Documentation
¶
Index ¶
Constants ¶
const ( ReasonGraceful = "graceful" // shutdown, app stop/disable, drain completed ReasonCrashed = "crashed" // health check failed or backend unreachable ReasonIdleTimeout = "idle_timeout" // autoscaler evicted an idle worker )
Reason labels for blockyard_workers_stopped_total. Keep the set small and bounded — cardinality is #reasons, shared across all apps.
const ( StateBusy = "busy" // has one or more active sessions StateIdle = "idle" // session count is zero, not draining StateDraining = "draining" // marked for drain; no new sessions routed )
State labels for blockyard_workers gauge. Bounded and shared across all apps. States are derived from ActiveWorker fields on each reconciliation tick rather than tracked as explicit transitions, so "starting" and "crashed" (both transient) are not represented — workers only appear in the gauge once WorkersActive sees them, and disappear the moment they're evicted.
const AppUnknown = "unknown"
AppUnknown is the label value used for proxy requests that do not resolve to an app (e.g. 404 on an unknown name). Using a sentinel keeps cardinality bounded against arbitrary URL paths.
Variables ¶
This section is empty.
Functions ¶
func InitTracing ¶
InitTracing sets up the OpenTelemetry trace provider. Returns a shutdown function that flushes pending spans. If endpoint is empty, tracing is not initialized (no-op provider is used).
func ProxyStatusClass ¶ added in v0.0.3
ProxyStatusClass returns the "Nxx" status class for an HTTP status code. Bucketing to classes (rather than raw codes) keeps label cardinality bounded — upstream apps and proxies emit long-tail codes (418, 499, 520, …) that would otherwise create a new series each.
Types ¶
type Metrics ¶ added in v0.0.3
type Metrics struct {
// Gauges — current state
WorkersActive prometheus.Gauge
WorkersByState *prometheus.GaugeVec // labels: state
SessionsActive prometheus.Gauge
// Counters — cumulative totals
WorkersSpawned prometheus.Counter
WorkersStopped *prometheus.CounterVec // labels: reason
BundlesUploaded prometheus.Counter
BundleRestoresSucceeded prometheus.Counter
BundleRestoresFailed prometheus.Counter
ProxyRequests *prometheus.CounterVec // labels: app, status
HealthChecksFailed *prometheus.CounterVec // labels: app
AuditEntriesDropped prometheus.Counter
// Histograms — distributions
ColdStartDuration prometheus.Histogram
ProxyRequestDuration prometheus.Histogram
BuildDuration prometheus.Histogram
}
Metrics holds every Prometheus collector used by the blockyard server. Create one instance per server (via NewMetrics) and thread it through the types that need to record observations. Tests should use a dedicated prometheus.Registry so parallel tests and before/after delta assertions stay isolated from each other.
func NewMetrics ¶ added in v0.0.3
func NewMetrics(reg prometheus.Registerer) *Metrics
NewMetrics constructs a Metrics registered with reg. Pass prometheus.DefaultRegisterer in production so the /metrics HTTP handler (which scrapes the default gatherer) can serve the values. Tests should pass prometheus.NewRegistry for isolation.
Source Files