Collect performance data reported by an Alibaba Cloud Java agent from Java programs

Step 2: Download a Java agent

• Regions in China

  wget https://logtail-release-cn-hangzhou.oss-cn-hangzhou.aliyuncs.com/jvm/continuous-profile-collector-agent-1.9.0.jar

• Regions outside China

  wget https://logtail-release-ap-southeast-1.oss-ap-southeast-1.aliyuncs.com/jvm/continuous-profile-collector-agent-1.9.0.jar

Step 3: Configure a Java program to push performance data

• Configure the Java program by using JVM parameters

java \
-Dprofiling.app.name=your_service_name \
-Dprofiling.agent.upload.server="http://{host}:{port}" \
-Dprofiling.cpu.engine={engine} \
-javaagent:{path for javaagent} \
-jar demo.jar

Parameter	Description

Parameter	Description
profiling.app.name	The name of the service.
profiling.agent.upload.server	The address for data upload. If you use an ECS instance, specify the value in the following format: IP address of the ECS instance:4040. If you use a server that resides in a Kubernetes cluster, set the value to logtail-kubernetes-metrics.sls-monitoring:4040. If you use a server that is from a third-party cloud service provider or a data center, specify the value in the following format: IP address of the server:4040.
profiling.cpu.engine	The engine used for CPU hotspot monitoring. Default value: off. Valid values: auto, async_profiler, jfr, and off. The value off specifies that CPU hotspot monitoring is disabled. Other values specify that CPU hotspot monitoring is enabled. We recommend that you set the value to auto.

• Configure the Java program by using environment variables

export PROFILING_APP_NAME="your_service_name"
export PROFILING_AGENT_UPLOAD_SERVER="http://{host}:{port}"
export PROFILING_CPU_ENGINE="{engine}"
export PROFILING_ALLOC_ENGINE="{engine}"

Parameter	Description

Parameter	Description
PROFILING_APP_NAME	The name of the service.
PROFILING_AGENT_UPLOAD_SERVER	The address for data upload. If you use an ECS instance, specify the value in the following format: IP address of the ECS instance:4040. If you use a server that resides in a Kubernetes cluster, set the value to logtail-kubernetes-metrics.sls-monitoring:4040. If you use a server that is from a third-party cloud service provider or a data center, specify the value in the following format: IP address of the server:4040.
PROFILING_CPU_ENGINE	The engine used for CPU hotspot monitoring. Default value: off. Valid values: auto, async_profiler, jfr, and off. The value off specifies that CPU hotspot monitoring is disabled. Other values specify that CPU hotspot monitoring is enabled. We recommend that you set the value to auto.

