The Memory Benchmark Nobody Asked For

Sun, 24 May 2026 00:00:00 +0000

Native image, memory, and the benchmark nobody warned you about.

A Micronaut 5.0.0 Space Observatory app — 16 REST endpoints, reactive R2DBC, 7-table PostgreSQL schema. The challenge: run it in a 256 MB container with swap disabled. GraalVM native image, garbage collector shootouts, invisible JVM memory, AOT bugs, and reactive backpressure tuning.

Micronaut 5 Native Image: 14 Entries vs 1,913
Native Image vs JVM at 128 MB
The Observability Tax
Every GC Failed at 256 MB (Except the Boring One)
107 MB of Invisible JVM RAM
Your JVM Reports 151 MB. Your Container Uses 248.
Three Bugs in a Trenchcoat Pretending to Be One Hang
The AOT Hang That Wasn’t
Seven Fixes, Zero Errors
flatMap’s Second Argument Saved 1.7 Million Requests
Loom Carrier on a Reactive Stack: 78 MB for Free
We Removed Reactor. Nothing Got Slower.
Error Is Not Exception: Micrometer’s Silent CPU Metric on Native Image
ExecutorType.VIRTUAL Doesn’t Exist
265 req/s on All Three Runtimes (Until You Add a Memory Limit)
Designed a Reactive Benchmark. Imperative Won.
The Throughput Was Always 115
Quarkus 12,563 vs Spring 820 (Both Numbers Are Wrong)
-Xmx Won’t Save Your Native Image
1% CPU and 6-Second Latency
Two ways to mistranslate a retry
Three Front Doors, One Engine
Six OOMs and a Connection Pool
Four Bugs the Tests Couldn’t See
Every Route Registered. Every Request 404’d.
16,700 Classes Before Your First Request
Eight Connections, Zero Available
2492, 2495, 2508
Native Image Grew to 734 MB

Series on tsvinc

The Memory Benchmark Nobody Asked For