JFR emergency dump didn't happen at OOM

[Erik Gahlin]

Hi Yasumasa,

This looks more reasonable, although most of it is a refactoring.

Could you turn it into a PR with a webrev so it's easier to review?

Thanks
Erik


________________________________
From: hotspot-jfr-dev <hotspot-jfr-dev-retn at openjdk.org> on behalf of Yasumasa Suenaga <suenaga at oss.nttdata.com>
Sent: Thursday, July 24, 2025 11:21 AM
To: hotspot-jfr-dev at openjdk.org <hotspot-jfr-dev at openjdk.org>
Subject: JFR emergency dump didn't happen at OOM

Hi all,

I reported and proposed the fix about 6 years ago (!), I faced the issue again that would be helpful if this is fixed.

6 years ago, we discussed about this issue in [1], then I hear JDK-8196050 might fix this issue, but it hasn't yet happened unfortunately...

JDK codebase has been made a lot of changes in 6 years, so I think we can fix simply than I proposed before.
All of changes is [2]. It is not so big change and it is good because no flags are added in .jfc files. What do you think? I will send PR if it is ok, but I want to hear your opinions before PR (because we discussed before).

For example, we can pass a flag for OOM rather than exception handler as following:

```
--- a/src/hotspot/share/jfr/recorder/repository/jfrEmergencyDump.cpp
+++ b/src/hotspot/share/jfr/recorder/repository/jfrEmergencyDump.cpp
@@ -556,22 +556,21 @@ class JavaThreadInVMAndNative : public StackObj {
    }
  };

-static void post_events(bool exception_handler, Thread* thread) {
-  if (exception_handler) {
+static void post_events(bool is_oom, Thread* thread) {
+  if (is_oom) {
+    LeakProfiler::emit_events(max_jlong, false, false);
+  } else {
      EventShutdown e;
      e.set_reason("VM Error");
      e.commit();
-  } else {
-    // OOM
-    LeakProfiler::emit_events(max_jlong, false, false);
    }
    EventDumpReason event;
-  event.set_reason(exception_handler ? "Crash" : "Out of Memory");
+  event.set_reason(is_oom ? "Out of Memory" : "Crash");
    event.set_recordingId(-1);
    event.commit();
  }
```

And also we can identify whether OOM or not (crash) via `id` in argument of `report_and_die`:
```
--- a/src/hotspot/share/utilities/vmError.cpp
+++ b/src/hotspot/share/utilities/vmError.cpp
@@ -1856,7 +1856,7 @@ void VMError::report_and_die(int id, const char* message, const char* detail_fmt
      log.set_fd(-1);
    }

-  JFR_ONLY(Jfr::on_vm_shutdown(true);)
+  JFR_ONLY(Jfr::on_vm_shutdown(static_cast<VMErrorType>(_id) == OOM_JAVA_HEAP_FATAL);)

```


Thanks,

Yasumasa


[1] https://mail.openjdk.org/pipermail/hotspot-jfr-dev/2019-January/000381.html
[2] https://github.com/YaSuenag/jdk/commit/3f18fd74679387c385a8f79c8bea5b6b1f2815c0
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/hotspot-jfr-dev/attachments/20250724/e1a03690/attachment-0001.htm>