Skip to content

Instantly share code, notes, and snippets.

View rkennke's full-sized avatar
🏠
Working from home

Roman Kennke rkennke

🏠
Working from home
66 followers · 9 following
  • Datadog
  • Zurich, Switzerland
  • X @rkennke
View GitHub Profile
@rkennke
rkennke / gist:35d20024749020ae1d2d345b103c2c14
Created December 14, 2020 15:57
> - You are running with +AbortOnVMCompilationFailure and still see the
> performance drop or not?
On DEV we currently try to run with +AbortOnVMCompilationFailure to quickly get methods not C2 compiling. There are a few with good reason not to compile. Likely nothing to be enabled for PROD. It -TieredCompilation fails back to C1 then we can keep it enabled.
> - Are you still observing the performance drop when you run with raised node limits?
Raised node limits helped the C2 to compile some methods which would be else failing.
Unfortunately the issue is not reproducible. It happened 5 times in the last 30 days on 120 machines. Last time, on Friday, for a very critical component. From the application point of view all was looking normal, several jstacks taken did not indicate anything wrong, nothing in the log file, heap/GC ok, but the process was running like 20-50 times slower.
@rkennke
rkennke / gist:8cbe2fcf35211b3dee243c171348b9c4
Created December 8, 2020 19:39
diff -r ab218a040145 src/share/vm/gc_implementation/shenandoah/shenandoahRootProcessor.cpp
--- a/src/share/vm/gc_implementation/shenandoah/shenandoahRootProcessor.cpp Tue Dec 08 16:07:59 2020 +0100
+++ b/src/share/vm/gc_implementation/shenandoah/shenandoahRootProcessor.cpp Tue Dec 08 20:39:28 2020 +0100
@@ -303,7 +303,7 @@
_serial_roots.oops_do(keep_alive, worker_id);
_dict_roots.oops_do(keep_alive, worker_id);
- _thread_roots.oops_do(keep_alive, &clds, &update_blobs, worker_id);
+ _thread_roots.oops_do(keep_alive, &clds, _update_code_cache ? NULL : &update_blobs, worker_id);
_cld_roots.cld_do(&clds, worker_id);
@rkennke
rkennke / gist:7565a7813ef9a7ab46e0665a3ac38748
Created December 8, 2020 19:37
_thread_roots.oops_do(keep_alive, &clds, &update_blobs, worker_id); <--- updates *some* nmethods
_cld_roots.cld_do(&clds, worker_id);
if(_update_code_cache) {
_code_roots.code_blobs_do(&update_blobs, worker_id); <--- updates *all* nmethods
}
@rkennke
rkennke / gist:61a7026e03a18f40ca51150ee793858e
Created December 8, 2020 18:48
Event: 295,288 Executing coalesced safepoint VM operation: ShenandoahInitMark
Event: 295,288 Pause Init Mark
Event: 295,413 Pause Init Mark done
Event: 295,413 Executing coalesced safepoint VM operation: ShenandoahInitMark done
Event: 295,413 Executing VM operation: RevokeBias done
Event: 295,414 Concurrent marking
Event: 295,417 Executing VM operation: RevokeBias
Event: 295,422 Executing VM operation: RevokeBias done
Event: 295,422 Executing VM operation: RevokeBias
Event: 295,423 Executing VM operation: RevokeBias done
@rkennke
rkennke / gist:a3b08f0dd365a23325dc647f4e8d8dd7
Created December 8, 2020 17:32
#
# A fatal error has been detected by the Java Runtime Environment:
#
# Internal Error (/home/rkennke/src/openjdk/shenandoah-jdk8/hotspot/src/share/vm/gc_implementation/shenandoah/shenandoahForwarding.inline.hpp:71), pid=512644, tid=0x00007f8600dfd640
# Error: Shenandoah assert_correct failed; Object klass pointer must go to metaspace
Referenced from:
no interior location recorded (probably a plain heap scan, or detached oop)
Object:
@rkennke
rkennke / gist:1c029cff1fc4a908046f7193777e75c7
Created December 2, 2020 19:35
