当您在 Kafka 消费者/客户端上收到“太多打开的文件”时，这意味着什么？

Question

I know that this usually means the ulimit needs to be increased. But what does this actually mean when it happens on the consumer side?

I'm using Apache Flink and I got this error on my Flink task node. When I reboot my Flink node and redeployed the job it worked fine. The brokers also seemed fine at the time.

I have a total of 9 tasks running over 3 nodes. Max parallelism for any one job is 1 to 2. So lets assume worst case 18 parallelism/threads over 3 nodes.

org.apache.kafka.common.KafkaException: Failed to construct kafka consumer
at org.apache.kafka.clients.consumer.KafkaConsumer.<init>(KafkaConsumer.java:799)
at org.apache.kafka.clients.consumer.KafkaConsumer.<init>(KafkaConsumer.java:650)
at org.apache.kafka.clients.consumer.KafkaConsumer.<init>(KafkaConsumer.java:630)
at org.apache.flink.streaming.connectors.kafka.internal.KafkaPartitionDiscoverer.initializeConnections(KafkaPartitionDiscoverer.java:58)
at org.apache.flink.streaming.connectors.kafka.internals.AbstractPartitionDiscoverer.open(AbstractPartitionDiscoverer.java:94)
at org.apache.flink.streaming.connectors.kafka.FlinkKafkaConsumerBase.open(FlinkKafkaConsumerBase.java:504)
at org.apache.flink.api.common.functions.util.FunctionUtils.openFunction(FunctionUtils.java:36)
at org.apache.flink.streaming.api.operators.AbstractUdfStreamOperator.open(AbstractUdfStreamOperator.java:102)
at org.apache.flink.streaming.runtime.tasks.StreamTask.openAllOperators(StreamTask.java:424)
at org.apache.flink.streaming.runtime.tasks.StreamTask.invoke(StreamTask.java:290)
at org.apache.flink.runtime.taskmanager.Task.run(Task.java:711)
at java.lang.Thread.run(Thread.java:748)
Caused by: org.apache.kafka.common.KafkaException: java.io.IOException: Too many open files
at org.apache.kafka.common.network.Selector.<init>(Selector.java:154)
at org.apache.kafka.common.network.Selector.<init>(Selector.java:188)
at org.apache.kafka.common.network.Selector.<init>(Selector.java:192)
at org.apache.kafka.clients.consumer.KafkaConsumer.<init>(KafkaConsumer.java:722)
... 11 more
Caused by: java.io.IOException: Too many open files
at sun.nio.ch.IOUtil.makePipe(Native Method)
at sun.nio.ch.EPollSelectorImpl.<init>(EPollSelectorImpl.java:65)
at sun.nio.ch.EPollSelectorProvider.openSelector(EPollSelectorProvider.java:36)
at java.nio.channels.Selector.open(Selector.java:227)
at org.apache.kafka.common.network.Selector.<init>(Selector.java:152)
... 14 more

Answer 1

Every Kafka client (producer, consumer) maintains a single socket per every broker in the cluster its connected to (worst case).

so youre looking at number of clients flink creates times number of brokers in your cluster

sockets count as handles for purposes of ulimit.

I dont know how many kafka clients flink creates internally - you could grab a heap dump and see how many client objects are in there