Query Java Flight Recorder files with SQL — load a .jfr recording into DuckDB, write queries,
and visualise the results as interactive charts in a notebook UI.
→ Live web app · Documentation
Early prototype. The database schema may change at any point.
JFR recordings are rich but hard to explore. JFR Query converts a recording into a DuckDB database and lets you write plain SQL against it — plus a full set of built-in views so you can ask questions like "what are the hottest methods?" or "how long did each GC phase take?" without knowing the raw event schema.
The web UI wraps that into a notebook where each cell can be a SQL query, a prose paragraph, or a chart — all live-linked so changing a query instantly updates every downstream chart and scalar.
Download the latest release or run with jbang:
jbang jfr-query@parttimenerd/jfr-queryStart the notebook UI:
java -jar query.jar serve myrecording.jfr
# → open http://localhost:4244Run a query directly from the terminal:
java -jar query.jar query myrecording.jfr "hot-methods"Method Samples Percent
------------------------------------------------------------------- ------- -------
java.util.concurrent.ForkJoinPool.deactivate(...) 1066 8.09%
scala.collection.immutable.RedBlackTree$.lookup(...) 695 5.27%
akka.actor.dungeon.Children.initChild(ActorRef) 678 5.14%
View names expand to SELECT * FROM <view>, so hot-methods is shorthand for the built-in view.
Run java -jar query.jar views to list all views, macros to list SQL macros.
Compose SQL queries, prose, and charts in a single document. Cells share data through named
aliases — one cell can feed multiple charts, and inline ${SELECT …} expressions embed query
results directly in text.
 |
 |
| Line, area, scatter — time-series and correlation | Interactive flamegraph with % parent / % total |
 |
 |
| Bar (grouped, stacked, horizontal) and bar+line | 2D heatmap for cross-dimensional intensity |
 |
 |
| Pie and donut charts | Histogram with optional log-scale bins |
Also: BOX_PLOT, GANTT, RANGE (confidence bands), TABLE.
See the Plot DSL reference for full syntax.
Parameterise queries with sliders, dropdowns, and text inputs. Every chart and inline scalar re-evaluates automatically when a variable changes.
The UI ships with ready-made notebooks for GC analysis, heap allocation, threading, and
exceptions. Open them from the New from template button. Point --templates-dir at your
own folder to add team-shared templates:
java -jar query.jar serve --templates-dir ~/jfr-templates myrecording.jfrContext-aware SQL completions powered by Google Gemini or OpenAI — the assistant knows your views, macros, and loaded schema.
Usage: query.jar [-hV] [COMMAND]
Commands:
import Import a JFR recording into a DuckDB database file
query Run a SQL query or built-in view on a JFR recording
serve Start the notebook web UI
macros List available SQL macros
views List available SQL views
context Dump table/view/macro descriptions for AI prompt injection
help Display help for a command
Import once, query many times:
java -jar query.jar import myrecording.jfr mydb.db
duckdb mydb.db "SELECT * FROM hot_methods"
# or: duckdb -ui mydb.db (browser-based DuckDB UI)Flags for query:
| Flag | Meaning |
|---|---|
--csv |
Output as CSV |
-o FILE / --output FILE |
Write results to a file |
-n / --no-cache |
Skip writing the .db cache file |
Flags for serve:
| Flag | Meaning |
|---|---|
-p PORT / --port PORT |
Listen port (default 4244) |
--no-open |
Don't auto-open the browser |
--templates-dir DIR |
Extra template directory |
mvn clean package- Stack traces are stored at a fixed depth (10 frames by default) — no line numbers, bytecode indices, or frame types. This keeps the database compact and queries fast.
- JFR-specific datatypes are mapped to the closest DuckDB equivalent; some precision may be lost.
GPL-2.0 — Copyright 2017–2025 SAP SE or an SAP affiliate company and contributors.