RFR: 8308995: Update Network IO JFR events to be static mirror events

Erik Gahlin egahlin at openjdk.org
Thu Jun 22 07:52:08 UTC 2023


On Tue, 6 Jun 2023 19:39:31 GMT, Tim Prinzing <tprinzing at openjdk.org> wrote:

> The socket read/write JFR events currently use instrumentation of java.base code using templates in the jdk.jfr modules. This results in some java.base code residing in the jdk.jfr module which is undesirable.
> 
> JDK19 added static support for event classes. The old instrumentor classes should be replaced with mirror events using the static support.
> 
> In the java.base module:
> Added two new events, jdk.internal.event.SocketReadEvent and jdk.internal.event.SocketWriteEvent.
> java.net.Socket and sun.nio.ch.SocketChannelImpl were changed to make use of the new events.
> 
> In the jdk.jfr module:
> jdk.jfr.events.SocketReadEvent and jdk.jfr.events.SocketWriteEvent were changed to be mirror events.
> In the package jdk.jfr.internal.instrument, the classes SocketChannelImplInstrumentor, SocketInputStreamInstrumentor, and SocketOutputStreamInstrumentor were removed. The JDKEvents class was updated to reflect all of those changes.
> 
> The existing tests in test/jdk/jdk/jfr/event/io continue to pass with the new implementation:
> Passed: jdk/jfr/event/io/TestSocketChannelEvents.java
> Passed: jdk/jfr/event/io/TestSocketEvents.java
> 
> I added a micro benchmark which measures the overhead of handling the jfr socket events.
> test/micro/org/openjdk/bench/java/net/SocketEventOverhead.java.
> It needs access the jdk.internal.event package, which is done at runtime with annotations that add the extra arguments.
> At compile time the build arguments had to be augmented in make/test/BuildMicrobenchmark.gmk

I believe the fields SOCKET_READ and SOCKET_WRITE in the jdk.jfr.events.EventConfigurations class can be removed.

-------------

PR Comment: https://git.openjdk.org/jdk/pull/14342#issuecomment-1602173783


More information about the core-libs-dev mailing list