Implement the JFR ObjectAllocationSample event - VM part by LinHu2016 · Pull Request #24609 · eclipse-openj9/openj9

LinHu2016 · 2026-08-21T17:06:15Z

Base on JFR event specification (from SAP JFR Events
17):

Define the event structure J9JFRObjectAllocationSample in
j9nonbuilder.h and J9JFR_EVENT_TYPE_OBJECT_ALLOCATION_SAMPLE id in
j9consts.h
Register a JFR-internal callback jfrObjectAllocationSample() on
J9HOOK_MM_OBJECT_ALLOCATION_SAMPLING_INTERNAL in
startJFRRecording(), implement the callback to write a
J9JFRObjectAllocationSample event with
a stack trace into the per-thread buffer, and unregister it in
stopJFRRecording().
JFR specifies ObjectAllocationSample throttling in events-per-second
(default: 150/s, profiling: 300/s). The GC layer works in bytes. A
conversion is needed: from the JVM's current heap allocation rate
(bytes/sec), derive a byte-granularity interval such that approximately
N events/second are emitted. JVM startup uses a reasonable default
byte interval 512KB. new objectAllocationSampleThrottleRate (throttle
value (events/s)) in vm->jfrState.
Add a periodic recalibration step in jfrSamplingThreadProc()'s
1-second tick block, Read total bytes allocated since last GC, Compute
new byte interval: newInterval = allocatedBytes / (throttleRate *
elapsedSinceLastGC).

#depends on: eclipse-omr/omr#8407
#depends on: #24774
#relate to: #24778
#fix: #24213

Signed-off-by: lhu linhu@ca.ibm.com

LinHu2016 · 2026-09-22T16:51:23Z

Hi @babsingh Could you please review the changes, Thanks

LinHu2016 · 2026-09-23T17:36:19Z

Optimize recalibration and integrate JFR standard settings/configuration in separate PRs.

babsingh

Also adding @gacholio to review this PR. They’re currently on vacation and will return next week.

babsingh · 2026-09-24T14:20:42Z

+	    J9ArrayClass *arrayClass = (J9ArrayClass *)data->clazz;
+	    U_8 *classLeafName = J9UTF8_DATA(J9ROMCLASS_CLASSNAME(arrayClass->leafComponentType->romClass));
+	    UDATA lenClassLeafName = J9UTF8_LENGTH(J9ROMCLASS_CLASSNAME(arrayClass->leafComponentType->romClass));


use tabs for indentation to match the coding standard

babsingh · 2026-09-24T14:21:32Z

+	MM_ObjectAllocationSamplingInternalEvent *data =
+		(MM_ObjectAllocationSamplingInternalEvent *)eventData;
+	J9VMThread *currentThread = data->currentThread;
+
+	U_8 *className = J9UTF8_DATA(J9ROMCLASS_CLASSNAME(data->clazz->romClass));
+	UDATA lenClassName = J9UTF8_LENGTH(J9ROMCLASS_CLASSNAME(data->clazz->romClass));


these read-only locals be made const: data, className, the name lengths, and arrayClass/classLeafName.

babsingh · 2026-09-24T14:23:54Z

+							UDATA newInterval = ((0 != totalBytes) && (0 != lastGCEnd))
+								? (totalBytes * 1000000 / (elapsedMicros * throttleRate))
+								: J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_INTERVAL;
+							if (newInterval < 1024) {
+								newInterval = 1024;
+							} else if (newInterval > (64 * 1024 * 1024)) {
+								newInterval = 64 * 1024 * 1024;
+							}


macros must be defined for: 1000000, 1024 and 64, as per the coding standard, similar to

#define J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_THROTTLE_RATE 150 #define J9JFR_OBJECT_ALLOCATION_SAMPLE_PROFILING_THROTTLE_RATE 300

babsingh · 2026-09-24T14:24:21Z

 #define J9JFR_GLOBAL_BUFFER_SIZE (10 * J9JFR_THREAD_BUFFER_SIZE)
 #define J9JFR_SAMPLING_RATE 10
 #define J9JFR_CLASSNAME_BUFFER_SIZE 128
+#define J9TIME_NANOSECONDS_PER_SECOND	(1000000000ULL)


this doesn't appear to be used

babsingh · 2026-09-24T14:25:15Z

+#define J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_INTERVAL (512 * 1024)  /* bytes; same as JVMTI default per JEP 331 */
+#define J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_THROTTLE_RATE 150	/* events per second */
+#define J9JFR_OBJECT_ALLOCATION_SAMPLE_PROFILING_THROTTLE_RATE 300	/* events per second */


inconsistent spacing before the comment

Suggested change

#define J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_INTERVAL (512 * 1024) /* bytes; same as JVMTI default per JEP 331 */

