chore(deps): update dependencies

Automated dependency update. All tests passed.

.gitignore

@@ -5,3 +5,6 @@ dist/
 coverage.out
 CLAUDE.md
 graphql-monitoring-proxy
+
+# Lean proofs: local only, never committed
+theorems/

README.md

@@ -821,7 +821,7 @@ Enable rate limiting based on user roles using the `ROLE_RATE_LIMIT` environment
 - `ROLE_RATE_LIMIT`: Enable role-based rate limiting (default: `false`)
 
 **Features:**
-- **Burst control**: Optional burst limits for handling traffic spikes
+- **Burst control**: Optional `burst` caps requests in a short window (one tenth of the interval). `req` counts requests per `interval`. A `burst` at or above `req` has no effect.
 
 `ratelimit.json` is loaded once at startup; changing the file requires a restart.
 

api.go

@@ -143,8 +143,13 @@ func periodicallyReloadBannedUsers(ctx context.Context) {
 	}
 }
 
-func checkIfUserIsBanned(c fiber.Ctx, userID string) bool {
+func isUserBanned(userID string) bool {
 	_, found := bannedUsersIDs.Load(userID)
+	return found
+}
+
+func checkIfUserIsBanned(c fiber.Ctx, userID string) bool {
+	found := isUserBanned(userID)
 
 	cfg.Logger.Debug(&libpack_logger.LogMessage{
 		Message: "Checking if user is banned",

backend_health.go

@@ -118,7 +118,7 @@ func (bhm *BackendHealthManager) WaitForBackendReady(timeout time.Duration) erro
 		}
 
 		// Exponential backoff (1.5x growth, capped at maxDelay), no jitter
-		time.Sleep(currentDelay)
+		time.Sleep(min(currentDelay, time.Until(deadline)))
 		currentDelay = time.Duration(float64(currentDelay) * 1.5)
 		if currentDelay > maxDelay {
 			currentDelay = maxDelay

backend_health_deadline_test.go

@@ -0,0 +1,32 @@
+package main
+
+import (
+	"net"
+	"testing"
+	"time"
+
+	libpack_logger "github.com/lukaszraczylo/graphql-monitoring-proxy/logging"
+	"github.com/valyala/fasthttp"
+)
+
+// Pins theorems/Theorems/ReadyBackoff.lean: the startup wait never sleeps past its timeout.
+func TestWaitForBackendReady_HonoursTimeout(t *testing.T) {
+	ln, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	addr := ln.Addr().String()
+	_ = ln.Close() // closed port: every probe is refused immediately
+
+	bhm := NewBackendHealthManager(&fasthttp.Client{}, "http://"+addr, "", libpack_logger.New())
+	defer bhm.cancel()
+
+	const timeout = 3 * time.Second
+	start := time.Now()
+	if err := bhm.WaitForBackendReady(timeout); err == nil {
+		t.Fatal("want error for unreachable backend")
+	}
+	if elapsed := time.Since(start); elapsed > timeout+500*time.Millisecond {
+		t.Fatalf("waited %v, want <= %v", elapsed, timeout+500*time.Millisecond)
+	}
+}

cache/memory/memory.go

@@ -158,7 +158,8 @@ func (c *Cache) Set(key string, value []byte, ttl time.Duration) {
 		c.evictToFreeMemory(memoryToFree)
 	} else if atomic.LoadInt64(&c.entryCount) >= c.maxCacheSize {
 		// Fall back to count-based eviction for backward compatibility
-		c.evictOldest(int(c.maxCacheSize / 10)) // Evict 10% of entries
+		// 10% of entries, at least one: max/10 is 0 for caps below 10.
+		c.evictOldest(int(max(1, c.maxCacheSize/10)))
 	}
 
 	// A non-positive TTL is clamped to defaultTTL so this backend expires

cache/memory/small_cap_test.go

@@ -0,0 +1,23 @@
+package libpack_cache_memory
+
+import (
+	"fmt"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// Pins theorems/Theorems/MemCacheCap.lean: every cap >= 1 bounds the entry count.
+func TestSet_SmallEntryCapIsEnforced(t *testing.T) {
+	for _, maxEntries := range []int64{1, 2, 5, 9, 10, 25} {
+		t.Run(fmt.Sprintf("cap_%d", maxEntries), func(t *testing.T) {
+			c := NewWithSize(time.Minute, DefaultMaxMemorySize, maxEntries)
+			defer c.Shutdown()
+			for i := 0; i < 100; i++ {
+				c.Set(fmt.Sprintf("k%d", i), []byte("v"), time.Minute)
+				assert.LessOrEqual(t, c.CountQueries(), maxEntries)
+			}
+		})
+	}
+}

cache/redis/redis.go

@@ -83,7 +83,15 @@ type RedisClientConfig struct {
 // entry.
 const defaultTTL = 60 * time.Second
 
+// DefaultKeyPrefix namespaces cache keys when no prefix is configured, so Clear
+// and CountQueries never touch keys of other applications on a shared Redis.
+const DefaultKeyPrefix = "gmp:"
+
 func New(redisClientConfig *RedisClientConfig) (*RedisConfig, error) {
+	prefix := redisClientConfig.Prefix
+	if prefix == "" {
+		prefix = DefaultKeyPrefix
+	}
 	c := &RedisConfig{
 		client: redis.NewClient(&redis.Options{
 			Addr:     redisClientConfig.RedisServer,
@@ -91,7 +99,7 @@ func New(redisClientConfig *RedisClientConfig) (*RedisConfig, error) {
 			DB:       redisClientConfig.RedisDB,
 		}),
 		ctx:    context.Background(),
-		prefix: redisClientConfig.Prefix,
+		prefix: prefix,
 		builderPool: &sync.Pool{
 			New: func() any {
 				return &strings.Builder{}

cache/redis/redis_default_prefix_test.go

@@ -0,0 +1,30 @@
+package libpack_cache_redis
+
+import (
+	"testing"
+
+	"github.com/alicebob/miniredis/v2"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+)
+
+// Pins theorems/Theorems/RedisClear.lean: with no configured prefix, Clear must still spare foreign keys.
+func TestClear_DefaultPrefixSparesForeignKeys(t *testing.T) {
+	t.Parallel()
+	s, err := miniredis.Run()
+	require.NoError(t, err)
+	t.Cleanup(s.Close)
+
+	require.NoError(t, s.Set("other-app:session", "keep"))
+
+	c, err := New(&RedisClientConfig{RedisServer: s.Addr()})
+	require.NoError(t, err)
+	require.NoError(t, c.Set("q1", []byte("v"), 0))
+
+	require.NoError(t, c.Clear())
+
+	assert.True(t, s.Exists("other-app:session"), "foreign key was deleted")
+	_, found, err := c.Get("q1")
+	require.NoError(t, err)
+	assert.False(t, found, "own key must be cleared")
+}

cache/redis/redis_test.go

@@ -131,11 +131,11 @@ func (suite *RedisConfigSuite) TestGetKeys() {
 	err = suite.redisConfig.Set("otherkey", []byte("othervalue"), ttl)
 	assert.NoError(suite.T(), err)
 
