jerqi commented on issue #2082: URL: https://github.com/apache/incubator-uniffle/issues/2082#issuecomment-2655745277
> I also hit the same issue. In my case, readBuffer is an instance of DirectByteBuffer and readBuffer#cleaner is null and readBuffer#attachment is not null. Since readBuffer#cleaner is null, PlatformDependent.freeDirectBuffer which eventually invokes cleaner won't do anything regardless, plus before reaching cleaner, it's already [error out when checking attachment](https://github.com/openjdk/jdk/blob/jdk-11-ga/src/jdk.unsupported/share/classes/sun/misc/Unsafe.java#L1238). The readBuffer we get is from [DirectByteBuffer.duplicate](https://github.com/ydq/OpenJDK-11/blob/master/src/java/nio/DirectByteBuffer.java#L168), which sets cleaner to null intentionally. It seems readBuffer is just a pointer to the base buffer, which would [cleanup anyway](https://github.com/apache/incubator-uniffle/blob/master/common/src/main/java/org/apache/uniffle/common/netty/TransportFrameDecoder.java#L202). It seems we can simply don't release readBuffer in ShuffleReadClientImpl. I tested my application with that and seems run successfully and didn't see memory issue as well. I think it's ok to remove the logic of `release read buffer`. Do you try to use `-Dio.netty.leakDetectionLevel=` to ensure that the client don't have memory issues? -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
