[mbreiser]

Summary

The arena rig now runs the generic host monitor from rig-nodes and records the PC's "pressure" into every run's MCAP bag, so each run carries how loaded the PC was. The monitor publishes CPU per core (window mean + busiest 100 ms), P-/E-core class, sampler lateness, memory, page faults, disks, network, NVIDIA GPU (incl. NVENC) and the top processes, once a second on /rig/host.

The node itself, rig_nodes.host_monitor, and its messages, rig_interfaces/HostStats (+ DiskStats, NetStats, GpuStats, ProcStats), shipped in rig-nodes 0.3.0. They were prototyped on ww18; the prototype is on branch rig/host-monitor-port @ f5cf5a6, and the message spec is on docs/host-stats-request. This PR is only the arena half:

• rig/nodes/bridge_node.py (+13): records /rig/host into the bag on arrival. It uses the typed HostStats when rig_interfaces has it, else the String JSON fallback.
• rig/nodes/rig.py (+22): starts python -m rig_nodes.host_monitor --log-dir rig/runlogs --py-root rig/ by default. --py-root lets it name this workspace's nodes py:<script>.
  • With an older rig-nodes package it prints a note and starts without the monitor.
  • New flags: --no-host-monitor, --host-monitor-args.
• rig/tests/test_ros2_poc.py (+81): starts the monitor the same way and adds six checks. It skips them with an info line when the installed rig_nodes has no monitor. The checks:
  • samples recorded at ≥ 0.8 per second;
  • fields equal the installed HostStats (when typed);
  • one entry per CPU, with peak ≥ mean;
  • graph nodes pinned by script;
  • the monitor's own CPU ≤ 2 % of one core;
  • /rig/host ≤ 5 % of the run's compressed payload.
• rig/README.md: one row.

fictrac-bridge/bridge.py is untouched; nothing in rig/pixi.toml / pixi.lock changes.

Depends on the webDisplayTools re-pin to the rig-nodes 0.3.0 channel (and #274) for the monitor to actually start. Until then this PR is inert apart from the bridge subscription.

Tier

Next candidate (rig code; no web-tool change, no version bumps).

Release notes

none (rig-PC ROS graph only; no web tool changes)

Test plan

• pixi run test (root): 72/72 files
• cd rig && pixi run test on ww18 (Windows 11, Core Ultra 9 285K, 24 CPUs, RTX PRO 4500), model fly, no hardware:
  • with the installed rig_nodes 0.2.1: exit 0, host-monitor checks skipped with the info line
  • with the rig-nodes 0.3.0 source (release/0.3.0 @ 5e20dee) first on PYTHONPATH: exit 0, all 6 host-monitor checks pass. 24 samples / 25 s, 24 cores, py:bridge_node / py:fictrac_udp_adapter / py:host_monitor pinned, monitor CPU median 0.9 % of one core, /rig/host 4.8 % of the compressed payload (≈ 1 MB/h, String JSON)
• Typed, on ww18 from candidate 037a5cc (contains rig: pin rig-nodes 0.3.0 + re-lock with pixi 0.83 (#280, LAB-338) #281) with this branch merged in, after pixi self-update to 0.83 and pixi install --locked -e typed. pixi run -e typed test: /rig/host recorded as rig_interfaces/msg/HostStats (24 samples / 25 s, fields = the installed HostStats, 24 cores, graph nodes pinned, monitor CPU median 0.9 % of one core). Every check passes except the storage gate: 6.7 % of the compressed payload (1.77 MB/h).
  • Typed is smaller raw (1.8 KB vs 4.6 KB per sample), but float32 noise compresses worse than the String form's 3-decimal text.
  • Re-serializing the same samples with the floats rounded to binary steps brings it to 4.1 % (1.06 MB/h): 0..1 fractions and cores to 1/1024, rates and MB to 1/64. Coarser steps (1/256, 1/16) give 3.1 %.
  • The fix belongs in the node, so it's a rig-nodes 0.3.1 item. Until it lands the gate fails in the typed environment only; the default environment (String JSON, what pixi run rig uses) passes at 4.8 %.
  • The harness decodes typed /rig/host with rclpy's deserializer (db51439). The converter's pure CDR decoder reads no nested message arrays, and the converter never needs this topic.
• A real run: pixi run rig with FicTrac + the Studio, confirm /rig/host in the run's bag

🤖 Generated with Claude Code