-	keys, _ := suite.redisConfig.client.Keys(suite.redisConfig.ctx, "testkey*").Result()
-	expectedKeys := []string{"testkey1", "testkey2"}
+	keys, _ := suite.redisConfig.client.Keys(suite.redisConfig.ctx, DefaultKeyPrefix+"testkey*").Result()
+	expectedKeys := []string{DefaultKeyPrefix + "testkey1", DefaultKeyPrefix + "testkey2"}
 	assert.ElementsMatch(suite.T(), expectedKeys, keys)
 
-	suite.redisConfig.client.Del(suite.redisConfig.ctx, "testkey1", "testkey2", "otherkey")
+	suite.redisConfig.client.Del(suite.redisConfig.ctx, DefaultKeyPrefix+"testkey1", DefaultKeyPrefix+"testkey2", DefaultKeyPrefix+"otherkey")
 }
 
 func (suite *RedisConfigSuite) TestGetKeysCount() {
@@ -154,5 +154,5 @@ func (suite *RedisConfigSuite) TestGetKeysCount() {
 	assert.NoError(suite.T(), err)
 	assert.Equal(suite.T(), int64(3), count3)
 
-	suite.redisConfig.client.Del(suite.redisConfig.ctx, "testkey1", "testkey2", "otherkey")
+	suite.redisConfig.client.Del(suite.redisConfig.ctx, DefaultKeyPrefix+"testkey1", DefaultKeyPrefix+"testkey2", DefaultKeyPrefix+"otherkey")
 }

coalesced_snapshot_test.go

@@ -0,0 +1,27 @@
+package main
+
+import (
+	"testing"
+
+	"github.com/gofiber/fiber/v3"
+	"github.com/stretchr/testify/assert"
+	"github.com/valyala/fasthttp"
+)
+
+// Pins theorems/Theorems/CoalesceAlias.lean: waiters never alias the primary's response buffer.
+func TestSnapshotCoalescedResponse_OwnsBody(t *testing.T) {
+	app := fiber.New()
+	fctx := &fasthttp.RequestCtx{}
+	c := app.AcquireCtx(fctx)
+	defer app.ReleaseCtx(c)
+
+	c.Response().SetStatusCode(201)
+	c.Response().SetBody([]byte(`{"user":"alice"}`))
+
+	snap := snapshotCoalescedResponse(c)
+
+	c.Response().SetBody([]byte(`{"user":"bobby"}`)) // buffer reused by a later request
+
+	assert.Equal(t, 201, snap.StatusCode)
+	assert.Equal(t, `{"user":"alice"}`, string(snap.Body))
+}

events.go

@@ -12,6 +12,9 @@ import (
 const (
 	initialDelay    = 60 * time.Second
 	cleanupInterval = 1 * time.Hour
+
+	// minRetentionDays guards against a cutoff at or after NOW(), which would delete all event history.
+	minRetentionDays = 1
 )
 
 // Use parameterized queries to prevent SQL injection
@@ -46,6 +49,10 @@ func enableHasuraEventCleaner(ctx context.Context) error {
 	logger := cfg.Logger
 	cfgMutex.RUnlock()
 
+	if clearOlderThan < minRetentionDays {
+		return fmt.Errorf("HASURA_EVENT_CLEANER_OLDER_THAN must be at least %d day, got %d: event cleaner not started", minRetentionDays, clearOlderThan)
+	}
+
 	logger.Info(&libpack_logger.LogMessage{
 		Message: "Event cleaner enabled",
 		Pairs:   map[string]any{"interval_in_days": clearOlderThan},

events_retention_test.go

@@ -0,0 +1,26 @@
+package main
+
+import (
+	"context"
+	"testing"
+
+	libpack_logger "github.com/lukaszraczylo/graphql-monitoring-proxy/logging"
+	"github.com/stretchr/testify/assert"
+)
+
+// Pins theorems/Theorems/EventCleaner.lean: a retention below one day would delete all event history.
+func TestEnableHasuraEventCleaner_RejectsNonPositiveRetention(t *testing.T) {
+	for _, days := range []int{0, -1, -30} {
+		cfgMutex.Lock()
+		cfg = &config{Logger: libpack_logger.New()}
+		cfg.HasuraEventCleaner.Enable = true
+		cfg.HasuraEventCleaner.EventMetadataDb = "postgres://fake:[email protected]:1/fake"
+		cfg.HasuraEventCleaner.ClearOlderThan = days
+		cfgMutex.Unlock()
+
+		ctx, cancel := context.WithCancel(context.Background())
+		err := enableHasuraEventCleaner(ctx)
+		cancel()
+		assert.Error(t, err, "days=%d", days)
+	}
+}

go.mod

@@ -5,12 +5,12 @@ go 1.26.0
 toolchain go1.27.3
 
 require (
-	github.com/MicahParks/keyfunc/v3 v3.8.2
+	github.com/MicahParks/keyfunc/v3 v3.9.1
 	github.com/VictoriaMetrics/metrics v1.44.1
 	github.com/alicebob/miniredis/v2 v2.40.0
 	github.com/avast/retry-go/v5 v5.0.0
-	github.com/goccy/go-json v0.11.2
-	github.com/gofiber/contrib/v3/websocket v1.2.9
+	github.com/goccy/go-json v0.12.0
+	github.com/gofiber/contrib/v3/websocket v1.2.10
 	github.com/gofiber/fiber/v3 v3.6.1
 	github.com/gofrs/flock v0.13.1
 	github.com/golang-jwt/jwt/v5 v5.3.1
@@ -18,9 +18,8 @@ require (
 	github.com/gookit/goutil v0.8.6
 	github.com/gorilla/websocket v1.5.3
 	github.com/graphql-go/graphql v0.8.1
-	github.com/jackc/pgx/v5 v5.11.0
+	github.com/jackc/pgx/v5 v5.12.0
 	github.com/lukaszraczylo/ask v0.0.0-20240916204100-6e9ef53a62d9
-	github.com/lukaszraczylo/go-ratecounter v0.1.13
 	github.com/lukaszraczylo/go-simple-graphql v1.2.89
 	github.com/lukaszraczylo/oss-telemetry v0.2.5
 	github.com/redis/go-redis/v9 v9.23.0
@@ -36,14 +35,14 @@ require (
 )
 