Remarks

JVM parameter	Environment variable	Description

JVM parameter	Environment variable	Description
profiling.app.name	PROFILING_APP_NAME	The name of the application.
profiling.agent.upload.server	PROFILING_AGENT_UPLOAD_SERVER	The address of the server to which the Java Flight Recorder (JFR) file is uploaded. Default value: http://localhost:4040. Do not start the address with http. The system automatically adds the http prefix to the address. Do not end the address with a forward slash (/). The system automatically appends a forward slash (/) to the address.
profiling.agent.timeout	PROFILING_AGENT_TIMEOUT	The timeout period for uploading the JFR file. Default value: 10. Unit: seconds.
profiling.agent.ingest.max.tries	PROFILING_AGENT_INGEST_MAX_TRIES	The maximum number of retries that are allowed for uploading the JFR file. Default value: 2.
profiling.app.http.headers	PROFILING_APP_HTTP_HEADERS	The HTTP header that is used when you upload the JFR file. This parameter is empty by default. Example: SESSION_ID=1111;XXX=YYY.
profiling.app.labels	PROFILING_APP_LABELS	The tag that is added to the JFR file when you upload the JFR file. This parameter is empty by default. Example: env=dev;lang=java;biz=member.
profiling.agent.log.level	PROFILING_AGENT_LOG_LEVEL	The log level. Default value: info. Valid values: info, debug, and error.
profiling.agent.log.file	PROFILING_AGENT_LOG_FILE	The path to the log file. You can set the value to /path/to/profiling.log. By default, the path is written to Java stdout and stderr.
profiling.period	PROFILING_PERIOD	The interval at which performance data is uploaded. Default value: 1. Unit: minutes.
profiling.delay	PROFILING_DELAY	The performance monitoring latency. Default value: 0, which indicates that performance monitoring starts immediately after the performance monitoring engine is enabled. If you set the value to N, performance monitoring starts N seconds after the performance monitoring engine is enabled.
profiling.start.at.zero.second	PROFILING_START_AT_ZERO_SECOND	Specifies whether to start performance monitoring at the 0th second of every minute. If you want to start performance monitoring at the 0th second of every minute, set the value to true. For example, if the value is set to true and the current time is 30 seconds of the current minute, the system automatically waits for 30 seconds before it starts performance monitoring. Default value: false.
profiling.compression.mode	PROFILING_COMPRESSION_MODE	The compression mode. Default value: none. Valid values: gzip and none. none: The file is not compressed and is suffixed with .jfr. gzip: The file is compressed and is suffixed with .jfr.gzip.
profiling.trigger.mode	PROFILING_TRIGGER_MODE	The trigger mode. You can trigger periodic or one-time performance monitoring. Default value: periodic. Valid values: periodic and api. We recommend that you set the value to periodic in agent mode.
profiling.output.format	PROFILING_OUTPUT_FORMAT	The format of the file. Default value: jfr. Valid values: jfr and collapsed.
profiling.cpu.engine	PROFILING_CPU_ENGINE	The engine used for CPU hotspot monitoring. Default value: off. Valid values: auto, async_profiler, jfr, and off. The value off specifies that CPU hotspot monitoring is disabled. Other values specify that CPU hotspot monitoring is enabled. We recommend that you set the value to auto.
profiling.cpu.interval	PROFILING_CPU_INTERVAL	The interval at which CPU hotspot monitoring is performed. A small value increases the overhead. Default value: 10. Unit: milliseconds.
profiling.wallclock.engine	PROFILING_WALLCLOCK_ENGINE	The engine used for the monitoring of wall clock hotspots. Default value: off. Valid values: auto, async_profiler, and off. The value off specifies that the monitoring of wall clock hotspots is disabled. Other values specify that the monitoring of wall clock hotspots is enabled. We recommend that you set the value to off.
profiling.wallclock.interval	PROFILING_WALLCLOCK_INTERVAL	The interval at which the monitoring of wall clock hotspots is performed. A small value increases the overhead. Default value: 20. Unit: milliseconds.
profiling.wallclock.thread.filter	PROFILING_WALLCLOCK_THREAD_FILTER	The thread filter used for the monitoring of wall clock hotspots. Default value: 0, which indicates that no threads are involved. The following list provides examples on how to specify values: Empty: "" Single thread: 123 Multiple threads: 122,123 Thread range: 122 to 134
profiling.wallclock.threads.per.tick	PROFILING_WALLCLOCK_THREADS_PER_TICK	The maximum number of threads used to monitor wall clock hotspots. Default value: 8.
profiling.alloc.engine	PROFILING_ALLOC_ENGINE	The engine used for Alloc hotspot monitoring. Default value: off. Valid values: auto, async_profiler, jfr, and off. Alloc hotspot monitoring refers to the monitoring of memory request hotspots. The value off specifies that Alloc hotspot monitoring is disabled. Other values specify that Alloc hotspot monitoring is enabled. We recommend that you set the value to auto.
profiling.alloc.interval	PROFILING_ALLOC_INTERVAL	The interval at which Alloc hotspot monitoring is performed. A small value increases the overhead. Default value: 256. Unit: kilo bytes.
profiling.jfr.max.size	PROFILING_JFR_MAX_SIZE	The upper limit of the size for the JFR file. If the size reaches the upper limit, the data in the file is automatically discarded. Default value: 64m. Example values: 256k and 10m.
profiling.jfr.max.age	PROFILING_JFR_MAX_AGE	The upper limit of the age for the JFR file. If the age reaches the upper limit, the data in the file is automatically discarded. Default value: 10m. Example values: 1m, 1h, and 1d.
profiling.jfr.max.stack.depth	PROFILING_JFR_MAX_STACK_DEPTH	The maximum stack depth that is allowed during JFR sampling. Default value: 64.