JEP 328 Flight Recorder

[pshipton]

https://bugs.openjdk.java.net/browse/JDK-8193393

This is an "Epic" to contain requirements for supporting this JEP on OpenJ9.

[pshipton]

I don't believe there are any plans to implement this.
@DanHeidinga

[fjeremic]

Isn't this something we are still looking into? What is the alternative for OpenJ9? HealthCenter?

[DanHeidinga]

Isn't this something we are still looking into? What is the alternative for OpenJ9? HealthCenter?

Currently, Healthcenter is the alternative for OpenJ9. This is an area we're looking more broadly to determine the right way to get metrics out of the JVM. FlightRecorder is one option but it's not the only one.

Closing this makes sense as this item is unlikely to be the best place to track OpenJ9's final approach to this area

[MariusVolkhart]

@DanHeidinga is there another public location where one can follow along or join the conversation about OpenJ9's approach?

I'm primarily interested in how to do custom events on OpenJ9 (jdk.jfr.* and jdk.jfr.consumer.* equivalents). I looked through various Eclipse and IBM documentation but didn't see this functionality pop up.

[danjhop3]

IT IS NEEDED AT A JAR LEVEL

[DanHeidinga]

I'm primarily interested in how to do custom events on OpenJ9 (jdk.jfr.* and jdk.jfr.consumer.* equivalents). I looked through various Eclipse and IBM documentation but didn't see this functionality pop up.

Does the Trace Java API [1] provide the required functionality? It allows the application to register tracepoints and inject them into the per-thread tracebuffers. These are - by default - an in-memory ring buffer that records the last few events or can be redirected to a file.

is there another public location where one can follow along or join the conversation about OpenJ9's approach?

This is a really good question. IBM HealthCenter is owned by IBM and to my knowledge, hasn't been opened sourced. The OpenJ9 project had relied on HealthCenter for historical reasons (and because HC supported OpenJ9 and was available). I'm not sure what the current state of HC is and what kind of resources are behind it today. Maybe @mstoodle or @pshipton can comment?

This may be an area we need to re-investigate if the current path still works or if we should give JFR / Mission Control another look

[mstoodle]

You're right, it hasn't been open sourced, although nothing to my knowledge would stop us from doing it (other than the fact that it would probably be pretty close to a dump and run :( ). There are a few people who help out with support for HC, but no dedicated resourcing right now.

I would support reopening the discussion to look at JFR / Mission Control, but unfortunately can't commit people to help with such an effort. This is an area where we could really use some help. I'd like to believe the existing hook/trace framework represents some significant fraction of the work that would be needed if we were to go down that path (so we wouldn't be starting from nothing: many of the needed hook points are probably already there), but that doesn't mean there wouldn't still be a fair amount of work needed :( .

[Wood-Xia]

I also strongly support to reopen this discussion.
When we switch from OpenJDK to OpenJ9, we are happy with its low memory footprint.
Well, when we are looking for some profiling tool as JFR which is cool, we are a little disappointed....

[monwolf]

We are at the same point, we are really happy with the low memory footprint but we lost a good capacity of our current APM system that is the real-time profiling of the APP.

[pshipton]

Re-opening since there is active development on JFR for OpenJ9 now.
https://github.com/eclipse-openj9/openj9/pulls?q=is%3Apr+is%3Aopen+jfr