145 lines
5.0 KiB
Go
145 lines
5.0 KiB
Go
package observability
|
|
|
|
import (
|
|
"sync"
|
|
|
|
"github.com/prometheus/client_golang/prometheus"
|
|
"github.com/prometheus/client_golang/prometheus/collectors"
|
|
)
|
|
|
|
// Metrics holds Relata's application metrics.
|
|
//
|
|
// These describe the health of the *service* — request rates, job outcomes,
|
|
// provider latency. They deliberately contain no stored-knowledge cardinality:
|
|
// no item IDs, no entity names, no query strings ever become label values.
|
|
type Metrics struct {
|
|
Registry *prometheus.Registry
|
|
|
|
HTTPRequests *prometheus.CounterVec
|
|
HTTPDuration *prometheus.HistogramVec
|
|
HTTPInFlight prometheus.Gauge
|
|
|
|
JobsProcessed *prometheus.CounterVec
|
|
JobDuration *prometheus.HistogramVec
|
|
JobsQueued *prometheus.GaugeVec
|
|
|
|
AIRequests *prometheus.CounterVec
|
|
AIDuration *prometheus.HistogramVec
|
|
|
|
StorageOps *prometheus.CounterVec
|
|
StorageDuration *prometheus.HistogramVec
|
|
|
|
IngestItems *prometheus.CounterVec
|
|
ModelInvalid *prometheus.CounterVec
|
|
AuthAttempts *prometheus.CounterVec
|
|
SearchOps *prometheus.CounterVec
|
|
SearchLatency prometheus.Histogram
|
|
}
|
|
|
|
var (
|
|
metricsOnce sync.Once
|
|
metricsInst *Metrics
|
|
)
|
|
|
|
// NewMetrics returns the process-wide metrics registry, creating it once.
|
|
func NewMetrics() *Metrics {
|
|
metricsOnce.Do(func() {
|
|
reg := prometheus.NewRegistry()
|
|
reg.MustRegister(collectors.NewGoCollector())
|
|
reg.MustRegister(collectors.NewProcessCollector(collectors.ProcessCollectorOpts{}))
|
|
|
|
m := &Metrics{Registry: reg}
|
|
|
|
m.HTTPRequests = prometheus.NewCounterVec(prometheus.CounterOpts{
|
|
Name: "relata_http_requests_total",
|
|
Help: "HTTP requests by route pattern, method and status class.",
|
|
}, []string{"route", "method", "status"})
|
|
|
|
m.HTTPDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
|
|
Name: "relata_http_request_duration_seconds",
|
|
Help: "HTTP request duration in seconds.",
|
|
Buckets: []float64{0.005, 0.025, 0.1, 0.25, 1, 2.5, 10, 30},
|
|
}, []string{"route", "method"})
|
|
|
|
m.HTTPInFlight = prometheus.NewGauge(prometheus.GaugeOpts{
|
|
Name: "relata_http_in_flight_requests",
|
|
Help: "HTTP requests currently being served.",
|
|
})
|
|
|
|
m.JobsProcessed = prometheus.NewCounterVec(prometheus.CounterOpts{
|
|
Name: "relata_jobs_processed_total",
|
|
Help: "Background jobs finished, by type and outcome.",
|
|
}, []string{"type", "outcome"})
|
|
|
|
m.JobDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
|
|
Name: "relata_job_duration_seconds",
|
|
Help: "Background job execution duration in seconds.",
|
|
Buckets: []float64{0.05, 0.25, 1, 5, 15, 60, 180, 600},
|
|
}, []string{"type"})
|
|
|
|
m.JobsQueued = prometheus.NewGaugeVec(prometheus.GaugeOpts{
|
|
Name: "relata_jobs_queued",
|
|
Help: "Jobs currently in each status.",
|
|
}, []string{"status"})
|
|
|
|
m.AIRequests = prometheus.NewCounterVec(prometheus.CounterOpts{
|
|
Name: "relata_ai_requests_total",
|
|
Help: "AI provider calls by provider, operation and outcome.",
|
|
}, []string{"provider", "operation", "outcome"})
|
|
|
|
m.AIDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
|
|
Name: "relata_ai_request_duration_seconds",
|
|
Help: "AI provider call duration in seconds.",
|
|
Buckets: []float64{0.1, 0.5, 1, 2.5, 5, 15, 60, 180},
|
|
}, []string{"provider", "operation"})
|
|
|
|
m.StorageOps = prometheus.NewCounterVec(prometheus.CounterOpts{
|
|
Name: "relata_storage_operations_total",
|
|
Help: "Storage operations by store, operation and outcome.",
|
|
}, []string{"store", "operation", "outcome"})
|
|
|
|
m.StorageDuration = prometheus.NewHistogramVec(prometheus.HistogramOpts{
|
|
Name: "relata_storage_operation_duration_seconds",
|
|
Help: "Storage operation duration in seconds.",
|
|
Buckets: []float64{0.001, 0.005, 0.025, 0.1, 0.5, 2, 10},
|
|
}, []string{"store", "operation"})
|
|
|
|
m.IngestItems = prometheus.NewCounterVec(prometheus.CounterOpts{
|
|
Name: "relata_ingest_items_total",
|
|
Help: "Items accepted for ingestion by source type and outcome.",
|
|
}, []string{"source_type", "outcome"})
|
|
|
|
m.ModelInvalid = prometheus.NewCounterVec(prometheus.CounterOpts{
|
|
Name: "relata_model_invalid_output_total",
|
|
Help: "Model responses rejected by schema validation, by task and reason.",
|
|
}, []string{"task", "reason"})
|
|
|
|
m.AuthAttempts = prometheus.NewCounterVec(prometheus.CounterOpts{
|
|
Name: "relata_auth_attempts_total",
|
|
Help: "Authentication attempts by outcome.",
|
|
}, []string{"outcome"})
|
|
|
|
m.SearchOps = prometheus.NewCounterVec(prometheus.CounterOpts{
|
|
Name: "relata_search_operations_total",
|
|
Help: "Search operations by mode and outcome.",
|
|
}, []string{"mode", "outcome"})
|
|
|
|
m.SearchLatency = prometheus.NewHistogram(prometheus.HistogramOpts{
|
|
Name: "relata_search_duration_seconds",
|
|
Help: "End-to-end search duration in seconds.",
|
|
Buckets: []float64{0.01, 0.05, 0.15, 0.5, 1.5, 5, 20},
|
|
})
|
|
|
|
reg.MustRegister(
|
|
m.HTTPRequests, m.HTTPDuration, m.HTTPInFlight,
|
|
m.JobsProcessed, m.JobDuration, m.JobsQueued,
|
|
m.AIRequests, m.AIDuration,
|
|
m.StorageOps, m.StorageDuration,
|
|
m.IngestItems, m.ModelInvalid, m.AuthAttempts,
|
|
m.SearchOps, m.SearchLatency,
|
|
)
|
|
metricsInst = m
|
|
})
|
|
return metricsInst
|
|
}
|