#define J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_THROTTLE_RATE 150 /* events per second */

#define J9JFR_OBJECT_ALLOCATION_SAMPLE_PROFILING_THROTTLE_RATE 300 /* events per second */

#define J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_INTERVAL (512 * 1024) /* bytes; same as JVMTI default per JEP 331 */

#define J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_THROTTLE_RATE 150 /* events per second */

#define J9JFR_OBJECT_ALLOCATION_SAMPLE_PROFILING_THROTTLE_RATE 300 /* events per second */

babsingh · 2026-09-24T14:42:45Z

+	J9JFRObjectAllocationSample *jfrEvent = (J9JFRObjectAllocationSample *)reserveBufferWithStackTrace(
+			currentThread, currentThread, J9JFR_EVENT_TYPE_OBJECT_ALLOCATION_SAMPLE, sizeof(J9JFRObjectAllocationSample), 0);


reserveBufferWithStackTrace() can eventually flush the JFR buffer, where allocateMemFromGlobalBuffer() releases VM access to acquire jfrBufferMutex. Since this callback is running from the allocation hook, can we confirm that it is valid for the subscriber to drop VM access here and that the allocator does not retain any GC sensitive refs across the hook dispatch? If not, this path needs to avoid dropping VM access or protect/update those references.

@gacholio for review

babsingh · 2026-09-24T14:44:46Z

 	SystemGCID = 36,
 	YoungGarbageCollectionID = 38,
 	OldGarbageCollectionID = 39,
+	ObjectAllocationSampleID = 83, /* jdk.ObjectAllocationSample -- must match JFR metadata blob */


does this also work with JFRv2?

for Java17 also JfrObjectAllocationSampleEvent = 83, but need to update for Java 25 JfrObjectAllocationSampleEvent = 91

Is this consistent (or correctly inconsistent, I suppose) with the RI? I suppose we only really care about LTS versions.

for Java17 also JfrObjectAllocationSampleEvent = 83, but need to update for Java 25 JfrObjectAllocationSampleEvent = 91

In V1 it will be the same for all Java versions. Don't worry about V2 for now @thallium is working on a change that will use the generated IDs in V2.

Also, the comment /* jdk.ObjectAllocationSample -- must match... is not needed as it applies the same to all Ids in the enum.

Here's the PR for using generated IDs in JFR v2: #24839

babsingh · 2026-09-24T14:46:01Z

+		Trc_VM_jfrObjectAllocationSample_indexableObject(currentThread,
+			lenClassName,
+			className,
+			lenClassLeafName,
+			classLeafName,
+			data->weight,
+			data->objectSize);
+	} else {
+		Trc_VM_jfrObjectAllocationSample(currentThread,
+			lenClassName,
+			className,


arguments should start on a new line

babsingh · 2026-09-24T14:46:20Z

+		jfrEvent->objectClass = data->clazz;
+		jfrEvent->weight      = data->weight;


Suggested change

jfrEvent->objectClass = data->clazz;

jfrEvent->weight = data->weight;

jfrEvent->objectClass = data->clazz;

jfrEvent->weight = data->weight;

babsingh · 2026-09-24T14:46:30Z

 	}
+
+	/* enable JFRObjectAllocationSample */
+	vm->jfrState.objectAllocationSampleThrottleRate  = J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_THROTTLE_RATE;


Suggested change

vm->jfrState.objectAllocationSampleThrottleRate = J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_THROTTLE_RATE;

vm->jfrState.objectAllocationSampleThrottleRate = J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_THROTTLE_RATE;

tajila · 2026-09-29T15:11:53Z

 	static constexpr int NETWORK_UTILIZATION_EVENT_SIZE = (4 * sizeof(U_64)) + sizeof(U_32);
 	static constexpr int DATA_LOSS_EVENT_SIZE = sizeof(U_8) + LEB128_32_SIZE + (3 * LEB128_64_SIZE);
 	static constexpr int THREAD_ALLOCATION_STATISTICS_EVENT_SIZE = sizeof(U_8) + LEB128_32_SIZE + (3 * LEB128_64_SIZE);
+	/* OBJECT_ALLOCATION_SAMPLE_EVENT_SIZE: eventSize(LEB128_64) + eventType(LEB128_32) + ticks(LEB128_64) + eventThread(LEB128_64) + stackTrace(LEB128_32) + objectClass(LEB128_32) + weight(LEB128_64) */


this comment is not needed

tajila · 2026-09-29T15:12:11Z

+	I_64 ticks;
+	U_64 eventThreadIndex;
+	U_32 stackTraceIndex;
+	U_32 objectClassIndex; /**< class constant-pool index for the allocated object class */