 require (
-	github.com/MicahParks/jwkset v0.11.3 // indirect
+	github.com/MicahParks/jwkset v0.11.4 // indirect
 	github.com/avast/retry-go/v4 v4.7.0 // indirect
 	github.com/cenkalti/backoff/v5 v5.0.3 // indirect
 	github.com/cespare/xxhash/v2 v2.3.0 // indirect
 	github.com/fasthttp/websocket v1.5.12 // indirect
 	github.com/go-logr/logr v1.4.4 // indirect
 	github.com/go-logr/stdr v1.2.2 // indirect
-	github.com/gofiber/schema v1.9.1 // indirect
+	github.com/gofiber/schema v1.9.2 // indirect
 	github.com/gofiber/utils/v2 v2.6.3 // indirect
 	github.com/grpc-ecosystem/grpc-gateway/v2 v2.33.0 // indirect
 	github.com/jackc/pgpassfile v1.0.0 // indirect
@@ -66,7 +65,7 @@ require (
 	go.opentelemetry.io/otel/metric v1.47.0 // indirect
 	go.opentelemetry.io/proto/otlp v1.11.1 // indirect
 	go.uber.org/atomic v1.12.0 // indirect
-	go.yaml.in/yaml/v3 v3.0.5 // indirect
+	go.yaml.in/yaml/v3 v3.0.6 // indirect
 	golang.org/x/crypto v0.58.0 // indirect
 	golang.org/x/net v0.62.0 // indirect
 	golang.org/x/sync v0.24.0 // indirect

