May 2024 - hotspot-jfr-dev

RFR: 8330205: Initial troff manpage generation for JDK 24
by Pavel Rappo 14 May '24

14 May '24

Please review this mechanical change to man pages. This PR should be integrated after https://github.com/openjdk/jdk/pull/18787. ------------- Commit messages: - Initial commit Changes: https://git.openjdk.org/jdk/pull/19119/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=19119&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8330205 Stats: 30 lines in 28 files changed: 0 ins; 0 del; 30 mod Patch: https://git.openjdk.org/jdk/pull/19119.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/19119/head:pull/19119 PR: https://git.openjdk.org/jdk/pull/19119

4 8

Priority redistribution in Old Object Sample event
by Stig Rohde Døssing 12 May '24

12 May '24

Hi, I'm having a hard time understanding the priority redistribution code for OldObjectSample. Some months ago, there was a thread about it at https://mail.openjdk.org/pipermail/hotspot-jfr-dev/2023-June/005066.html, but it didn't conclude one way or the other whether this code looked correct. I'm hoping to understand the intent, and maybe also discuss whether the current implementation follows that intent. I'll be describing my understanding in order to surface any misunderstandings, so apologies for the incoming wall of text. Please correct me if I'm getting things wrong. # The intent If we think of allocations as a "timeline", where each sampled object extends the timeline by adding more bytes to the right side of the line, my understanding is that the goal of this code is to ensure that the samples we keep around in the priority list are as evenly distributed across this timeline as possible. We'd presumably like a good distribution whether objects are garbage collected or not. Going by the earlier mailing list comment at https://mail.openjdk.org/pipermail/hotspot-jfr-dev/2024-January/005847.html, it seems like the intent was that the sample spans should cover the entire allocation timeline when summed. I think the invariant that's being aimed for is that a sample always covers the span of bytes allocated since the older neighboring sample was taken. When a node is removed, the span of the removed node should be given to the younger neighbor, which ensures the neighbor will cover the span left missing by the node being removed. When a node is removed that has no younger neighbor, the span should be given to the next sample added to the queue, which ensures the new sample will cover the span to the older neighbor. The total spans of the nodes in the queue will then nearly sum to the total allocated bytes, and any missing bit (due to removal of the youngest sample/rejection of incoming samples because their spans are too small) will be covered by the next sample added to the queue. I think the idea is that by continually adjusting sample spans as neighbors are GC'ed or otherwise removed, and by keeping the samples with the largest spans, we'll end up with samples that are well distributed across the timeline. # The current implementation When an object is added, the sampler calculates a new span by counting how many bytes have been sampled so far (including the sample potentially being added), and subtracting the spans of the items already in the queue. https://github.com/openjdk/jdk/blob/32c7681cf310c87669c502c4a8b62a7fecc9336… . This calculated span is used only for deciding whether to add the new sample to the queue. If added, the sample gets a span equal to the byte size of the object being allocated. https://github.com/openjdk/jdk/blob/32c7681cf310c87669c502c4a8b62a7fecc9336… When a sample is dropped due to GC, the span it covers is handed to the `prev` sample in the list. As far as I can tell, the list is ordered youngest-sample-first, so this adds the span to the younger neighbor sample. If there is no younger neighbor, the span is discarded. https://github.com/openjdk/jdk/blob/32c7681cf310c87669c502c4a8b62a7fecc9336… When a sample is dropped in favor of a new sample with a larger span, the old sample's span is discarded. https://github.com/openjdk/jdk/blob/32c7681cf310c87669c502c4a8b62a7fecc9336… # The bits I think might not fit the intent * The span stored in each sample is not the calculated span, it's just the object's byte size (`allocated`). That means as soon as any object falls out of the queue, the spans in the queue no longer sum to cover the allocation timeline. This causes all future samples to be added to be unduly prioritized for adding to the queue, because they are given an artificially high span. In effect, future samples are weighted as if they cover both the interval between themselves and the older neighbor sample, plus all "missing spans" from nodes that have been discarded since the program started. Just to demonstrate that there might be a problem, I tried limiting the sampler queue size to 3, and calling the `ObjectSampler::add` method 10 times in a row with a dummy object and a fixed byte size of 10. The result is that the 1st, 2nd and 10th allocations are kept. This is because after the first 3 inserts to fill the queue, the 4th sample is discarded because the calculated span is 40 - 30 = 10. The 5th item will calculate the span 50 - 30 = 20, and so replace the smallest sample (in span terms), which will be the youngest since all samples have a stored span of 10. This will repeat for the 6th-10th items (the calculated span increasing by 10 per item), always replacing the most recently inserted sample. I think the span stored in a sample should be the span calculated when comparing to existing samples, and not the object's byte size. * When a sample is removed from the queue because a sample with a larger span is being added, the span of the removed node is not handed to the younger neighbor, this only happens when a sample is removed due to GC. This means that the span will be given to the next sample added to the queue. When the sample being removed is the youngest sample, this is fine, but when it's a sample that has a younger neighbor, the span should probably be given to that neighbor rather than the newcomer. Handing it to the newcomer gives the new sample a high weight it doesn't deserve. It ends up covering not just the span to the older neighbor, but also the span of the removed node, which isn't what we want. I think when a sample is removed because it's being evicted in favor of a new sample, its span should be given to the younger neighbor. If there is no younger neighbor, the span should instead be given to the sample being added.

