andygrove commented on issue #1894: URL: https://github.com/apache/datafusion-comet/issues/1894#issuecomment-4178475112
Here are a few approaches depending on what you're trying to measure: ### Measuring Comet's off-heap memory usage **Option 1: Tracing with jemalloc (recommended for overall native memory)** Build Comet with jemalloc enabled: ``` make release COMET_FEATURES="jemalloc" ``` Then enable tracing: ``` spark.comet.tracing.enabled=true ``` This generates a `comet-event-trace.json` file (Chrome Trace Event Format) that includes: - `jemalloc_allocated` — total native memory allocated by the executor process - `task_memory_comet_NNN` — off-heap memory allocated by Comet for query execution - `comet_shuffle_NNN` — off-heap memory allocated for columnar shuffle You can view this in `chrome://tracing` or [Perfetto](https://ui.perfetto.dev/). **Option 2: Debug memory logging** Set `spark.comet.debug.memory=true` to log all memory pool reservation grow/shrink operations. This gives fine-grained visibility into DataFusion memory pool usage. ### Profiling native threads For CPU profiling of native code, [cargo-flamegraph](https://github.com/flamegraph-rs/flamegraph) works well for benchmarks. For profiling within a Spark job, async-profiler (which you're already using) is the right tool — it can capture both JVM and native frames in the same flamegraph. For more details, see our contributor guides: - [Tracing](https://datafusion.apache.org/comet/contributor-guide/tracing.html) - [Profiling Native Code](https://datafusion.apache.org/comet/contributor-guide/profiling_native_code.html) - [Debugging](https://datafusion.apache.org/comet/contributor-guide/debugging.html) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