go.sum

@@ -1,7 +1,7 @@
-github.com/MicahParks/jwkset v0.11.3 h1:Phli4RdTDdIdLXZpuO7abkwZyzIk0RDTUPVVBHPRdkQ=
-github.com/MicahParks/jwkset v0.11.3/go.mod h1:U2oRhRaLgDCLjtpGL2GseNKGmZtLs/3O7p+OZaL5vo0=
-github.com/MicahParks/keyfunc/v3 v3.8.2 h1:eydEwk/pBAVrDIpmFfB/gkCcrp++xQ7YYXirrI2zlWE=
-github.com/MicahParks/keyfunc/v3 v3.8.2/go.mod h1:T4snFPe26GwMg45bBAdM5P6qWQyLxZHLwBhxR/9PnCs=
+github.com/MicahParks/jwkset v0.11.4 h1:sm6pqlEpa1YDu82Hoxho/O+Y291lPcksNImVIsV74gg=
+github.com/MicahParks/jwkset v0.11.4/go.mod h1:U2oRhRaLgDCLjtpGL2GseNKGmZtLs/3O7p+OZaL5vo0=
+github.com/MicahParks/keyfunc/v3 v3.9.1 h1:10dlh2kS3kRtfj5x2bokCxLhTpmIpf3WtsvBdJ2TfOo=
+github.com/MicahParks/keyfunc/v3 v3.9.1/go.mod h1:h8DMhdulFa3yjG1q74FqaehNDekB18HrVCXPl9Raa4I=
 github.com/VictoriaMetrics/metrics v1.44.1 h1:OMyGrMkhI3hnq+IpJ1gZK9VhXcsOjUX1CoOFv//bJ0I=
 github.com/VictoriaMetrics/metrics v1.44.1/go.mod h1:xDM82ULLYCYdFRgQ2JBxi8Uf1+8En1So9YUwlGTOqTc=
 github.com/alicebob/miniredis/v2 v2.40.0 h1:6IgTQbvm2JiMYZ8yur/acNJzVjtSOWCelxXbk/LUBXc=
@@ -28,16 +28,16 @@ github.com/go-logr/logr v1.4.4 h1:tG4xh9yMsRCAiodLVTxyrkzSZ9+o0L1Kg/+cPVcbP/8=
 github.com/go-logr/logr v1.4.4/go.mod h1:9T104GzyrTigFIr8wt5mBrctHMim0Nb2HLGrmQ40KvY=
 github.com/go-logr/stdr v1.2.2 h1:hSWxHoqTgW2S2qGc0LTAI563KZ5YKYRhT3MFKZMbjag=
 github.com/go-logr/stdr v1.2.2/go.mod h1:mMo/vtBO5dYbehREoey6XUKy/eSumjCCveDpRre4VKE=
-github.com/goccy/go-json v0.11.2 h1:jdZv93Tt4ioR8yW1CoNsvSxrcZlCXAUU1aZXN7gpXUA=
-github.com/goccy/go-json v0.11.2/go.mod h1:3NdmfEkZlB7YI5UFw/qdFKq8XN1aiWR0YyRPWZNQltY=
+github.com/goccy/go-json v0.12.0 h1:pD+oWouAiz4yruwXIt/iT+lOMeyhr1aTtxzaMP/RgVI=
+github.com/goccy/go-json v0.12.0/go.mod h1:3NdmfEkZlB7YI5UFw/qdFKq8XN1aiWR0YyRPWZNQltY=
 github.com/goccy/go-reflect v1.2.0 h1:O0T8rZCuNmGXewnATuKYnkL0xm6o8UNOJZd/gOkb9ms=
 github.com/goccy/go-reflect v1.2.0/go.mod h1:n0oYZn8VcV2CkWTxi8B9QjkCoq6GTtCEdfmR66YhFtE=
-github.com/gofiber/contrib/v3/websocket v1.2.9 h1:DgZgxWK5Yua3g5+I4TmoA0RRx6MILIaqooSGKMgjWZE=
-github.com/gofiber/contrib/v3/websocket v1.2.9/go.mod h1:+unMmdAcLfIywqYDaH8eTynn503n7Lf4Cr4kXM5ghO0=
+github.com/gofiber/contrib/v3/websocket v1.2.10 h1:tUyRcHSH9GUsh6WO03Xn+Kjvt3850W6raF6WPngdJLE=
+github.com/gofiber/contrib/v3/websocket v1.2.10/go.mod h1:O6MOrNHCVdCrrstdLrYmRIL0eJy2N87Wa/y2dszuxvQ=
 github.com/gofiber/fiber/v3 v3.6.1 h1:wYjbgqpJm6VErIVLSGzGpX7lLZyTNgElsj8nQ1i+iFo=
 github.com/gofiber/fiber/v3 v3.6.1/go.mod h1:+nxjBal3gw0EPf3gqK5EmtxaxTL01+zbSTagFletlDI=
-github.com/gofiber/schema v1.9.1 h1:QjicFdTDMlkzCOoZAlNWqdkR2YHhgH8/Q/MpcbHurjs=
-github.com/gofiber/schema v1.9.1/go.mod h1:219nYAgzIEBWl/aHCyhQtFoXBvfLjsC6+FI3DUN1+Ck=
+github.com/gofiber/schema v1.9.2 h1:Gzw2PF1K5DpWgG2L51/9t9dkmxw4kqK6auKMvgD9mIU=
+github.com/gofiber/schema v1.9.2/go.mod h1:219nYAgzIEBWl/aHCyhQtFoXBvfLjsC6+FI3DUN1+Ck=
 github.com/gofiber/utils/v2 v2.6.3 h1:VlnOB89rFnonlXoYOkESOkXtIkAw4CBBwpBHu+/a/YQ=
 github.com/gofiber/utils/v2 v2.6.3/go.mod h1:Mq3OXwyBPxORb3jPafy6Swki4/TJJW9TDPHeerXkXpg=
 github.com/gofrs/flock v0.13.1 h1:jjREztyBeSKBZYAC+mgc1laB+xsgy4kYMf3FbKF2UBo=
@@ -62,8 +62,8 @@ github.com/jackc/pgpassfile v1.0.0 h1:/6Hmqy13Ss2zCq62VdNG8tM1wchn8zjSGOBJ6icpsI
 github.com/jackc/pgpassfile v1.0.0/go.mod h1:CEx0iS5ambNFdcRtxPj5JhEz+xB6uRky5eyVu/W2HEg=
 github.com/jackc/pgservicefile v0.0.0-20240606120523-5a60cdf6a761 h1:iCEnooe7UlwOQYpKFhBabPMi4aNAfoODPEFNiAnClxo=
 github.com/jackc/pgservicefile v0.0.0-20240606120523-5a60cdf6a761/go.mod h1:5TJZWKEWniPve33vlWYSoGYefn3gLQRzjfDlhSJ9ZKM=
-github.com/jackc/pgx/v5 v5.11.0 h1:IzBBtyK9AHqf98cctWFifYSci2hgQR/cd56wB4p+ogg=
-github.com/jackc/pgx/v5 v5.11.0/go.mod h1:mal1tBGAFfLHvZzaYh77YS/eC6IX9OWbRV1QIIM0Jn4=
+github.com/jackc/pgx/v5 v5.12.0 h1:hWwax8YiGT0YiN50U1oIDn5kgR7KC2tG3rmavwpPkxY=
+github.com/jackc/pgx/v5 v5.12.0/go.mod h1:u2Fi6no03HCSBYE+ZVdLxbwdu3xyi2vU+K/YNJioSds=
 github.com/jackc/puddle/v2 v2.2.3 h1:f8HoREISb3l5eO9coPgK7U3hKLhWGokOFVcThl5jrao=
 github.com/jackc/puddle/v2 v2.2.3/go.mod h1:vriiEXHvEE654aYKXXjOvZM39qJ0q+azkZFrfEOc3H4=
 github.com/klauspost/compress v1.20.3 h1:zWNoFqq/lSuYQRim3dUXlXeA0IqwuZGicdG5ZTfY5j4=
@@ -72,8 +72,6 @@ github.com/klauspost/cpuid/v2 v2.4.0 h1:S6Hrbc7+ywsr0r+RLapfGBHfyefhCTwEh3A0tV91
 github.com/klauspost/cpuid/v2 v2.4.0/go.mod h1:19jmZ9mjzoF//ddRSUsv0zfBTJWh3QJh9FNxZTMrGxU=
 github.com/lukaszraczylo/ask v0.0.0-20240916204100-6e9ef53a62d9 h1:pL8B9mjv6RPUfKYYGm/uJ7QL6Ndf+z+OEl0qJE6KmEc=
 github.com/lukaszraczylo/ask v0.0.0-20240916204100-6e9ef53a62d9/go.mod h1:M+UVdyqZs++xtEPrascaVmZdOMhCnxjZ2SgH+xHpR0c=
-github.com/lukaszraczylo/go-ratecounter v0.1.13 h1:p1r53oyMb9ZnMbqub6IF4CWhURFiz7k967I0GbUyQMA=
-github.com/lukaszraczylo/go-ratecounter v0.1.13/go.mod h1:TqXEOCtFJStk1i0tkipprv1kiDHGon1MVUisjSTBSKM=
 github.com/lukaszraczylo/go-simple-graphql v1.2.89 h1:Xbu1Ny+a0lT2Sr2SaSC8mcHmGQDwGD4TJKk4DDd+PwA=
 github.com/lukaszraczylo/go-simple-graphql v1.2.89/go.mod h1:PxQYblQDZISmYYj8sNfazAWxAOh1rhAtU208y+uPV8s=
 github.com/lukaszraczylo/oss-telemetry v0.2.5 h1:W6Qgwp7IY4yBRO361MmnEAmM59motdQpbAFAx8v6fJM=
@@ -144,8 +142,8 @@ go.uber.org/automaxprocs v1.6.0 h1:O3y2/QNTOdbF+e/dpXNNW7Rx2hZ4sTIPyybbxyNqTUs=
 go.uber.org/automaxprocs v1.6.0/go.mod h1:ifeIMSnPZuznNm6jmdzmU3/bfk01Fe2fotchwEFJ8r8=
 go.uber.org/goleak v1.3.0 h1:2K3zAYmnTNqV73imy9J1T3WC+gmCePx2hEGkimedGto=
 go.uber.org/goleak v1.3.0/go.mod h1:CoHD4mav9JJNrW/WLlf7HGZPjdw8EucARQHekz1X6bE=
-go.yaml.in/yaml/v3 v3.0.5 h1:N6y/pJk8buWs9NY5ERU2HSMfm+IuD/OtfdAnq6kESPw=
-go.yaml.in/yaml/v3 v3.0.5/go.mod h1:HVTZu1O7/Vkt2N+BFy8Zza+lnLsABggaTM2ZpNIGuKg=
+go.yaml.in/yaml/v3 v3.0.6 h1:zcDfvjXEqTQmpjniUBLc91xYJMB2c4QOFxEF/LCEckM=
+go.yaml.in/yaml/v3 v3.0.6/go.mod h1:HVTZu1O7/Vkt2N+BFy8Zza+lnLsABggaTM2ZpNIGuKg=
 golang.org/x/crypto v0.58.0 h1:COkYLr4k7nDI4r6QtMML2AfzeIswWHAJthLS6K/M/54=
 golang.org/x/crypto v0.58.0/go.mod h1:Xh+kl5A+M0gMiWDytH2Hhr5Z7WbNCpv/eB/EFgfSk+w=
 golang.org/x/net v0.62.0 h1:18ELalfM5+XNRwzM2nYqXsZQ9ByS1uhj2O/U7+5jqRY=

