From 1eafebedbc3af5b709fdb4371268ae3627932863 Mon Sep 17 00:00:00 2001
From: Xe Iaso <me@xeiaso.net>
Date: Fri, 18 Jul 2025 20:13:03 +0000
Subject: [PATCH] feat(osiris): serve metrics and healthz

Signed-off-by: Xe Iaso <me@xeiaso.net>
---
 cmd/osiris/internal/entrypoint/entrypoint.go | 79 +++++++++++++++++++-
 cmd/osiris/internal/entrypoint/router.go     | 21 +++++-
 2 files changed, 95 insertions(+), 5 deletions(-)

diff --git a/cmd/osiris/internal/entrypoint/entrypoint.go b/cmd/osiris/internal/entrypoint/entrypoint.go
index fdc643e1..611f8097 100644
--- a/cmd/osiris/internal/entrypoint/entrypoint.go
+++ b/cmd/osiris/internal/entrypoint/entrypoint.go
@@ -1,13 +1,20 @@
 package entrypoint
 
 import (
+	"context"
 	"fmt"
 	"log/slog"
+	"net"
 	"net/http"
+	"os"
+	"os/signal"
+	"syscall"
 
 	"github.com/TecharoHQ/anubis/cmd/osiris/internal/config"
 	"github.com/TecharoHQ/anubis/internal"
 	"github.com/hashicorp/hcl/v2/hclsimple"
+	"github.com/prometheus/client_golang/prometheus/promhttp"
+	"golang.org/x/sync/errgroup"
 	healthv1 "google.golang.org/grpc/health/grpc_health_v1"
 )
 
@@ -18,6 +25,9 @@ type Options struct {
 func Main(opts Options) error {
 	internal.SetHealth("osiris", healthv1.HealthCheckResponse_NOT_SERVING)
 
+	ctx, cancel := signal.NotifyContext(context.Background(), os.Interrupt, syscall.SIGTERM)
+	defer cancel()
+
 	var cfg config.Toplevel
 	if err := hclsimple.DecodeFile(opts.ConfigFname, nil, &cfg); err != nil {
 		return fmt.Errorf("can't read configuration file %s:\n\n%w", opts.ConfigFname, err)
@@ -32,6 +42,71 @@ func Main(opts Options) error {
 		return err
 	}
 
-	slog.Info("listening on", "http", cfg.Bind.HTTP)
-	return http.ListenAndServe(cfg.Bind.HTTP, rtr)
+	g, gCtx := errgroup.WithContext(ctx)
+
+	// HTTP
+	g.Go(func() error {
+		ln, err := net.Listen("tcp", cfg.Bind.HTTP)
+		if err != nil {
+			return fmt.Errorf("(HTTP) can't bind to tcp %s: %w", cfg.Bind.HTTP, err)
+		}
+		defer ln.Close()
+
+		go func(ctx context.Context) {
+			<-ctx.Done()
+			ln.Close()
+		}(gCtx)
+
+		slog.Info("listening for HTTP", "bind", cfg.Bind.HTTP)
+
+		srv := http.Server{Handler: rtr, ErrorLog: internal.GetFilteredHTTPLogger()}
+
+		return srv.Serve(ln)
+	})
+
+	// HTTPS
+
+	// Metrics
+	g.Go(func() error {
+		ln, err := net.Listen("tcp", cfg.Bind.Metrics)
+		if err != nil {
+			return fmt.Errorf("(metrics) can't bind to tcp %s: %w", cfg.Bind.Metrics, err)
+		}
+		defer ln.Close()
+
+		go func(ctx context.Context) {
+			<-ctx.Done()
+			ln.Close()
+		}(gCtx)
+
+		mux := http.NewServeMux()
+
+		mux.Handle("/metrics", promhttp.Handler())
+		mux.HandleFunc("/healthz", func(w http.ResponseWriter, r *http.Request) {
+			st, ok := internal.GetHealth("osiris")
+			if !ok {
+				slog.Error("health service osiris does not exist, file a bug")
+			}
+
+			switch st {
+			case healthv1.HealthCheckResponse_NOT_SERVING:
+				http.Error(w, "NOT OK", http.StatusInternalServerError)
+				return
+			case healthv1.HealthCheckResponse_SERVING:
+				fmt.Fprintln(w, "OK")
+				return
+			default:
+				http.Error(w, "UNKNOWN", http.StatusFailedDependency)
+				return
+			}
+		})
+
+		slog.Info("listening for Metrics", "bind", cfg.Bind.Metrics)
+
+		srv := http.Server{Handler: mux, ErrorLog: internal.GetFilteredHTTPLogger()}
+
+		return srv.Serve(ln)
+	})
+
+	return g.Wait()
 }
diff --git a/cmd/osiris/internal/entrypoint/router.go b/cmd/osiris/internal/entrypoint/router.go
index 4f16f34a..d120fbab 100644
--- a/cmd/osiris/internal/entrypoint/router.go
+++ b/cmd/osiris/internal/entrypoint/router.go
@@ -14,11 +14,25 @@ import (
 
 	"github.com/TecharoHQ/anubis/cmd/osiris/internal/config"
 	"github.com/lum8rjack/go-ja4h"
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
 )
 
 var (
 	ErrTargetInvalid = errors.New("[unexpected] target invalid")
 	ErrNoHandler     = errors.New("[unexpected] no handler for domain")
+
+	requestsPerDomain = promauto.NewGaugeVec(prometheus.GaugeOpts{
+		Namespace: "techaro",
+		Subsystem: "osiris",
+		Name:      "request_count",
+	}, []string{"domain"})
+
+	unresolvedRequests = promauto.NewGauge(prometheus.GaugeOpts{
+		Namespace: "techaro",
+		Subsystem: "osiris",
+		Name:      "unresolved_requests",
+	})
 )
 
 type Router struct {
@@ -86,12 +100,12 @@ func NewRouter(c config.Toplevel) (*Router, error) {
 		return nil, err
 	}
 
-	fmt.Printf("%#v\n", result.routes)
-
 	return result, nil
 }
 
 func (rtr *Router) ServeHTTP(w http.ResponseWriter, r *http.Request) {
+	requestsPerDomain.WithLabelValues(r.Host).Inc()
+
 	var h http.Handler
 	var ok bool
 
@@ -104,11 +118,12 @@ func (rtr *Router) ServeHTTP(w http.ResponseWriter, r *http.Request) {
 	rtr.lock.RUnlock()
 
 	if !ok {
+		unresolvedRequests.Inc()
 		http.NotFound(w, r) // TODO(Xe): brand this
 		return
 	}
 
-	r.Header.Set("X-HTTP-JA4H-Fingerprint", ja4hFP)
+	r.Header.Set("X-Http-Ja4h-Fingerprint", ja4hFP)
 
 	h.ServeHTTP(w, r)
 }