denet: a streaming process monitor

denet /de.net/ v. 1. Turkish: to monitor, to supervise, to audit. 2. to track metrics of a running process.

Denet is a lightweight streaming process monitor. It tracks CPU, memory, I/O, and threads for a running process (and its children), with adaptive sampling and optional GPU / eBPF support.

Installation

pip install denet    # Python package
cargo install denet  # Rust binary

Or install the conda package from the almost-conductor channel:

# pixi
pixi workspace channel add https://prefix.dev/almost-conductor
pixi add denet
# conda
conda install -c https://prefix.dev/almost-conductor denet

Optional features for the Rust binary:

cargo install denet --features gpu   # NVIDIA GPU monitoring (requires drivers; enable per run with --gpu)
cargo install denet --features ebpf  # eBPF profiling (Linux only, requires clang)

Usage

# Run a command and monitor it
denet run python train.py

# Attach to an existing process
denet attach 1234

# Save metrics as JSONL
denet --json --out metrics.jsonl run python train.py

# Turn a saved run into a report (timelines, regime detection, eBPF syscalls)
pip install 'denet[report]'
denet-report metrics.jsonl -o report.html

CPU usage follows the top convention: 100% = one fully utilized core, so a 4-core workload shows 400%.

Sampling starts at 100 ms and ramps up to 1 s after 10 s of runtime (adaptive). Override with -i / -m:

denet -i 50 -m 500 run python train.py

Further reading

License

GPL-3

Acknowledgements

• sysinfo - Rust library for system information
• PyO3 - Rust bindings for Python
• aya - Rust library for eBPF programs