Profiling configuration options

If enabled, the apm agent will correlate it’s transaction with the profiling data from elastic universal profiling running on the same host.

The feature needs to buffer ended local-root spans for a short duration to ensure that all of its profiling data has been received.This configuration option configures the buffer size in number of spans. The higher the number of local root spans per second, the higher this buffer size should be set. The agent will log a warning if it is not capable of buffering a span due to insufficient buffer size. This will cause the span to be exported immediately instead with possibly incomplete profiling correlation data.

The extension needs to bind a socket to a file for communicating with the universal profiling host agent.This configuration option can be used to change the location. Note that the total path name (including the socket) must not exceed 100 characters due to OS restrictions. If unset, the value of the java.io.tmpdir system property will be used.

Set to true to make the agent create spans for method executions based on async-profiler, a sampling aka statistical profiler.

Due to the nature of how sampling profilers work, the duration of the inferred spans are not exact, but only estimations. The profiling_inferred_spans_sampling_interval lets you fine tune the trade-off between accuracy and overhead.

The inferred spans are created after a profiling session has ended. This means there is a delay between the regular and the inferred spans being visible in the UI.

Only platform threads are supported. Virtual threads are not supported and will not be profiled.

By default, async profiler prints warning messages about missing JVM symbols to standard output. Set this option to false to suppress such messages

The frequency at which stack traces are gathered within a profiling session. The lower you set it, the more accurate the durations will be. This comes at the expense of higher overhead and more spans for potentially irrelevant operations. The minimal duration of a profiling-inferred span is the same as the value of this setting.

Supports the duration suffixes ms, s and m. Example: 50ms.

The minimum duration of an inferred span. Note that the min duration is also implicitly set by the sampling interval. However, increasing the sampling interval also decreases the accuracy of the duration of inferred spans.

Supports the duration suffixes ms, s and m. Example: 0ms.

If set, the agent will only create inferred spans for methods which match this list. Setting a value may slightly reduce overhead and can reduce clutter by only creating spans for the classes you are interested in. Example: org.example.myapp.*

This option supports the wildcard *, which matches zero or more characters. Examples: /foo/*/bar/*/baz*, *foo*. Matching is case insensitive by default. Prepending an element with (?-i) makes the matching case sensitive.

Excludes classes for which no profiler-inferred spans should be created.

This option supports the wildcard *, which matches zero or more characters. Examples: /foo/*/bar/*/baz*, *foo*. Matching is case insensitive by default. Prepending an element with (?-i) makes the matching case sensitive.

Profiling requires that the async-profiler shared library is exported to a temporary location and loaded by the JVM. The partition backing this location must be executable, however in some server-hardened environments, noexec may be set on the standard /tmp partition, leading to java.lang.UnsatisfiedLinkError errors. Set this property to an alternative directory (e.g. /var/tmp) to resolve this. If unset, the value of the java.io.tmpdir system property will be used.