Skip to main content
This guide walks you through profiling a Java application and analyzing the results with ap-query.

​
Prerequisites

​
Profile your application

Use asprof to collect a profile. The most common profiling scenarios:
Profile CPU usage for 30 seconds:
asprof -d 30 -o jfr -f profile.jfr <pid>
CPU profiling shows where your application spends time executing code.
Replace <pid> with your Java process ID. Find it with jps or ps aux | grep java.

​
Analyze with ap-query

​
Start with triage

The info command gives a comprehensive overview: 
ap-query info profile.jfr

Profile: profile.jfr
Duration: 30.0s
Event: cpu (12,450 samples)
Total: 12,450 samples

Top 20 hot methods (self time):
    12.3%   1,532   sun.nio.ch.EPollArrayWrapper.poll
     8.7%   1,083   java.util.HashMap.resize
     6.2%     772   com.example.Parser.parseToken
     4.8%     598   java.lang.String.split
     ...

Thread distribution:
    45.2%   5,627   http-nio-8080-exec-
    32.1%   3,997   kafka-consumer-thread-
    12.8%   1,594   scheduling-thread-
     9.9%   1,232   background-worker-

​
Find hot methods

Rank methods by self and total time: 
ap-query hot profile.jfr --top 20
Output shows methods sorted by self time (time spent in the method itself) and total time (including callees).

​
Drill into a method

See what a hot method calls: 
ap-query tree profile.jfr -m HashMap.resize --depth 6

HashMap.resize (1,083 samples, 8.7%)
├─ HashMap.putVal (892 samples, 7.2%)
│  ├─ HashMap.put (892 samples, 7.2%)
│  │  ├─ CacheManager.store (445 samples, 3.6%)
│  │  │  └─ RequestHandler.process (445 samples, 3.6%)
│  │  └─ UserService.addUser (447 samples, 3.6%)
│  │     └─ UserController.createUser (447 samples, 3.6%)
└─ [other callers] (191 samples, 1.5%)

​
Find the hottest path

Trace from a method to the leaf: 
ap-query trace profile.jfr -m HashMap.resize
Shows the single most expensive path through your code.

​
See who calls a method

Find caller paths: 
ap-query callers profile.jfr -m HashMap.resize
Inverted call tree showing which methods call into HashMap.resize.

​
Analyze over time

View sample distribution across time: 
ap-query timeline profile.jfr

Timeline (30.0s, 15 buckets)
  0.0s-2.0s       423  ████████████░░░░░░░░  HashMap.resize (12%)
  2.0s-4.0s       512  ███████████████░░░░░  Parser.parseToken (15%)
  4.0s-6.0s       398  ████████████░░░░░░░░  String.split (9%)
  6.0s-8.0s       445  █████████████░░░░░░░  EPoll.poll (8%)
  ...
 28.0s-30.0s      387  ███████████░░░░░░░░░  HashMap.resize (11%)

​
Common workflows

​
Compare before and after

Profile before and after an optimization: 
# Profile before
asprof -d 30 -o jfr -f before.jfr <pid>

# Make your optimization
# ...

# Profile after
asprof -d 30 -o jfr -f after.jfr <pid>

# Compare
ap-query diff before.jfr after.jfr --min-delta 0.5
The diff output shows REGRESSION (slower), IMPROVEMENT (faster), NEW, and GONE methods.

​
Analyze a time window

Focus on a specific time range: 
# First, see the timeline
ap-query timeline profile.jfr

# Then zoom into a spike at 12-14 seconds
ap-query hot profile.jfr --from 12s --to 14s

​
Filter by thread

Analyze a specific thread pool: 
# See thread distribution
ap-query threads profile.jfr

# Focus on HTTP threads
ap-query hot profile.jfr -t "http-nio"

​
Remove idle samples

For wall-clock profiles, filter out idle waiting: 
ap-query hot profile.jfr --event wall --no-idle
The --no-idle flag removes samples where threads are blocked on I/O, sleep, or park operations.

​
Agent-driven workflow

If you installed the agent skill (installation guide), you can ask your coding agent to handle all of this: 
Profile this application with async-profiler and analyze the result.
Start with hotspot triage, then drill down and report concrete next actions.
The agent will:
  1. Profile your application with asprof
  2. Run ap-query info for triage
  3. Drill into hot methods with tree, trace, and callers
  4. Provide specific optimization recommendations

​
Next steps

Command reference

Explore all commands and flags

Profiling workflow

Learn the complete analysis workflow

Event types

Understand CPU, wall, alloc, and lock events

Scripting

Write custom analysis scripts

Build docs developers (and LLMs) love

Get started for freeTalk to us