Using Java Flight Recorder
The Java Flight Recorder (JFR) feature allows you to record JVM, system, and Java level events to monitor the behavior and troubleshoot problems in your Java applications. Keeping the overhead of the recording process low, Azul Zulu Prime Builds of OpenJDK (Azul Zulu Prime JVM) Java Flight Recorder records a large amount of information about your system from command-line parameters and system properties to time spent executing a piece of code.
JFR is available in Azul Zulu Prime JVM without the need for command-line options to enable it.
To analyze the JVM or your application, you can either interactively connect to a Java process from the Azul Mission Control desktop application or start a JFR recording on the command line with the
JFR Recording with jcmd
On the same host and userid as the java process under analysis, run the following command to start a recording:
jcmd JVMID JFR.start filename=rec.jfr
JVMID is either the process ID or the classname assigned with the java process as shown when running
jcmd without parameters.
To stop a recording, use:
jcmd JVMID JFR.stop name=1
To start a recording with more detailed metrics, use the following instead:
jcmd JVMID JFR.start filename=rec.jfr settings=profile
Analyzing a JFR Recording
It is recommended to open and read Azul Zulu Prime JVM Java Flight recordings using Azul Mission Control tool available for download from the Azul website.
Alternatively, a Java Flight recording can be viewed using the
jfr command-line tool.
Example usage of the
$JAVA_HOME/bin/jfr print rec.jfr > file.txt
Using the Azul Mission Control Desktop Application
With the Azul Mission Control application, you can also start and stop recording or interactively browse metrics of the java process under analysis.
By default, JFR uses a Unix socket for communication and therefore does not offer remote connections. To allow remote connection, you need to enable the JVM’s JMX connector.
See "Find JVM to connect" to in the Zulu® Mission Control Quick Start Guide for details.
Starting a Recording at JVM Start Time
For detailed performance analysis, you might prefer to start a JFR recording directly when launching a java process.
To enable recording, add either of the following command-line options to the java process under analysis:
-XX:StartFlightRecording=filename=rec.jfr- for an on-disk recording
-XX:StartFlightRecording=disk=false,filename=rec.jfr- for an in-memory recording
More Java Flight Recorder command-line options are listed in Using ZVM Command-Line Options.
Using Zing Live Objects
Zing Live Objects help determine the type of objects that are leaking while troubleshooting memory leaks. To identify and address a memory leak using information gathered in a Zing flight recording, complete the following steps.
To create a Zing flight recording for inspection, use the following arguments:
The Flight Recorder has two configuration files: default and profile.
To enable the Live Objects feature, you need to enable ZingOldObjectsEvent in the configuration file. By default, ZingOldObjectsEvent is enabled only in the profile configuration file in versions of Zing prior to 22.214.171.124. Starting with ZVM 126.96.36.199, ZingOldObjectEvent is disabled by default in all configurations.
By default, the amount of live set events recorded by Java Flight Recorder is set to 99%. To decrease the recording file size, you can set the value of the
JFRDescribeLiveSetPercentage command-line option to a number less than 99:
java -XX:JFRDescribeLiveSetPercentage=90 -XX:StartFlightRecording=disk=false,duration=<recording_time_duration>,settings=profile,filename=<recording_filename>.jfr
When a recording is started, the output should look similar to the snippet below:
$ /home/username/<JDK_installation_path>/bin/java ‑XX:StartFlightRecording=disk=false,settings=profile,filename=myrecording.jfr Started recording 1. Use jcmd 32747 JFR.dump name=1 to copy recording data to file.
If just a filename without a path is specified (i.e.,
filename=<recording_filename>.jfr), the flight recording is saved to the directory from which you ran the command.
To specify the path of your choice, use:
Once you create a Zing Flight recording, start the Azul Mission Control tool and open your Azul Zulu Prime JVM Flight recording from the File menu.
If you do not have Azul Mission Control on your machine, download the installer file from Download Azul Mission Control and see Downloading and Starting Azul Mission Control for installation instructions if needed.
In the Outline navigation tree, select Memory > Zing Live Objects to display the set of live objects captured in the recording.
Study and compare GC cycles in the table with garbage collection identifiers (GC ID) for a live set suddenly increasing in size which indicates a memory leak.
Live objects types marked during each selected GC cycle display in the table to the right when a specific GC cycle is chosen.
Sort out live objects in a GC cycle by the total size to track the live object type with the biggest size and select a potentially leaking object type in the Referent column.
All items that refer to a potentially leaking object type display in the Referrer column in the bottom table. The latter helps identify the source of the leaking object by reference count.
(Optional) To retrieve more information, repeat step 5 if needed.
When a potentially leaking class is identified on the Zing Live Objects sub-tab, use the Memory tab to get information about the allocation site. The Memory tab contains allocation sites, which lead to allocation in a new thread local allocation buffer (TLAB) or allocation outside a TLAB, sorted by type. Checking stack traces for a potentially leaking class can help to stop the real place in code where the leak occurs.
In case the Memory tab does not help to find the leak, but the leak is still suspected, the Method Profiling tab might be used for this purpose. It is less convenient compared to the Memory tab, as it shows method calls of concrete types, not allocation sites.
Usually, the Zing Live Objects sub-tab and the Memory tab are all you need to find a leak.
Inspect the Stack Trace of your candidates to track where most memory allocation pressure comes from.
Using TTSP JFR Events
Azul Platform Prime 20.07.0.0 introduces support for profiling Time To Safepoint (TTSP) issues based on data about the Azul Zulu Prime JVM events collected in a Java Flight Recording (JFR).
To profile an existing TTSP issue using information gathered in a flight recording, complete the following steps:
Create a flight recording for inspection using the following arguments:java -XX:StartFlightRecording=name=<recording_name>,filename=<recording_filename>.jfr,settings=profile
The Flight Recorder has
profileconfiguration files. By default, the
TimeToSafepointIssueevent, which activates TTSP issue profiling, is enabled in all configurations. Use the following options to configure the event settings file for the Java Flight Recorder:
Configuring Option Description Default
Generates asynchronous signals to log safepoint issues.
Specifies the number of real time signals used for safepoint profiling.
Specifies the number of milliseconds to start safepoint profiling.
Specifies the threshold percentage value to start safepoint profiling.
Specifies the interval percentage value for safepoint profiling.
Specifies the number of safepoint profiler buffers to track.
Specifies the log file to write safepoint profiling data to.
The profile settings may increase performance overhead due to collecting more data by the Java Flight Recorder. Customizing any settings file, either default or profile, may incur increased performance overhead.
Once you create a Zing Flight recording, start the Azul Mission Control tool and open your Zing Flight recording from the File menu.
If you do not have Azul Mission Control on your machine, download the installer file from Download Azul Mission Control and see Install and Run Azul Mission Control for installation instructions if needed.
In the Outline navigation tree, choose Event Browser.
In the Event Types Tree, select Java Virtual Machine > Runtime > Safepoint to display the set of Safepoint event types captured in the recording.
In case of a TTSP issue, check the Event Browser table.
The contents of the Stack Trace window may help you understand the cause of the issue.