proxy.go

@@ -740,11 +740,7 @@ func performProxyRequest(c fiber.Ctx, proxyURL string) error {
 				}, proxyErr
 			}
 
-			return &CoalescedResponse{
-				Body:       c.Response().Body(),
-				StatusCode: c.Response().StatusCode(),
-				// Headers intentionally left nil; not populated or read anywhere.
-			}, nil
+			return snapshotCoalescedResponse(c), nil
 		})
 
 		// Check for error from rc.Do (though it typically returns nil)
@@ -773,6 +769,17 @@ func performProxyRequest(c fiber.Ctx, proxyURL string) error {
 	return performProxyRequestCore(c, proxyURL, cacheKey)
 }
 
+// snapshotCoalescedResponse captures the primary's response for waiters. It
+// copies the body because fasthttp reuses the buffer once the primary handler
+// returns, while waiters can still be reading.
+func snapshotCoalescedResponse(c fiber.Ctx) *CoalescedResponse {
+	return &CoalescedResponse{
+		Body:       bytes.Clone(c.Response().Body()),
+		StatusCode: c.Response().StatusCode(),
+		// Headers intentionally left nil; not populated or read anywhere.
+	}
+}
+
 // performProxyRequestCore executes the proxy request with retries and circuit breaker
 // This is the core implementation used by both direct calls and coalesced requests
 func performProxyRequestCore(c fiber.Ctx, proxyURL string, cacheKey string) error {

ratelimit.go

@@ -8,17 +8,17 @@ import (
 	"time"
 
 	"github.com/goccy/go-json"
-	goratecounter "github.com/lukaszraczylo/go-ratecounter"
 	libpack_logger "github.com/lukaszraczylo/graphql-monitoring-proxy/logging"
 )
 
 // RateLimitConfig holds the rate limit configuration for a role
 type RateLimitConfig struct {
-	RateCounterTicker *goratecounter.RateCounter
+	// sustained enforces Req over the full Interval.
+	sustained *attemptWindow
 	// burstCounter enforces Burst over a short trailing window, independent
-	// of RateCounterTicker's sustained-rate window. Nil when Burst <= 0
+	// of the sustained window. Nil when Burst <= 0
 	// (burst disabled). See burstWindow and checkRateLimit for semantics.
-	burstCounter *burstWindow
+	burstCounter *attemptWindow
 	Endpoints    []string      `json:"endpoints,omitempty"`
 	Interval     time.Duration `json:"interval"`
 	Req          int           `json:"req"`
@@ -107,51 +107,6 @@ func burstWindowDuration(interval time.Duration) time.Duration {
 	return w
 }
 
-// burstWindow is a minimal, clock-injectable, trailing-window request
-// counter used to enforce RateLimitConfig.Burst independently of the
-// sustained-rate RateCounterTicker.
-//
-// [email protected] cannot provide this second, shorter window
-// cleanly: the window length lives on *RateCounter itself, not per named
-// Counter (helpers.go: GetRate -> c.parent.interval.Seconds()), so two
-// different window lengths need two separate RateCounter instances. Each
-// instance drives its own background goroutine off wall-clock time.Now with
-// no seam to inject a fake clock (helpers.go: start(), cleanUpOldValues(),
-// addValue() all call time.Now() directly). That would force any test of
-// window rollover to sleep on real time. burstWindow is implemented here
-// instead, with a swappable clock, so tests can advance time deterministically.
-type burstWindow struct {
-	mu     sync.Mutex
-	now    func() time.Time
-	window time.Duration
-	hits   []time.Time
-}
-
-// newBurstWindow returns a burstWindow that tracks hits over the trailing
-// duration window, using the real wall clock.
-func newBurstWindow(window time.Duration) *burstWindow {
-	return &burstWindow{now: time.Now, window: window}
-}
-
-// recordAndCount records a hit at the current time, drops hits older than
-// the trailing window, and returns the resulting hit count within the
-// window. Safe for concurrent use.
-func (b *burstWindow) recordAndCount() int {
-	b.mu.Lock()
-	defer b.mu.Unlock()
-	now := b.now()
-	cutoff := now.Add(-b.window)
-	live := b.hits[:0]
-	for _, t := range b.hits {
-		if t.After(cutoff) {
-			live = append(live, t)
-		}
-	}
-	live = append(live, now)
-	b.hits = live
-	return len(b.hits)
-}
-
 // Variable to hold the current load config function - allows for testing
 var loadConfigFunc = loadConfigFromPath
 
@@ -237,14 +192,12 @@ func loadConfigFromPath(path string) error {
 
 	newRateLimits := make(map[string]RateLimitConfig, len(config.RateLimit))
 	for key, value := range config.RateLimit {
-		value.RateCounterTicker = goratecounter.NewRateCounter().WithConfig(goratecounter.RateCounterConfig{
-			Interval: value.Interval,
-		})
+		value.sustained = newAttemptWindow(value.Interval, value.Req)
 