Base on JFR event specification (from [SAP JFR Events 17](https://sap.github.io/jfrevents/17.html#objectallocationsample)): - Define the event structure J9JFRObjectAllocationSample in j9nonbuilder.h and J9JFR_EVENT_TYPE_OBJECT_ALLOCATION_SAMPLE id in j9consts.h - Register a JFR-internal callback jfrObjectAllocationSample() on J9HOOK_MM_OBJECT_ALLOCATION_SAMPLING_INTERNAL in startJFRRecording(), implement the callback to write a J9JFRObjectAllocationSample event with a stack trace into the per-thread buffer, and unregister it in stopJFRRecording(). - JFR specifies `ObjectAllocationSample` throttling in events-per-second (default: 150/s, profiling: 300/s). The GC layer works in bytes. A conversion is needed: from the JVM's current heap allocation rate (bytes/sec), derive a byte-granularity interval such that approximately `N` events/second are emitted. JVM startup uses a reasonable default byte interval 512KB. new objectAllocationSampleThrottleRate (throttle value (events/s)) in vm->jfrState. - Add a periodic recalibration step in jfrSamplingThreadProc()'s 1-second tick block, Read total bytes allocated since last GC, Compute new byte interval: newInterval = allocatedBytes / (throttleRate * elapsedSinceLastGC). Signed-off-by: lhu <linhu@ca.ibm.com>

LinHu2016 force-pushed the jfrAllocationSampling2 branch 2 times, most recently from 19a978f to 878ab16 Compare August 24, 2026 16:03

tajila added comp:vm comp:gc jfr labels Aug 25, 2026

tajila requested review from dmitripivkine and tajila August 25, 2026 15:11

LinHu2016 force-pushed the jfrAllocationSampling2 branch 13 times, most recently from fcbcd85 to 3496f10 Compare August 27, 2026 14:13

LinHu2016 mentioned this pull request Aug 28, 2026

Implement the JFR ObjectAllocationSample event - step2 #24659

Open

LinHu2016 force-pushed the jfrAllocationSampling2 branch from 3496f10 to 9261d6f Compare August 28, 2026 15:38

LinHu2016 force-pushed the jfrAllocationSampling2 branch 8 times, most recently from 66940ae to c9fa930 Compare September 10, 2026 21:13

LinHu2016 force-pushed the jfrAllocationSampling2 branch 2 times, most recently from 7c315a1 to 63d9beb Compare September 15, 2026 16:29

LinHu2016 changed the title ~~Implement the JFR ObjectAllocationSample event - step1~~ Implement the JFR ObjectAllocationSample event - VM part Sep 15, 2026

LinHu2016 force-pushed the jfrAllocationSampling2 branch 3 times, most recently from defe6aa to 2647565 Compare September 18, 2026 02:08

babsingh reviewed Sep 24, 2026

View reviewed changes

babsingh requested a review from gacholio September 24, 2026 14:49

LinHu2016 force-pushed the jfrAllocationSampling2 branch 2 times, most recently from 8dc6291 to 8d23c13 Compare September 25, 2026 17:32

LinHu2016 mentioned this pull request Sep 28, 2026

Wire up en/disable and throttle control for jfrObjectAllocationSample #24828

Open

tajila reviewed Sep 29, 2026

View reviewed changes

LinHu2016 force-pushed the jfrAllocationSampling2 branch 2 times, most recently from 6696a8e to 4f90688 Compare September 29, 2026 16:20

LinHu2016 force-pushed the jfrAllocationSampling2 branch from 4f90688 to 76316ae Compare September 29, 2026 16:22

		J9JFRObjectAllocationSample jfrEvent = (J9JFRObjectAllocationSample )reserveBufferWithStackTrace(
		currentThread, currentThread, J9JFR_EVENT_TYPE_OBJECT_ALLOCATION_SAMPLE, sizeof(J9JFRObjectAllocationSample), 0);

		jfrEvent->objectClass = data->clazz;
		jfrEvent->weight = data->weight;

	vm->jfrState.objectAllocationSampleThrottleRate = J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_THROTTLE_RATE;
	vm->jfrState.objectAllocationSampleThrottleRate = J9JFR_OBJECT_ALLOCATION_SAMPLE_DEFAULT_THROTTLE_RATE;

Conversation

LinHu2016 commented Aug 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

LinHu2016 commented Sep 22, 2026

Uh oh!

LinHu2016 commented Sep 23, 2026

Uh oh!

babsingh left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

LinHu2016 commented Aug 21, 2026 •

edited

Loading