Documentation Grafana Cloud Instrument and send data Grafana Alloy Reference Components pyroscope pyroscope.java

pyroscope.java

pyroscope.java continuously profiles Java processes running on the local Linux OS using async-profiler.

Usage

alloy Copy

pyroscope.java "LABEL" {
  targets    = TARGET_LIST
  forward_to = RECEIVER_LIST
}

Target JVM configuration

When you use pyroscope.java to profile Java applications, you can configure the target JVMs with some command line flags that ensure accurate profiling, especially for inlined methods. Add the following flags to your Java application’s startup command:

java Copy

-XX:+UnlockDiagnosticVMOptions -XX:+DebugNonSafepoints

For more details, refer to Restrictions/Limitations in the async-profiler documentation.

Additional Configuration for Linux Capabilities

If your Kubernetes environment has Linux capabilities enabled, configure the following in your Helm values to ensure pyroscope.java functions properly:

yaml Copy

alloy:
  securityContext:
    runAsUser: 0
    runAsNonRoot: false
    capabilities:
      add:
        - PERFMON
        - SYS_PTRACE
        - SYS_RESOURCE
        - SYS_ADMIN

These capabilities enable Alloy to access performance monitoring subsystems, trace processes, override resource limits, and perform necessary system administration tasks for profiling.

Arguments

The following arguments are supported:

Profiling behavior

The special label __process_pid__ must always be present in each target of targets and corresponds to the PID of the process to profile.

After component startup, pyroscope.java creates a temporary directory under tmp_dir and extracts the async-profiler binaries for both glibc and musl into the directory with the following layout.

Copy

/tmp/alloy-asprof-glibc-{SHA1}/bin/asprof
/tmp/alloy-asprof-glibc-{SHA1}/lib/libasyncProfiler.so
/tmp/alloy-asprof-musl-{SHA1}/bin/asprof
/tmp/alloy-asprof-musl-{SHA1}/lib/libasyncProfiler.so

After process profiling startup, the component detects libc type and copies according libAsyncProfiler.so into the target process file system at the exact same path.

targets argument

The special __process_pid__ label must always be present and corresponds to the process PID that’s used for profiling.

Labels starting with a double underscore (__) are treated as internal, and are removed prior to scraping.

The special label service_name is required and must always be present. If it’s not specified, pyroscope.scrape will attempt to infer it from either of the following sources, in this order:

1. __meta_kubernetes_pod_annotation_pyroscope_io_service_name which is a pyroscope.io/service_name pod annotation.
2. __meta_kubernetes_namespace and __meta_kubernetes_pod_container_name
3. __meta_docker_container_name
4. __meta_dockerswarm_container_label_service_name or __meta_dockerswarm_service_name

If service_name isn’t specified and couldn’t be inferred, then it’s set to unspecified.

Blocks

The following blocks are supported inside the definition of pyroscope.java:

profiling_config block

The profiling_config block describes how async-profiler is invoked.

The following arguments are supported:

Refer to profiler-options for more information about async-profiler configuration.

event argument

Sets the CPU profiling event:

• itimer - Default. Uses the setitimer(ITIMER_PROF) syscall, which generates a signal every time a process consumes CPU.
• cpu - Uses PMU-case sampling (like Intel PEBS or AMD IBS), can be more accurate than itimer, but it’s not available on every platform.
• wall - This samples all threads equally every given period of time regardless of thread status: Running, Sleeping, or Blocked. For example, this can be helpful when profiling application start-up time or IO-intensive processes.

per_thread argument

Sets per thread mode on async profiler. Threads are profiled separately and each stack trace will end with a frame that denotes a single thread.

The Wall-clock profiler (event=wall) is most useful in per-thread mode.

Exported fields

pyroscope.java doesn’t export any fields that can be referenced by other components.

Component health

pyroscope.java is only reported as unhealthy when given an invalid configuration. In those cases, exported fields retain their last healthy values.

Debug information

pyroscope.java doesn’t expose any component-specific debug information.

Debug metrics

pyroscope.java doesn’t expose any component-specific debug metrics.

Examples

Profile every java process on the current host

alloy Copy

pyroscope.write "staging" {
  endpoint {
    url = "http://localhost:4040"
  }
}

discovery.process "all" {
  refresh_interval = "60s"
  discover_config {
    cwd = true
    exe = true
    commandline = true
    username = true
    uid = true
    container_id = true
  }
}

discovery.relabel "java" {
  targets = discovery.process.all.targets
  rule {
    action = "keep"
    regex = ".*/java$"
    source_labels = ["__meta_process_exe"]
  }
}

pyroscope.java "java" {
  targets = discovery.relabel.java.output
  forward_to = [pyroscope.write.staging.receiver]
  profiling_config {
    interval = "60s"
    alloc = "512k"
    cpu = true
    sample_rate = 100
    lock = "1ms"
  }
}

Compatible components

pyroscope.java can accept arguments from the following components:

• Components that export Targets
• Components that export Pyroscope ProfilesReceiver