Resource consumption

The resource consumption manager samples edgelet stack and host usage against configured limits and publishes results to StatusReporter. It also participates in limit enforcement signaling for the agent.

Code: internal/resourceconsumption/

Purpose

• Sample control-plane RSS/CPU via gopsutil (process.MemoryInfo, process.Percent)
• When containerEngine=edgelet on an embedded build, include the --edgelet-containerd-child data-plane process(es)
• Compare stack totals against configured limits (bytes / CPU percentage)
• Update ResourceConsumptionManagerStatus on StatusReporter
• Collect initial sample immediately on start (no wait for first tick)

Metrics

Field	Meaning
agentCpu / agentMemory	Control-plane edgelet daemon process (cores / bytes): local API only
runtimeCpu / runtimeMemory	Embedded containerd child (embedded engine only; cores / bytes): local API only
edgeletStackCpu / edgeletStackMemory	Stack totals (cores / bytes): local API only
cpuUsage / memoryUsage	Edgelet stack total (agent + runtime when available): also sent to Controller in PUT status. Stack CPU uses per-core scale (100 = one logical CPU); human CLI prints stack breakdown in cores and hides duplicate cpuUsage when edgeletStackCpu is present.
diskUsage	Edgelet data directory usage in GiB (not host filesystem total)
systemCpus	Logical CPU count
systemOs / systemOsVersion / systemKernelVersion	GOOS family (linux, darwin, windows); Linux PRETTY_NAME or platform+version; Linux kernel only ("" on non-Linux)
systemTotalMemory / systemAvailableMemory	Host RAM capacity / free (bytes)
systemTotalDisk / systemAvailableDisk	Filesystem of diskDirectory total / free (bytes)
systemTotalCpu	Host CPU busy 0-100% (not core count)

External docker / podman engines report agent-only stack totals (no runtime child tracking).

Dependencies

Depends on	Reason
statusreporter	Publish usage metrics
config	Limits and poll frequency
gopsutil	Process and host metrics
pkg/containerd	Discover embedded containerd child PIDs on Linux

Used by	Reason
supervisor	Started early (before Field Agent)

Lifecycle

Start

1. InstanceConfigUpdated(). Load limits from config
2. collectUsageData() immediately
3. Start periodic worker at configured frequency

Config update

InstanceConfigUpdated() refreshes disk/CPU/memory limits and may restart sampling logic.

Configuration

Limits loaded from config profile (typical keys):

Key	Unit
Disk limit	bytes
Memory limit	bytes
CPU limit	stack cores × 100 (5-400; default 80 = 0.8 CPU; monitor-only)

Exact YAML names match config.yaml profiles. See default config in internal/config/default_config.yaml.

Module status

Property	Value
StatusReporter index	0 (utils.ResourceConsumptionManager)
First slot in modulesStatus[]	Resource Consumption

Status fields include stack breakdown plus legacy memoryUsage, cpuUsage, diskUsage.

External APIs

Exposed indirectly via GET /v1/system/status resource section and Controller status POST.

Observability

• Log module: "Resource Consumption Manager"
• Initial and periodic debug logs with MiB/GiB formatting
• Warn when multiple embedded containerd child PIDs are detected

Failure modes

Symptom	Typical cause
Zero usage reported	gopsutil error on platform
runtimeDegraded=true	Embedded engine configured but containerd child not running
Limit warnings	Stack usage exceeded configured thresholds

Code map

File	Role
manager.go	Sampling loop, limit comparison, status updates
stats.go	Parallel process/host sampling and CPU smoothing
runtime_linux.go	Embedded runtime PID discovery
host_cpu_linux.go	Linux host CPU fallback

Related: Status reporter, Supervisor.

See anything wrong with the document? Help us improve it!

Edit this page on GitHub!