Heap sampling

• How heap sampling works
• Advantages
• Limitations
• Performance overhead

How heap sampling works

Heap sampling, introduced in Java 11 (JEP 331), is a JVM feature that enables low-overhead, probabilistic tracing of object allocations. It periodically samples allocations on the Java heap, capturing stack traces and metadata (such as object size and type) for a subset of allocated objects. This is achieved using the JVM's memory allocation mechanisms, with a configurable sampling interval that determines how often allocations are recorded.

Advantages

1. It provides the same level of detail as bytecode instrumentation (including thread information, full stack traces of object creation, and object size) but with lower overhead.
2. Heap sampling is less invasive because it does not require bytecode instrumentation. It does not increase bytecode size or slow down class loading.
3. This is especially useful in attach mode, as it prevents the pause that occurs on the first attempt to start object allocation profiling due to instrumenting classes loaded before the agent attaches.

Limitations

Because heap sampling only collects data on a fraction of the total allocations — rather than capturing every object allocation — it may miss objects. This can lead to partial or skewed insights into which code paths are responsible for the bulk of allocations.

Performance overhead

The overhead of heap sampling directly depends on the number of objects for which the profiler records a stack trace. The fewer stack traces the profiler records, the lower the overhead.

You can control overhead using the sampling interval parameter. The larger the sampling interval, the fewer samples are recorded and the lower the overhead. Conversely, the smaller the sampling interval, the higher the overhead.