config

Documentation

Index

• type CollectionConfig
• type CollectorCfg
• type Config
  • func Default() *Config
  • func Load(path string) (*Config, error)
• type FaultSubConfig
• type FaultSubDefaults
• type HealthConfig
• type ServerConfig
• type SnapshotConfig
• type StorageConfig
• type StragglerOutputConfig

Types

type CollectionConfig

type CollectionConfig struct {
	MinPriority string `yaml:"min_priority"` // low | medium | high
}

CollectionConfig controls which metrics are collected (pre-filter by priority).

type CollectorCfg

type CollectorCfg struct {
	Enabled  bool          `yaml:"enabled"`
	Interval time.Duration `yaml:"interval"`
}

CollectorCfg holds per-collector configuration.

type Config

type Config struct {
	Server          ServerConfig            `yaml:"server"`
	Collectors      map[string]CollectorCfg `yaml:"collectors"`
	Storage         StorageConfig           `yaml:"storage"`
	Health          HealthConfig            `yaml:"health"`
	Stress          stress.Config           `yaml:"stress"`
	Collection      CollectionConfig        `yaml:"collection"`
	Features        []string                `yaml:"features"` // enabled features; daemon loads features/<name>/metrics.yaml overrides + derives C_comp from their intervals
	FaultSub        FaultSubConfig          `yaml:"faultsub"`
	StragglerOutput StragglerOutputConfig   `yaml:"straggler_output"`
	Snapshot        SnapshotConfig          `yaml:"snapshot"`
}

Config represents the full CATMonitor configuration.

func Default

func Default() *Config

Default returns the default configuration.

func Load

func Load(path string) (*Config, error)

Load reads configuration from a YAML file. If the file does not exist, default configuration is returned.

type FaultSubConfig

type FaultSubConfig struct {
	Enabled        bool             `yaml:"enabled"`         // opt-in switch
	RestAddr       string           `yaml:"rest_addr"`       // subscription REST API listen address
	WebhookTimeout time.Duration    `yaml:"webhook_timeout"` // per-request webhook timeout
	WebhookRetry   int              `yaml:"webhook_retry"`   // failed-webhook retry count
	EventBuffer    int              `yaml:"event_buffer"`    // recent-event ring buffer size
	Defaults       FaultSubDefaults `yaml:"defaults"`
	Rules          map[string]bool  `yaml:"rules"`
}

FaultSubConfig controls the fault subscription & push mechanism (features/faultsub). When Enabled is false (the default) the daemon skips the feature entirely and behaves exactly as before.

type FaultSubDefaults

type FaultSubDefaults struct {
	DebounceMs  int    `yaml:"debounce_ms"`
	MinSeverity string `yaml:"min_severity"`
}

FaultSubDefaults holds subscription defaults applied when a subscriber omits the corresponding field.

type HealthConfig

type HealthConfig struct {
	Enabled      bool   `yaml:"enabled"`
	WeightScheme string `yaml:"weight_scheme"` // auto | cpu_only | accelerated_8card | accelerated_4card
}

HealthConfig holds health evaluation configuration. (Health is evaluated by the snapshot global writer at C_global; no separate health interval — removed dead config.) WeightScheme selects the scoring weights ("auto" detects gpu/npu).

type ServerConfig

type ServerConfig struct {
	Type string `yaml:"type"` // auto | cpu_only | accelerated
}

ServerConfig holds server-level configuration.

type SnapshotConfig

type SnapshotConfig struct {
	Enabled bool   `yaml:"enabled"`
	Dir     string `yaml:"dir"` // directory for snapshot_<comp>.json + snapshot.json
}

SnapshotConfig controls daemon-side snapshot production (per-component files snapshot_<comp>.json + one global snapshot.json), consumed by read-only features (web/dfee). When Enabled is false (the default) the daemon writes no snapshot files and behaves exactly as before — this is the migration switch. When enabled, the daemon is the sole snapshot producer and web must run as a read-only consumer (no self-collection). The per-component history ring depth is fixed at 60 (not configurable).

type StorageConfig

type StorageConfig struct {
	DataDir    string        `yaml:"data_dir"`
	MaxFileAge time.Duration `yaml:"max_file_age"`
	Rotation   string        `yaml:"rotation"`
}

StorageConfig holds data storage configuration.

type StragglerOutputConfig

type StragglerOutputConfig struct {
	Enabled       bool          `yaml:"enabled"`        // opt-in switch
	DataDir       string        `yaml:"data_dir"`       // KPI file directory
	Retention     time.Duration `yaml:"retention"`      // file retention (default 15d)
	FlushInterval time.Duration `yaml:"flush_interval"` // in-memory buffer flush cadence
	Metrics       []string      `yaml:"metrics"`        // which straggler fields to emit (empty=all)
}

StragglerOutputConfig controls the straggler-dedicated KPI file output (features/stragglerout). When Enabled is false (the default) the daemon skips the feature and no KPI file is produced.