Resource Limits

Runix can enforce CPU and memory limits on processes using Linux cgroups v2.

Implementation

Located at internal/cgroups/:

File	Responsibility
manager.go	Manager interface and helper functions
manager_linux.go	Linux cgroups v2 implementation
manager_other.go	No-op implementation for non-Linux platforms

Configuration

processes:
  api:
    entrypoint: ./cmd/api
    resources:
      cpu_quota: "50%"
      memory_limit: "512MB"

  heavy-worker:
    entrypoint: worker.py
    resources:
      cpu_quota: "200%"
      memory_limit: "2GB"

Fields

Field	Type	Description
cpu_quota	string	CPU limit as percentage (e.g., "50%", "200%")
memory_limit	string	Memory limit with unit (e.g., "512MB", "1GB", "1024KB")

Memory Units

Unit	Multiplier
KB	1024
MB	1,048,576
GB	1,073,741,824

Platform Support

Platform	Support
Linux (cgroups v2)	Full — CPU and memory limits enforced
macOS	Not supported — returns error if limits are configured
Windows	Not supported — returns error if limits are configured

How It Works (Linux)

cgroup Structure

/sys/fs/cgroup/runix/<pid>/
├── cgroup.procs    # Process membership
├── cpu.max         # CPU quota and period
└── memory.max      # Memory limit in bytes

CPU Quota

CPU quota is converted from percentage to microseconds per 100ms period:

Config	cpu.max
"50%"	50000 100000
"100%"	100000 100000
"200%"	200000 100000

Memory Limit

Memory limit is converted to bytes and written to memory.max:

Config	Bytes
"512MB"	536870912
"1GB"	1073741824

Lifecycle

1. On start: Create cgroup at /sys/fs/cgroup/runix/<pid>/, add PID, set limits
2. On stop: Remove the cgroup directory
3. Stats: Read memory.current for current memory usage

No-op Implementation

On non-Linux platforms, the noopManager returns an error if limits are configured:

func (m *noopManager) Apply(pid int, limits Limits) error {
    if limits.CPUQuota != "" || limits.MemoryLimit != "" {
        return fmt.Errorf("resource limits require cgroups v2 (Linux only)")
    }
    return nil
}

What's Next

• Configuration Reference — Resource config fields
• Metrics — Per-process resource monitoring