1 0

RFR: 8331931: JFR: Avoid loading regex classes during startup
by Erik Gahlin 10 May '24

10 May '24

Could I have review of a change that avoids loading 45 regex related classes when using -XX:StartFlightRecording. Testing: jdk/jdk/jfr Thanks Erik ------------- Commit messages: - Remove debug statement - Initial Changes: https://git.openjdk.org/jdk/pull/19146/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=19146&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8331931 Stats: 4 lines in 1 file changed: 1 ins; 0 del; 3 mod Patch: https://git.openjdk.org/jdk/pull/19146.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/19146/head:pull/19146 PR: https://git.openjdk.org/jdk/pull/19146

2 2

RFR: 8319997: JFR: Reduce use of dynamic proxies
by Erik Gahlin 07 May '24

07 May '24

Hi, Could I have a review of a change that removes the use of dynamic proxies for known event metadata, for example labels and descriptions. Testing: test/jdk/jdk/jfr Thanks ------------- Commit messages: - Initial Changes: https://git.openjdk.org/jdk/pull/19106/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=19106&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8319997 Stats: 81 lines in 5 files changed: 21 ins; 35 del; 25 mod Patch: https://git.openjdk.org/jdk/pull/19106.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/19106/head:pull/19106 PR: https://git.openjdk.org/jdk/pull/19106

2 2

RFR: 8331653: JFR: Improve logging for jdk/jfr/api/consumer/recordingstream;TestStop.java
by Erik Gahlin 06 May '24

06 May '24

Could I have a review of a change that improves logging for a test that fails intermittently. Testing: 300 x jdk/jfr/api/consumer/recordingstream;TestStop.java Thanks Erik ------------- Commit messages: - Add back blank line - Initial Changes: https://git.openjdk.org/jdk/pull/19090/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=19090&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8331653 Stats: 33 lines in 1 file changed: 20 ins; 0 del; 13 mod Patch: https://git.openjdk.org/jdk/pull/19090.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/19090/head:pull/19090 PR: https://git.openjdk.org/jdk/pull/19090

2 2

RFR: 8331507: JFR: Improve example usage in -XX:StartFlightRecording:help
by Erik Gahlin 03 May '24

03 May '24

Could I have review that changes the the first example for -XX:StartFlightRecording:help. See bug for more details. Testing: test/jdk/jdk/jfr Thanks Erik ------------- Commit messages: - Initial Changes: https://git.openjdk.org/jdk/pull/19067/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=19067&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8331507 Stats: 3 lines in 1 file changed: 2 ins; 0 del; 1 mod Patch: https://git.openjdk.org/jdk/pull/19067.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/19067/head:pull/19067 PR: https://git.openjdk.org/jdk/pull/19067

2 2

RFR: 8294978: Convert 5 test/jdk/jdk/jfr tests from ASM library to Classfile API
by Chen Liang 03 May '24

03 May '24

Converted these tests to Classfile API and some of the constants to Constant API. InstrumentationCallback.callback wasn't changed to a record since I cannot figure out how to ship a nested record class with InstrumentationCallback in a jar. ------------- Commit messages: - Convert jdk.jfr tests to ASM Changes: https://git.openjdk.org/jdk/pull/13072/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=13072&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8294978 Stats: 283 lines in 5 files changed: 75 ins; 108 del; 100 mod Patch: https://git.openjdk.org/jdk/pull/13072.diff Fetch: git fetch https://git.openjdk.org/jdk pull/13072/head:pull/13072 PR: https://git.openjdk.org/jdk/pull/13072

3 19

RFR: 8331478: JFR: Rename printHelp methods for jdk.jfr.internal.dcmd classes
by Erik Gahlin 02 May '24

02 May '24

Could I have a review of a PR that changes the name of the print methods used by the JFR diagnostic commands. The names are legacy from when they printed the help which hasn't been the case for several years. Testing: tier1 + tier2 + test/jdk/jdk/jfr Thanks Erik ------------- Commit messages: - Initial Changes: https://git.openjdk.org/jdk/pull/19041/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=19041&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8331478 Stats: 22 lines in 9 files changed: 0 ins; 0 del; 22 mod Patch: https://git.openjdk.org/jdk/pull/19041.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/19041/head:pull/19041 PR: https://git.openjdk.org/jdk/pull/19041

2 2

RFR: 8323058: Revisit j.l.classfile.CodeBuilder API surface
by Adam Sotona 02 May '24

02 May '24

`java.lang.classfile.CodeBuilder` contains more than 230 API methods. Existing ClassFile API use cases proved the concept one big CodeBuilder is comfortable. However there are some redundancies, glitches in the naming conventions, some frequently used methods are hard to find and some methods have low practical use. This patch revisits the `CodeBuilder` API methods and introduces some changes. For more details, please, visit the [CSR ](https://bugs.openjdk.org/browse/JDK-8323067) Please review. Thank you, Adam ------------- Commit messages: - Merge branch 'master' into JDK-8323058-CodeBuilder - extended CodeBuilder::conversion functionality - removed CodeBuilder::newPrimitiveArray, newReferenceArray, newMultidimensionalArray and operator methods - fixed tests - fixed jdk/classfile tests - fixed benchmarks - fixed jdk.jfr and jdk.jlink - fixed java.base - 8323058: Revisit j.l.classfile.CodeBuilder API surface Changes: https://git.openjdk.org/jdk/pull/17282/files Webrev: https://webrevs.openjdk.org/?repo=jdk&pr=17282&range=00 Issue: https://bugs.openjdk.org/browse/JDK-8323058 Stats: 876 lines in 43 files changed: 41 ins; 188 del; 647 mod Patch: https://git.openjdk.org/jdk/pull/17282.diff Fetch: git fetch https://git.openjdk.org/jdk.git pull/17282/head:pull/17282 PR: https://git.openjdk.org/jdk/pull/17282

5 13