 		// Burst disabled (<=0) leaves burstCounter nil, so checkRateLimit
 		// performs the Req-only check exactly as before burst existed.
 		if value.Burst > 0 {
-			value.burstCounter = newBurstWindow(burstWindowDuration(value.Interval))
+			value.burstCounter = newAttemptWindow(burstWindowDuration(value.Interval), value.Burst)
 		}
 
 		if cfg.LogLevel == "DEBUG" {
@@ -279,7 +232,7 @@ func rateLimitedRequest(userID, userRole string) bool {
 	// Try to get config from atomic value first for better performance
 	if configInterface := rateLimitConfigAtomic.Load(); configInterface != nil {
 		if config, ok := configInterface.(map[string]RateLimitConfig); ok {
-			if roleConfig, exists := config[userRole]; exists && roleConfig.RateCounterTicker != nil {
+			if roleConfig, exists := config[userRole]; exists && roleConfig.sustained != nil {
 				return checkRateLimit(userID, userRole, roleConfig, "")
 			}
 		}
@@ -290,7 +243,7 @@ func rateLimitedRequest(userID, userRole string) bool {
 	roleConfig, ok := rateLimits[userRole]
 	rateLimitMu.RUnlock()
 
-	if !ok || roleConfig.RateCounterTicker == nil {
+	if !ok || roleConfig.sustained == nil {
 		cfg.Logger.Warning(&libpack_logger.LogMessage{
 			Message: "Rate limit role not found or ticker not initialized - defaulting to deny",
 			Pairs:   map[string]any{"user_role": userRole},
@@ -304,72 +257,48 @@ func rateLimitedRequest(userID, userRole string) bool {
 
 // checkRateLimit performs the actual rate limit check.
 //
-// Two independent gates apply, both must pass:
+// Two gates apply, both must pass:
 //
-//  1. Sustained: tickerRate (requests measured over the full Interval) must
-//     stay at or below Req. This is the only gate when Burst <= 0, and its
-//     comparison is unchanged from before the burst fix.
-//  2. Burst (only when roleConfig.burstCounter != nil, i.e. Burst > 0): the
-//     raw request count within a short trailing window (see
-//     burstWindowDuration), much shorter than Interval, must stay at or
-//     below Burst.
+//  1. Sustained: the request count over the full Interval must stay at or
+//     below Req (Req is per Interval, not per second).
+//  2. Burst (only when Burst > 0): the request count within a short trailing
+//     window (see burstWindowDuration) must stay at or below Burst.
 //
-// Because the burst window is short relative to Interval, a client can
-// legally land up to Burst requests in a tight cluster - more than a
-// steady Req-paced stream would deliver in that same short span - as long
-// as the sustained gate still holds over the full Interval. That is what
-// makes burst > req a real spike allowance instead of a no-op. Burst is a
-// short-window ceiling layered ON TOP of the sustained Req rate, not a
-// replacement for it. burst < req is an unusual config: the burst gate
-// becomes the tighter, binding constraint whenever requests cluster in
-// time, but the sustained cap for evenly-spread traffic is still Req
-// alone. The pre-fix code compared the SAME sustained-window tickerRate
-// against both Burst and Req, so when burst < req its Burst check tripped
-// first and Burst was effectively the sustained cap too. This gate's
-// Req-only sustained check can therefore allow MORE through than the
-// pre-fix code did for evenly-spread traffic under a burst < req config.
-// In both cases the burst gate never allows more than Burst requests
-// within its own window (fail-safe).
+// The burst window lies inside the Interval, so its count never exceeds the
+// sustained count. Burst >= Req therefore never rejects anything the sustained
+// gate admits; Burst only takes effect when it is below Req.
 func checkRateLimit(userID, userRole string, roleConfig RateLimitConfig, endpoint string) bool {
-	roleConfig.RateCounterTicker.Incr(1)
-	tickerRate := roleConfig.RateCounterTicker.GetRate()
+	debug := cfg.Logger.IsLevelEnabled(libpack_logger.LEVEL_DEBUG)
 
-	logDetails := map[string]any{
-		"user_role":   userRole,
-		"user_id":     userID,
-		"rate":        tickerRate,
-		"config_rate": roleConfig.Req,
-		"interval":    roleConfig.Interval,
-		"endpoint":    endpoint,
-	}
-
-	cfg.Logger.Debug(&libpack_logger.LogMessage{
-		Message: "Rate limit ticker",
-		Pairs:   map[string]any{"log_details": logDetails},
-	})
-
-	if tickerRate > float64(roleConfig.Req) {
-		cfg.Logger.Debug(&libpack_logger.LogMessage{
-			Message: "Rate limit exceeded",
-			Pairs:   map[string]any{"log_details": logDetails},
-		})
+	if roleConfig.sustained.recordAndExceeds() {
+		if debug {
+			cfg.Logger.Debug(&libpack_logger.LogMessage{
+				Message: "Rate limit exceeded",
+				Pairs: map[string]any{
+					"user_role":   userRole,
+					"user_id":     userID,
+					"config_rate": roleConfig.Req,
+					"interval":    roleConfig.Interval,
+					"endpoint":    endpoint,
+				},
+			})
+		}
 		return false
 	}
 
 	if roleConfig.burstCounter != nil {
-		burstCount := roleConfig.burstCounter.recordAndCount()
-		burstLogDetails := map[string]any{
-			"user_role":    userRole,
-			"user_id":      userID,
-			"burst_count":  burstCount,
-			"config_burst": roleConfig.Burst,
-			"endpoint":     endpoint,
-		}
-		if burstCount > roleConfig.Burst {
-			cfg.Logger.Debug(&libpack_logger.LogMessage{
-				Message: "Burst limit exceeded",
-				Pairs:   map[string]any{"log_details": burstLogDetails},
-			})
+		if roleConfig.burstCounter.recordAndExceeds() {
+			if debug {
+				cfg.Logger.Debug(&libpack_logger.LogMessage{
+					Message: "Burst limit exceeded",
+					Pairs: map[string]any{
+						"user_role":    userRole,
+						"user_id":      userID,
+						"config_burst": roleConfig.Burst,
+						"endpoint":     endpoint,
+					},
+				})
+			}
 			return false
 		}
 	}

ratelimit_test.go

@@ -10,7 +10,6 @@ import (
 	"time"
 
 	"github.com/goccy/go-json"
-	goratecounter "github.com/lukaszraczylo/go-ratecounter"
 	libpack_logger "github.com/lukaszraczylo/graphql-monitoring-proxy/logging"
 )
 
@@ -65,8 +64,8 @@ func (suite *Tests) Test_loadRatelimitConfig() {
 		suite.Contains(rateLimits, "user")
 		suite.Equal(100, rateLimits["admin"].Req)
 		suite.Equal(10, rateLimits["user"].Req)
-		suite.NotNil(rateLimits["admin"].RateCounterTicker)
-		suite.NotNil(rateLimits["user"].RateCounterTicker)
+		suite.NotNil(rateLimits["admin"].sustained)
+		suite.NotNil(rateLimits["user"].sustained)
 	})
 
