[FLINK-24302] Direct buffer memory leak on Pulsar connector with Java 11

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Workaround
• Affects Version/s: 1.14.0
• Fix Version/s: pulsar-4.1.0
• Component/s: Connectors / Pulsar
• Labels: pull-request-available test-stability
• Assignee: Zili Chen
• Reporter: Yufan Sheng
• Created: 15/Sep/21 17:34
• Updated: 24/Jul/23 07:05
• Resolved: 20/Jul/23 10:23

Description

Running the Pulsar connector with multiple split readers on Java 11 could throw a java.lang.OutOfMemoryError exception.

Caused by: java.util.concurrent.CompletionException: org.apache.pulsar.client.admin.internal.http.AsyncHttpConnector$RetryException: Could not complete the operation. Number of retries has been exhausted. Failed reason: java.lang.OutOfMemoryError: Direct buffer memory
	at java.base/java.util.concurrent.CompletableFuture.encodeThrowable(CompletableFuture.java:331)
	at java.base/java.util.concurrent.CompletableFuture.completeThrowable(CompletableFuture.java:346)
	at java.base/java.util.concurrent.CompletableFuture$OrApply.tryFire(CompletableFuture.java:1503)
	... 42 more
Caused by: org.apache.pulsar.client.admin.internal.http.AsyncHttpConnector$RetryException: Could not complete the operation. Number of retries has been exhausted. Failed reason: java.lang.OutOfMemoryError: Direct buffer memory
	at org.apache.pulsar.client.admin.internal.http.AsyncHttpConnector.lambda$retryOperation$4(AsyncHttpConnector.java:249)
	... 39 more
Caused by: org.apache.pulsar.shade.io.netty.handler.codec.EncoderException: java.lang.OutOfMemoryError: Direct buffer memory
	at org.apache.pulsar.shade.io.netty.handler.codec.MessageToMessageEncoder.write(MessageToMessageEncoder.java:104)
	at org.apache.pulsar.shade.io.netty.channel.CombinedChannelDuplexHandler.write(CombinedChannelDuplexHandler.java:346)
	at org.apache.pulsar.shade.io.netty.channel.AbstractChannelHandlerContext.invokeWrite0(AbstractChannelHandlerContext.java:717)
	at org.apache.pulsar.shade.io.netty.channel.AbstractChannelHandlerContext.invokeWrite(AbstractChannelHandlerContext.java:709)
	at org.apache.pulsar.shade.io.netty.channel.AbstractChannelHandlerContext.write(AbstractChannelHandlerContext.java:792)
	at org.apache.pulsar.shade.io.netty.channel.AbstractChannelHandlerContext.write(AbstractChannelHandlerContext.java:702)
	at org.apache.pulsar.shade.io.netty.handler.stream.ChunkedWriteHandler.doFlush(ChunkedWriteHandler.java:303)
	at org.apache.pulsar.shade.io.netty.handler.stream.ChunkedWriteHandler.flush(ChunkedWriteHandler.java:132)
	at org.apache.pulsar.shade.io.netty.channel.AbstractChannelHandlerContext.invokeFlush0(AbstractChannelHandlerContext.java:750)
	at org.apache.pulsar.shade.io.netty.channel.AbstractChannelHandlerContext.invokeWriteAndFlush(AbstractChannelHandlerContext.java:765)
	at org.apache.pulsar.shade.io.netty.channel.AbstractChannelHandlerContext.write(AbstractChannelHandlerContext.java:790)
	at org.apache.pulsar.shade.io.netty.channel.AbstractChannelHandlerContext.writeAndFlush(AbstractChannelHandlerContext.java:758)
	at org.apache.pulsar.shade.io.netty.channel.DefaultChannelPipeline.writeAndFlush(DefaultChannelPipeline.java:1020)
	at org.apache.pulsar.shade.io.netty.channel.AbstractChannel.writeAndFlush(AbstractChannel.java:311)
	at org.apache.pulsar.shade.org.asynchttpclient.netty.request.NettyRequestSender.writeRequest(NettyRequestSender.java:420)
	... 23 more

The reason is that under Java 11, the Netty will allocate memory from the pool of Java Direct Memory and is affected by the MaxDirectMemory limit. Under Java 8, it allocates native memory and is not affected by that setting.

We have to reduce the direct memory usage by using a newer Pulsar client which has a memory-limits configuration.

This issue is addressed on Pulsar, and PIP-74 has been created for resolving this issue.

We should keep this issue open with no resolved versions until Pulsar provides a new client with memory limits.

Update: 2022/08/04

The memory limit on consumer API has been released https://github.com/apache/pulsar/pull/15216, we need to add autoScaledReceiverQueueSizeEnabled option to enable this feature. This memory limit will get released on Pulsar 2.11.0. We will get this fixed after that.

Update: 2023/02/13

Pulsar has a client memory allocator configuration which can drop the use of direct buffer memory.

Issue Links

• causes: FLINK-29495 PulsarSinkE2ECase hang
• is fixed by: FLINK-30606 Bump Pulsar to 2.11.0
• links to: GitHub Pull Request #27
• links to: GitHub Pull Request #55