feat: show dataplane CPU of the packet-processing cores

[tbortolossi]

Problem

During a PA-5580 load test, the API and SNMP dashboards both showed 45 % dataplane CPU while the packet-processing cores were at about 72 %.

A probe on the firewall showed why. Each dataplane lists 128 cores in resource-monitor, but only the 80 pan task cores (1–80) process packets. Core 0 and cores 81–127 stay at exactly 0 % in both the minute average and the minute maximum. PAN-OS averages over all 128 cores in the XML API and in SNMP hrProcessorLoad alike: the two were measured within 0.3 point of each other. The all-core figure is correct, but it hides how close the working cores are to saturation.

Change

• Collector: each dataplane's core=average point gets cpu_active_pct, the mean of the cores with any load during the minute (by maximum when available), and active_cores. An idle dataplane reports 0.0. These are new fields only; no existing field changes type.
• API dashboards (standard and chassis):
  • a new DP CPU (active cores) tile next to DP CPU (avg). The strip now holds 9 tiles; MP CPU, MP RAM and Session Table are narrower;
  • a Dataplanes (active cores) line in CPU MP / DP;
  • DP CPU (active cores) in the Load Test ramp and the CPU vs Throughput scatter plot.
• The all-core value is unchanged and still matches SNMP.
• README and AGENTS.md explain both figures.

Verification

• The real PA-5580 captures replayed through the new parser give 80 active cores per dataplane at 71–73 %, against 45 % all-core.
• Tests: 281 OK. New tests cover the PA-5580 case, the idle and no-maximum fallbacks, and the 9-tile strip filling 24 columns edge to edge.
• Every query of both API dashboards ran against the local InfluxDB: the active-core queries compile, and Grafana reloaded both dashboards with the new strip.

Operator impact

Rebuild the Telegraf image (./generate.sh). The new tile and lines fill in from the first poll after the upgrade.

🤖 Generated with Claude Code