 	// Test loading config from non-existent path
@@ -154,23 +153,17 @@ func (suite *Tests) Test_rateLimitedRequest() {
 	rateLimits = make(map[string]RateLimitConfig)
 
 	// Admin role with high limit
-	adminCounter := goratecounter.NewRateCounter().WithConfig(goratecounter.RateCounterConfig{
-		Interval: 1 * time.Second,
-	})
 	rateLimits["admin"] = RateLimitConfig{
-		RateCounterTicker: adminCounter,
-		Interval:          1 * time.Second,
-		Req:               100,
+		sustained: newAttemptWindow(1*time.Second, 100),
+		Interval:  1 * time.Second,
+		Req:       100,
 	}
 
 	// User role with low limit
-	userCounter := goratecounter.NewRateCounter().WithConfig(goratecounter.RateCounterConfig{
-		Interval: 1 * time.Second,
-	})
 	rateLimits["user"] = RateLimitConfig{
-		RateCounterTicker: userCounter,
-		Interval:          1 * time.Second,
-		Req:               2, // Set very low for testing
+		sustained: newAttemptWindow(1*time.Second, 2),
+		Interval:  1 * time.Second,
+		Req:       2, // Set very low for testing
 	}
 	rateLimitMu.Unlock()
 
@@ -327,19 +320,17 @@ func (f *fakeClock) Advance(d time.Duration) {
 }
 
 // newBurstTestRoleConfig mirrors the RateLimitConfig construction performed
-// by loadConfigFromPath: a real RateCounterTicker for the sustained check,
+// by loadConfigFromPath: a real sustained window for the Req check,
 // and (only when burst > 0) a burstCounter sized by burstWindowDuration.
 func newBurstTestRoleConfig(interval time.Duration, req, burst int) RateLimitConfig {
 	rc := RateLimitConfig{
-		RateCounterTicker: goratecounter.NewRateCounter().WithConfig(goratecounter.RateCounterConfig{
-			Interval: interval,
-		}),
-		Interval: interval,
-		Req:      req,
-		Burst:    burst,
+		sustained: newAttemptWindow(interval, req),
+		Interval:  interval,
+		Req:       req,
+		Burst:     burst,
 	}
 	if burst > 0 {
-		rc.burstCounter = newBurstWindow(burstWindowDuration(interval))
+		rc.burstCounter = newAttemptWindow(burstWindowDuration(interval), burst)
 	}
 	return rc
 }
@@ -372,59 +363,26 @@ func Test_checkRateLimit_BurstSemantics(t *testing.T) {
 		}
 	})
 
-	t.Run("b_burst_above_req_allows_short_spike_sustained_still_capped", func(t *testing.T) {
+	t.Run("b_burst_above_req_is_noop_sustained_caps_at_req", func(t *testing.T) {
 		const (
 			interval = 10 * time.Second
 			req      = 3
-			burst    = 6 // burst > req
+			burst    = 6 // burst >= req: dominated by the sustained gate
 		)
 		rc := newBurstTestRoleConfig(interval, req, burst)
 		fc := newFakeClock(time.Now())
 		rc.burstCounter.now = fc.Now
 
-		// A rapid cluster of `burst` requests must all be allowed, even
-		// though burst (6) > req (3): the sustained gate only bounds the
-		// total over the full 10s Interval (tickerRate <= 3 needs count <=
-		// 30), so a small cluster like this never trips it.
-		for i := 1; i <= burst; i++ {
+		for i := 1; i <= req; i++ {
 			if !checkRateLimit("u", "role-b", rc, "") {
-				t.Fatalf("request %d in first burst: got denied, want allowed (<= Burst=%d)", i, burst)
+				t.Fatalf("request %d: got denied, want allowed (<= Req=%d per Interval)", i, req)
 			}
 		}
 
-		// The (burst+1)th request in the SAME short window must be denied
-		// by the burst gate specifically - not by the sustained gate,
-		// which alone would still allow it (tickerRate = 7/10 = 0.7 <= 3).
-		// This is the proof burst > req is not a no-op: it is the gate
-		// that actually fires here.
+		// Clearing the burst window must not readmit: Req is per Interval.
+		fc.Advance(2 * burstWindowDuration(interval))
 		if checkRateLimit("u", "role-b", rc, "") {
-			t.Fatalf("request %d: got allowed, want denied by burst gate (> Burst=%d)", burst+1, burst)
-		}
-
-		// Roll the burst window forward (fake clock only - the real
-		// sustained ticker has not aged, since no real time passed) and
-		// repeat clusters of `burst` requests. Each cluster alone passes
-		// burst, but the cumulative sustained count keeps climbing until
-		// it exceeds req*interval.Seconds() = 30, at which point the
-		// sustained gate must deny regardless of how the burst window
-		// resets. This proves "sustained still capped at Req".
-		sustainedTripped := false
-		for batch := 0; batch < 10 && !sustainedTripped; batch++ {
-			fc.Advance(2 * burstWindowDuration(interval)) // clears the burst window
-			for i := 1; i <= burst; i++ {
-				allowed := checkRateLimit("u", "role-b", rc, "")
-				rate := rc.RateCounterTicker.GetRate()
-				if rate > float64(req) {
-					if allowed {
-						t.Fatalf("batch %d req %d: allowed with sustained rate %.2f > Req=%d", batch, i, rate, req)
-					}
-					sustainedTripped = true
-					break
-				}
-			}
-		}
-		if !sustainedTripped {
-			t.Fatalf("sustained gate never tripped after repeated bursts; expected it to cap total requests at Req over the Interval")
+			t.Fatalf("request %d: got allowed, want denied by the sustained gate (Req=%d per Interval)", req+1, req)
 		}
 	})
 

