Performance Improvements in JDK 27
Remote
Markus Eisele
@myfear@mastodon.online
mastodon 4.8.0-nightly.2026-10-10Something at #IBMResearch. Legal thinks you need to know that I'm speaking for myself here. I believe you knew.
#Java #Kubernetes #quarkus
and more #fedi22
2476 Followers
641 Following
28 Posts
Joined April 25, 2022
The Main Thread:
GitHub:
website:
My Roq documentation site had page lists that drifted away from the Markdown. I moved membership and ordering into page metadata, then used Qute @TemplateExtension methods to build the views.
I also connected shared Excalidraw sources through Roq Data and ran Pagefind after static generation in Maven. The article covers the code choices and their limits.
https://www.the-main-thread.com/p/quarkus-roq-documentation-sites
An operator needs to retry one shipment. A server shell gives them an operating-system account.
This Quarkus Aesh tutorial takes a narrower path: one tested command model, exposed through SSH and a role-protected browser terminal. It adds dry runs, confirmation, connection limits, readiness checks, and audit events.
The extensions are preview, so the production boundary is part of the tutorial.
https://www.the-main-thread.com/p/quarkus-aesh-ssh-websocket-terminal #Java #Quarkus #Aesh #SSH #WebSockets
A Quarkus + Chart.js tutorial using NASA's exoplanet catalogue.
Java builds discovery counts, method shares, and a period/radius scatter plot with XDEV's chartjs-java-model. Planets with missing measurements stay in the counts; only usable pairs enter the scatter.
Includes a reproducible snapshot, an explicit live source, and 13 tests.
https://www.the-main-thread.com/p/quarkus-chartjs-java-exoplanet-dashboard
Quarkus 4.0.0.Beta1 is tagged!
Java 21 becomes the minimum, with Jackson 3, Hibernate ORM 8, Jakarta REST 4, and CDI 5 among the changes.
A good time to test a familiar application against the beta and report any rough edges. I’d start with JSON serialization and persistence tests.
Thanks to everyone building and testing this release.
https://github.com/quarkusio/quarkus/releases/tag/4.0.0.Beta1
A web chat UI for IBM Bob is the easy part. The useful work is the ACP control plane: keep `session/request_permission` unfinished until a human answers, stream plans and tool calls over SSE, and let Quarkus own the policy around `bob acp`.
I used a tiny Java off-by-one so every permission, plan, and command stays visible. https://www.the-main-thread.com/p/ibm-bob-acp-quarkus-web #IBMBob #ACP #Quarkus #Java
I tried native WebMCP in a Quarkus order page: three browser tools, with return eligibility and refund calculations kept in Java.
Bob calls the page through Chrome DevTools MCP. The tutorial checks the returned draft against the visible page, including rejected inputs, cancellation, and late responses after clearing.
Experimental API, pinned versions, runnable source. The workflow prepares a draft; it submits nothing.
https://www.the-main-thread.com/p/quarkus-webmcp-ibm-bob-browser-tools
Quarkus has `@Timeout`, `@Retry`, and `@Fallback`. A mocked exception skips the HTTP client, so it does not prove those annotations.
Goblin injects latency and HTTP failures at the JAX-RS boundary. I targeted one inventory resource, left the public quote endpoint reachable, and showed the caller returning fallback data.
The 0.0.1 engine stays inactive in the packaged app.
https://www.the-main-thread.com/p/quarkus-goblin-chaos-engineering
Markdown in /src
Bob TV: Episode 2: wit yours truly and Max!
Tokenmaxxing: How Much Context Does AI Coding Really Need?
Tue, Oct 6, 2026 11:00 AM EDT (5:00 PM Europe/Berlin)
https://ibm.webcasts.com/starthere.jsp?ei=1774868&tp_key=60e89a0154
#Register #Free
🏃♂️ Building #OpenPace - federated activity tracking for the Fediverse
Think Strava, but:
✅ Federates with Mastodon
✅ Self-hosted
✅ Your data, your rules
✅ Apache licensed
Thanks @javahippie@mainz.social for the inspiration!
🛠️ Java + Quarkus + Vert.x + ActivityPub
🌐 https://open-pace.com
💻 https://github.com/myfear/open-pace
Post a run → friends on Mastodon see it
Own your miles, share your journey
Contributions welcome. Let's build in the open!
