Documentation
¶
Overview ¶
Package system encapsulates host-level resource probing used by the WebSocket system stats endpoint: cgroup limits and GPU detection/stats. The types here own their own caching and synchronization so handlers can hold them as plain fields.
Index ¶
Constants ¶
const AMDGPUSysfsPath = "/sys/class/drm"
AMDGPUSysfsPath is the sysfs base used to discover AMD GPUs.
Variables ¶
This section is empty.
Functions ¶
Types ¶
type CgroupCache ¶
type CgroupCache struct {
// contains filtered or unexported fields
}
CgroupCache memoizes the result of cgroup limit detection for a configurable TTL. A single in-flight refresh is shared across concurrent callers (singleflight semantics via the write lock), so a stampede of system-stats samplers never duplicates the work.
func NewCgroupCache ¶
func NewCgroupCache(ttl time.Duration) *CgroupCache
NewCgroupCache returns a cache that caches detection results for ttl using the default docker.DetectCgroupLimits detector.
func NewCgroupCacheWithDetector ¶
func NewCgroupCacheWithDetector(ttl time.Duration, detector func() (*docker.CgroupLimits, error)) *CgroupCache
NewCgroupCacheWithDetector returns a cache backed by a custom detector. Useful when callers (e.g. tests) need to control detection behaviour or simulate failures.
func (*CgroupCache) Get ¶
func (c *CgroupCache) Get() *docker.CgroupLimits
Get returns the cached cgroup limits, refreshing if the entry is older than the TTL. Returns nil when no cgroup limits have been detected (host is not running under cgroups).
type GPUMonitor ¶
type GPUMonitor struct {
// contains filtered or unexported fields
}
GPUMonitor probes for an attached GPU (NVIDIA / AMD / Intel) and reports VRAM usage. Detection is cached for gpuDetectionTTL; once a vendor is detected, subsequent Stats calls invoke the vendor-specific tool directly.
func NewGPUMonitor ¶
func NewGPUMonitor(enabled bool, configuredType string) *GPUMonitor
NewGPUMonitor creates a monitor. enabled gates Stats; when false, Stats returns (nil, nil). configuredType is the user-pinned vendor ("nvidia"|"amd"|"intel"|"auto"|"") — anything else falls back to auto-detection.
func (*GPUMonitor) Enabled ¶
func (m *GPUMonitor) Enabled() bool
Enabled reports whether GPU monitoring is on.
func (*GPUMonitor) Stats ¶
func (m *GPUMonitor) Stats(ctx context.Context) ([]systemtypes.GPUStats, error)
Stats returns per-GPU VRAM stats. Returns (nil, 0, nil) when monitoring is disabled or no GPU is detected; vendor-specific errors are propagated otherwise.
Source Files