feat(metrics): Prometheus collectors + typed API + HTTP middleware (ADR-030, #15)

New internal/metrics package: AI metrics (summarize/caption/chat latency, LLM
tokens), session metrics (requests+latency by bounded route pattern, logins), and
a /metrics Handler. Adapters call a typed API; never touch prometheus types.
Dep: github.com/prometheus/client_golang (standard Go client; cluster runs
prometheus-operator). TDD: collectors + middleware route-pattern cardinality
covered.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-06-12 08:32:41 +02:00
co-authored by Claude Opus 4.8
parent 64d11af9ef
commit a5a8cf6f6d
4 changed files with 268 additions and 7 deletions
+92
View File
@@ -0,0 +1,92 @@
package metrics
import (
"net/http"
"net/http/httptest"
"testing"
"time"
"github.com/prometheus/client_golang/prometheus"
"github.com/prometheus/client_golang/prometheus/testutil"
dto "github.com/prometheus/client_model/go"
"github.com/stretchr/testify/require"
)
// histCount reads a histogram child's observation count (testutil.ToFloat64 only
// works on counters/gauges; a histogram's WithLabelValues child is an Observer).
func histCount(t *testing.T, o prometheus.Observer) uint64 {
t.Helper()
m, ok := o.(prometheus.Metric)
require.True(t, ok, "histogram child must be a prometheus.Metric")
var d dto.Metric
require.NoError(t, m.Write(&d))
return d.GetHistogram().GetSampleCount()
}
// TestObserveSummarizeRecordsModelOutcomeFallback: a success observation lands on
// the right model/outcome/fallback series.
func TestObserveSummarizeRecordsModelOutcomeFallback(t *testing.T) {
before := histCount(t, summarizeDuration.WithLabelValues("koala/phi4-mini", "success", "false"))
ObserveSummarize("koala/phi4-mini", "success", false, 1200*time.Millisecond)
after := histCount(t, summarizeDuration.WithLabelValues("koala/phi4-mini", "success", "false"))
require.Equal(t, before+1, after, "one success observation recorded for the model")
}
// TestObserveSummarizeRecordsFailureOutcomes: error and parse_error are distinct
// series so a fallback chain's failures are visible.
func TestObserveSummarizeRecordsFailureOutcomes(t *testing.T) {
e0 := histCount(t, summarizeDuration.WithLabelValues("m", "error", "false"))
p0 := histCount(t, summarizeDuration.WithLabelValues("m", "parse_error", "false"))
ObserveSummarize("m", "error", false, time.Second)
ObserveSummarize("m", "parse_error", false, time.Second)
require.Equal(t, e0+1, histCount(t, summarizeDuration.WithLabelValues("m", "error", "false")))
require.Equal(t, p0+1, histCount(t, summarizeDuration.WithLabelValues("m", "parse_error", "false")))
}
func TestObserveCaptionFetchByOutcome(t *testing.T) {
b := histCount(t, captionFetchDuration.WithLabelValues("captions"))
ObserveCaptionFetch("captions", 3*time.Second)
require.Equal(t, b+1, histCount(t, captionFetchDuration.WithLabelValues("captions")))
}
func TestChatAnswerLatencyRecorded(t *testing.T) {
b := histCount(t, chatDuration.WithLabelValues("iguana/gemma4-26b"))
ObserveChat("iguana/gemma4-26b", 2*time.Second)
require.Equal(t, b+1, histCount(t, chatDuration.WithLabelValues("iguana/gemma4-26b")))
}
// TestRecordTokens: prompt + completion land on their kind series; zero is skipped.
func TestRecordTokens(t *testing.T) {
p0 := testutil.ToFloat64(llmTokens.WithLabelValues("m", "prompt"))
c0 := testutil.ToFloat64(llmTokens.WithLabelValues("m", "completion"))
RecordTokens("m", 100, 40)
RecordTokens("m", 0, 0) // skipped, no panic
require.Equal(t, p0+100, testutil.ToFloat64(llmTokens.WithLabelValues("m", "prompt")))
require.Equal(t, c0+40, testutil.ToFloat64(llmTokens.WithLabelValues("m", "completion")))
}
func TestLoginCounted(t *testing.T) {
b := testutil.ToFloat64(logins)
IncLogin()
require.Equal(t, b+1, testutil.ToFloat64(logins))
}
// TestHTTPMiddlewareRecordsByRoutePattern: the request is counted under the bounded
// registered pattern (r.Pattern after routing), not the raw path with its ids.
func TestHTTPMiddlewareRecordsByRoutePattern(t *testing.T) {
mux := http.NewServeMux()
mux.HandleFunc("GET /v/{videoId}", func(w http.ResponseWriter, _ *http.Request) {
w.WriteHeader(http.StatusTeapot)
})
h := HTTPMiddleware(mux)
before := testutil.ToFloat64(httpRequests.WithLabelValues("GET", "GET /v/{videoId}", "418"))
rec := httptest.NewRecorder()
h.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, "/v/abc-123", nil))
require.Equal(t, http.StatusTeapot, rec.Code)
after := testutil.ToFloat64(httpRequests.WithLabelValues("GET", "GET /v/{videoId}", "418"))
require.Equal(t, before+1, after, "counted under the pattern, not /v/abc-123")
require.Equal(t, float64(0), testutil.ToFloat64(httpRequests.WithLabelValues("GET", "/v/abc-123", "418")),
"raw path must never be a label value")
}