPartition, out var b))
{
database.Write(b, rocksDbWriteOptions);
b.Dispose();
writeBatches.Remove(msg.TopicPartition);
}
if (eofListeners.TryGetValue(msg.TopicPartition, out var listeners))
{
foreach (var taskCompletionSource in listeners)
{
taskCompletionSource.SetResult(true);
}
eofListeners.Remove(msg.TopicPartition);
}
if (eofsFound.Count != tposForSubscription.Count + topicsNotInKafka)
{
return;
}
allEofsFound.SetResult(true);
cts.Cancel();
cts.Dispose();
return;
}
if (eofsFound.Contains(msg.TopicPartition))
{
return;
}
// presume we are the only one writing to this topic partition (otherwise akka persistence gets messy real quick)
if (!writeBatches.TryGetValue(msg.TopicPartition, out var writebatch))
{
writeBatches[msg.TopicPartition] = writebatch = new WriteBatch();
}
//add event to the writebatch
var persistent = PersistentFromMessage(msg);
WriteToRocksDbBatch(persistent, writebatch);
//store current offset into (using the writebatch)
var key = TopicPartitionKey(TopicPartitionNumericId(msg.TopicPartition));
writebatch.Put(KeyToBytes(key), CounterToBytes(msg.Offset));
}, materializer);
sourceTask.ContinueWith(x =>
{
Console.WriteLine("Source stopped for kafka journal");
});
}