Just published a new deep-dive on building enterprise-grade RAG in Java.
In this tutorial, we combine:
• Quarkus
• Docling (layout-aware PDF parsing)
• pgvector + PostgreSQL
• Local LLMs via Ollama
• And a simple guardrail layer
This is the most complete RAG pipeline I’ve built so far, and it’s fully open for you to copy, run, and adapt.
Read here:
https://www.the-main-thread.com/p/enterprise-rag-quarkus-docling-pgvector-tutorial
Vault's TOTP API can validate a six-digit code. That is not yet a complete MFA flow.
This Quarkus tutorial adds the missing boundaries: authenticated subject, stored enrollment mapping, strict input validation, a two-minute step-up JWT, normal RBAC, and a replay test against a patched Vault runtime.
https://www.the-main-thread.com/p/quarkus-vault-totp-step-up-auth
#Java #Quarkus #Vault #TOTP #APISecurity
AI coding agents make code cheaper. They do not make specification free.
I wrote about the spec trap: loose prompts look fast until rework, token burn, human review, and multi-agent drift show up on the ledger. The sweet spot is not a 40-page requirements document. It is validated acceptance criteria, executable tests, and stronger handoff contracts where they pay.
What is JBang? The Complete Guide to Java Scripting
https://blog.marcnuri.com/jbang-introduction
#Java #Jbang
Java frameworks are usually discussed in terms of benchmarks and best practices.
But sometimes, the fastest way to understand a platform is to do something completely ridiculous with it.
This article breaks down six and a half intentionally weird Quarkus demos from JChampions Conference.
Under the jokes is a serious point: Quarkus is far more flexible than most people realize.
https://www.the-main-thread.com/p/six-and-a-half-ridiculous-things-quarkus
Slides from my #JConEurope session this morning!
JCON Chasing the Main Thread - Adventures in AI Assisted Coding https://speakerdeck.com/myfear/jcon-chasing-the-main-thread-adventures-in-ai-assisted-coding
Replying to
AI agents behave much more like terminal users than API clients.
So it’s not surprising that CLI tools often outperform MCP integrations for many workflows.
For Java developers this opens an interesting path:
JBang + small CLI tools as agent capabilities.
I wrote a deep dive here:
https://www.the-main-thread.com/p/ai-agents-cli-tools-jbang-java-architects
I beg you to follow Crocker's Rules, even if you will be rude to me
https://lr0.org/blog/p/crocker/
🏈 Predicting NFL Games with Java, Quarkus & Tribuo
Even if you don't watch sports, the NFL is a perfect dataset for machine-learning pipelines.
I show how to pull real ESPN data, engineer features, train a Tribuo model, and serve predictions through a Quarkus microservice.
Pure Java. No Python.
One clean pipeline from ingestion → training → inference.
👉 https://www.the-main-thread.com/p/nfl-game-prediction-java-quarkus-tribuo-tutorial
The LLM Curve of Impact on Software Engineers https://serce.me/posts/2025-02-07-the-llm-curve-of-impact-on-software-engineers
#software #developers #llm #genai
Build Beautiful Java Dashboards with Quarkus and IBM Carbon Design System https://www.the-main-thread.com/p/quarkus-carbon-design-system-dashboard-tutorial
Moving a Spring service to Quarkus? Check what happens at build time, which settings still change at startup, and where blocking calls run.
I've refreshed the article with a Quarkus 3.39.1 / JDK 25 example, a migration checklist, and the community's migration skill. It also covers the Quarkus team's published JVM benchmarks, with the versions and setup.
I can pass 50 JUnit examples on a Quarkus price calculator that rounds too early. Coverage has no reason to complain. The suite never asked about rounding order.
jqwik searches the same valid domain. Shrinking reduced a nine-digit miss to six units at one cent with a 9% discount.
That shrunk case is now test 51.
https://www.the-main-thread.com/p/quarkus-jqwik-property-based-testing
“Graph” is a data structure, not a quality guarantee.
I connected code-review-graph to IBM Bob over MCP and benchmarked both on Open Liberty: 120,016 tracked files. The graph runs were 18.8% faster and returned 18.7% less tool output—but found 1/10 required call sites. Solo Bob found 10/10.
The useful lesson is how to benchmark agent tooling without rewarding fast wrong answers.
https://www.the-main-thread.com/p/ibm-bob-code-review-graph-benchmark
