New blog post: Chasing a Metaspace leak #OpenJDK #Metaspace #MemoryLeak #Instrumentation https://t.co/R9WhVOzh0D

In this week's blog post, I'll showcase how the meta-agent hot-patches the JVM to record the byte-code transformations of native Java agents:

https://t.co/7Y4bqMMeNy Can't keep that daddy's old java around forever.

"Help! My LocalDate Isn’t Flattened" Enjoyed reading this post by @cayhorstmann about the subtleties of turning Java's LocalDate into a value type for Project Valhalla. Null-ability and built-in object serialization keep haunting us. https://t.co/YG8PS4scjy

In Linux, closing a file descriptor doesn't always mean you've freed its memory. This is a classic source of subtle memory leaks in long-running applications. The reason? A powerful design choice centered around two syscalls: mmap and its counterpart, munmap. Let's break it

https://t.co/XIWZZuMOV2 C++ Compile times. Ethereum byte code and runtime. Java and exceptions; more C++ compile times. Fat binaries & code portability. Language design around memory safety

The Virtual @OpenJDK Serviceability Meetup 2025 will be the 17th of December: https://t.co/RP8EM1fLBG

Latency Profiling and Optimization

How we optimize RocksDB in TiKV (Part 1) : The Battle Against the DB Mutex

The JFR events collection website has a new home: https://t.co/Uw867D01MF Old links will still work and are forwarded to this location.

https://t.co/LldAcbV81x Why LLVM is slow; inst select, one pass, unrolling. gcc vs clang. PGO. Array init IR, try catch finally, throws, unwind tables, deopt, exceptional control flow. interior pointers. Stack unwinding vs branching checks

Kernel 6.17 has a bug causing async-profiler to freeze an entire system. I noticed this yesterday evening. I thought I was just too tired, but it happened even today. Independently, @forked_franz observed it too: https://t.co/IYcS7Kv6Md Kernel fix:

JEP 8371379 ("Deprecate UTF-16-only String Representation") added. See

JEP 526 ("Lazy Constants (Second Preview)") updated: Status changed from "Candidate" to "Proposed to target". See

JEP 500 ("Prepare to Make Final Mean Final") updated: Status changed from "Candidate" to "Proposed to target for version 26". See

@abhi9u There is still too much cost. Instead in JVM JMH what we do is to collect the number of iterations/invocations and perform the before/after measurements via rdtsc. It means microbenchs should have steady state and enough iterations. And any mem load should be repeated x op.

Meet async-profiler 4.2: Method tracing, Process sampling, better stacks. Now included in Amazon Corretto. https://t.co/rCxrcLYRlp

Learn how to import JFR data into DuckDB, to easily query and analyze your profiling data in this week's blog post:

Get yer extremely niche RFC commentary here:

I wrote a blog post about fast call-stack backtracing. Hopefully, someone making an intrusive profiler, memory tracker, or logger will find it useful... https://t.co/Lb6j5WYkjr

CppCon 2014: Chandler Carruth "Efficiency with Algorithms, Performance with Data Structures" https://t.co/jKPc3AsBvc It's been in my archives for a while and makes a good watch in the context of caches. Things have evolved, but the core value prop remains relevant in '25.