🔧 chore(config): defense-in-depth for WatchAndApply test race (Q-038) (#50)

Co-authored-by: Gabriel Radureau <arcodange@gmail.com>
Co-committed-by: Gabriel Radureau <arcodange@gmail.com>

pkg/config/config.go

@@ -730,10 +730,15 @@ func (c *Config) WatchAndApply(ctx context.Context) {
 	// Stop the watcher on context cancel — we set a flag that the
 	// OnConfigChange handler checks, avoiding the race with viper's
 	// internal state that would occur if we called OnConfigChange again.
-	// We deliberately do NOT log here: viper's internal watcher goroutine
-	// has no public Stop, so it can outlive ctx, and a zerolog call here
-	// would race with the next test's LoadConfig → SetupLogging →
-	// zerolog.SetGlobalLevel under -race (observed 2026-05-05).
+	//
+	// We deliberately do NOT log inside this goroutine: this goroutine
+	// outlives ctx (parent's defer cancel only fires when the test's
+	// outer scope exits, not when t.Cleanup runs), so a log call here
+	// races with the next test's LoadConfig → SetupLogging →
+	// zerolog.SetGlobalLevel under -race (observed 2026-05-05, Q-038).
+	// The flag-set is the load-bearing operation; the missing log line
+	// is a small ops cost (operators learn the watcher stops on shutdown
+	// via the parent shutdown logs, not a dedicated message).
 	go func() {
 		<-ctx.Done()
 		c.reloadMu.Lock()

pkg/config/main_test.go

@@ -0,0 +1,26 @@
+package config
+
+import (
+	"os"
+	"testing"
+
+	"github.com/rs/zerolog"
+)
+
+// TestMain quiets the global zerolog level for the duration of the test
+// suite. Rationale (Q-038, 2026-05-05): viper's internal watcher goroutine
+// (started by viper.WatchConfig in WatchAndApply) has no public Stop and
+// can outlive a test's context. Any log call from a leaked goroutine
+// races with the next test's LoadConfig → SetupLogging →
+// zerolog.SetGlobalLevel under `go test -race`. Disabling the logger here
+// is the root-cause fix: the racing memory location is zerolog's gLevel
+// global, and if no log call ever evaluates against it we sidestep the
+// race entirely without changing production behavior.
+//
+// In production, log calls happen against an unchanging global level
+// (SetupLogging runs once at startup), so the race condition does not
+// occur there.
+func TestMain(m *testing.M) {
+	zerolog.SetGlobalLevel(zerolog.Disabled)
+	os.Exit(m.Run())
+}