[rkennke@localhost substratevm]$ mx native-image -H:+ReportExceptionStackTraces --allow-incomplete-classpath "-J-XX:FlightRecorderOptions=retransform=false" --no-fallback -ea -cp ~/src/graal-jfr/graalvm-jfr-tests/flat EventCommit
[eventcommit:35500] classlist: 988.29 ms, 0.96 GB
[eventcommit:35500] (cap): 628.34 ms, 0.96 GB
[eventcommit:35500] setup: 1,212.93 ms, 0.96 GB
Fatal error:java.lang.NoClassDefFoundError
at java.base/jdk.internal.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method)
at java.base/jdk.internal.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62)
at java.base/jdk.internal.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45)
at java.base/java.lang.reflect.Constructor.newInstance(Constructor.java:490)
at java.base/java.util.concurrent.ForkJoinTask.getThrowableException(ForkJoinTask.java:603)
@rkennke
rkennke / gist:d6337e96074efa468c0e85c54f4d36da
Created December 2, 2020 18:37
pick b3a827c707d Add stub for JfrThreadLocal and insert an instance into SubstrateVM Java Threads implementation
pick d1290475311 Add JfrOptions to read command line arguments for JFR
pick a344171200c Implement metadata parsing at build-time
pick 87c82a027ae Remove substitution for getTypeId(String), it does not exist in JDK11 and is probably not needed
pick fb669e9f838 Add javax.xml.datatype to build-time-initialized packages
pick edb52c55e72 Add some machinery to replace jfrTypeIds.hpp generation and lookup
pick cd0291ed2b0 Provide Metadata.getJfrTypeId() method as replacement for jfrTypeIds.hpp
pick 5b52723d267 Substitute Type.register() in order to get correct typeIDs for known types
pick ab877ac471e Simplify alias of knownTypes field
pick e23be2aab2f Remove substitution for MetadataRepository
@rkennke
rkennke / gist:144295024edfcd0221b9da6ba4f9a474
Created November 30, 2020 20:45
Creating symlink jdk/lib/libverify.debuginfo
/usr/bin/ld: /home/rkennke/src/labs-openjdk-11/build/linux-x86_64-normal-server-release/support/native/java.base/libjava/childproc.o:/home/rkennke/src/labs-openjdk-11/src/java.base/unix/native/libjava/childproc.h:121: multiple definition of `parentPathv'; /home/rkennke/src/labs-openjdk-11/build/linux-x86_64-normal-server-release/support/native/java.base/libjava/ProcessImpl_md.o:/home/rkennke/src/labs-openjdk-11/src/java.base/unix/native/libjava/childproc.h:121: first defined here
collect2: error: ld returned 1 exit status
@rkennke
rkennke / gist:d057f906814a4e022b2bf69ffe996712
Created November 4, 2020 14:13
diff --git a/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.cpp b/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.cpp
index f19a93ea487..fb1b9ccbf20 100644
--- a/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.cpp
+++ b/src/hotspot/cpu/x86/gc/shenandoah/shenandoahBarrierSetAssembler_x86.cpp
@@ -348,27 +348,33 @@ void ShenandoahBarrierSetAssembler::load_reference_barrier(MacroAssembler* masm,
}
save_xmm_registers(masm);
+ address calladdr = NULL;
switch (kind) {
@rkennke
rkennke / gist:3e9f156225e5e40f1d1c324861af0b93
Created October 6, 2020 14:57
0x00007f2f2cb7650f: mov 0x10(%rbx),%rbp <==== Reference.get()
0x00007f2f2cb76513: testb $0x1,0x20(%r15)
0x00007f2f2cb76518: jne 0x00007f2f2cb76c42 <==== LRB may turn rbp into NULL
0x00007f2f2cb7651e: mov %rbp,%r12 <==== copy rbp (possibly NULL) to r12
0x00007f2f2cb76521: testb $0x2,0x20(%r15)
0x00007f2f2cb76526: jne 0x00007f2f2cb76c61 <==== keep-alive barrier
0x00007f2f2cb7652c: mov 0x50(%rsp),%r10
0x00007f2f2cb76531: cmp %r12,%r10
0x00007f2f2cb76534: jne 0x00007f2f2cb76562 <==== Not sure what that is, but branches to crashing subroutine, with r12 still NULL
0x00007f2f2cb76536: mov 0x38(%rbx),%r10