ratelimit_units_test.go

@@ -0,0 +1,44 @@
+package main
+
+import (
+	"testing"
+	"time"
+
+	libpack_logger "github.com/lukaszraczylo/graphql-monitoring-proxy/logging"
+)
+
+func setupRateLimitCfg() {
+	cfg = &config{}
+	parseConfig()
+	cfg.Logger = libpack_logger.New()
+}
+
+// Pins theorems/Theorems/RateLimitUnits.lean: Req counts requests per Interval, not per second.
+func TestCheckRateLimit_ReqIsPerInterval(t *testing.T) {
+	const req = 3
+	setupRateLimitCfg()
+	rc := newBurstTestRoleConfig(time.Minute, req, 0)
+	for i := 1; i <= req; i++ {
+		if !checkRateLimit("u", "r", rc, "") {
+			t.Fatalf("request %d denied, want allowed (Req=%d per minute)", i, req)
+		}
+	}
+	if checkRateLimit("u", "r", rc, "") {
+		t.Fatalf("request %d allowed, want denied: Req=%d per minute", req+1, req)
+	}
+}
+
+// With Burst >= Req the burst gate never rejects what the sustained gate admits.
+func TestCheckRateLimit_BurstAboveReqIsNoOp(t *testing.T) {
+	const req, burst = 3, 6
+	setupRateLimitCfg()
+	rc := newBurstTestRoleConfig(time.Minute, req, burst)
+	for i := 1; i <= req; i++ {
+		if !checkRateLimit("u", "r", rc, "") {
+			t.Fatalf("request %d denied", i)
+		}
+	}
+	if checkRateLimit("u", "r", rc, "") {
+		t.Fatalf("request %d allowed, want denied by the sustained gate", req+1)
+	}
+}

ratelimit_window.go

@@ -0,0 +1,49 @@
+package main
+
+import (
+	"sync"
+	"time"
+)
+
+// attemptWindow decides "more than limit attempts in the trailing window" while
+// storing only the newest limit+1 attempt times. The attempt number limit+1
+// counted back from now lies inside the window exactly when the full count
+// exceeds limit (see theorems/Theorems/SlidingWindow.lean), so memory and time
+// per attempt stay constant however hard a client floods.
+type attemptWindow struct {
+	now    func() time.Time
+	base   time.Time // monotonic origin for stored offsets
+	ring   []int64   // offsets from base, grows to capacity then wraps
+	mu     sync.Mutex
+	window time.Duration
+	head   int // index of the oldest stored attempt once the ring is full
+	size   int // limit+1: attempts kept; the ring grows lazily up to this
+}
+
+func newAttemptWindow(window time.Duration, limit int) *attemptWindow {
+	return &attemptWindow{
+		now:    time.Now,
+		base:   time.Now(),
+		window: window,
+		size:   max(limit, 0) + 1,
+	}
+}
+
+// recordAndExceeds records an attempt now and reports whether the attempts in
+// the trailing window, this one included, outnumber the limit.
+func (w *attemptWindow) recordAndExceeds() bool {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+
+	t := int64(w.now().Sub(w.base))
+	if len(w.ring) < w.size {
+		w.ring = append(w.ring, t)
+	} else {
+		w.ring[w.head] = t
+		w.head = (w.head + 1) % len(w.ring)
+	}
+	if len(w.ring) < w.size {
+		return false
+	}
+	return w.ring[w.head] > t-int64(w.window)
+}

ratelimit_window_test.go

@@ -0,0 +1,64 @@
+package main
+
+import (
+	"math/rand"
+	"testing"
+	"time"
+)
+
+// naiveExceeds is the exact reference: count every attempt in the trailing window.
+func naiveExceeds(attempts []time.Time, now time.Time, window time.Duration, limit int) bool {
+	cutoff := now.Add(-window)
+	n := 0
+	for _, t := range attempts {
+		if t.After(cutoff) {
+			n++
+		}
+	}
+	return n > limit
+}
+
+// Pins theorems/Theorems/SlidingWindow.lean: storing the newest limit+1 attempts decides like the full count.
+func TestAttemptWindow_MatchesExactCount(t *testing.T) {
+	for _, limit := range []int{0, 1, 3, 10} {
+		rng := rand.New(rand.NewSource(int64(limit) + 1))
+		window := 10 * time.Second
+		fc := newFakeClock(time.Unix(1_000_000, 0))
+		w := newAttemptWindow(window, limit)
+		w.now = fc.Now
+		w.base = fc.Now()
+
+		var attempts []time.Time
+		for i := 0; i < 5000; i++ {
+			fc.Advance(time.Duration(rng.Intn(3000)) * time.Millisecond)
+			attempts = append(attempts, fc.Now())
+			want := naiveExceeds(attempts, fc.Now(), window, limit)
+			if got := w.recordAndExceeds(); got != want {
+				t.Fatalf("limit=%d step=%d: got %v, want %v", limit, i, got, want)
+			}
+		}
+	}
+}
+
+func TestAttemptWindow_StorageIsBounded(t *testing.T) {
+	const limit = 10
+	fc := newFakeClock(time.Unix(1_000_000, 0))
+	w := newAttemptWindow(time.Hour, limit)
+	w.now = fc.Now
+	w.base = fc.Now()
+	for i := 0; i < 100_000; i++ {
+		w.recordAndExceeds()
+		fc.Advance(time.Millisecond)
+	}
+	if got := len(w.ring); got > limit+1 {
+		t.Fatalf("stored %d timestamps, want <= %d", got, limit+1)
+	}
+}
+
+func BenchmarkAttemptWindow(b *testing.B) {
+	w := newAttemptWindow(time.Hour, 1000)
+	b.ReportAllocs()
+	for i := 0; i < b.N; i++ {
+		w.recordAndExceeds()
+	}
+}

retry_budget.go

@@ -148,31 +148,39 @@ func (rb *RetryBudget) refill() {
 	now := time.Now().UnixNano()
 	last := rb.lastRefill.Load()
 
-	// Calculate elapsed time in seconds
-	elapsed := float64(now-last) / float64(time.Second)
+	tokensToAdd, newLast := refillStep(tokensPerSecond, last, now)
+	if tokensToAdd <= 0 {
+		return
+	}
+	if rb.lastRefill.CompareAndSwap(last, newLast) {
+		for {
+			current := rb.currentTokens.Load()
+			newValue := current + tokensToAdd
+			if newValue > maxTokens {
+				newValue = maxTokens
+			}
 
-	// Calculate tokens to add
-	tokensToAdd := int64(elapsed * tokensPerSecond)
-
-	if tokensToAdd > 0 {
-		// Update last refill time
-		if rb.lastRefill.CompareAndSwap(last, now) {
-			// Add tokens, capped at maxTokens
-			for {
-				current := rb.currentTokens.Load()
-				newValue := current + tokensToAdd
-				if newValue > maxTokens {
-					newValue = maxTokens
-				}
-
-				if rb.currentTokens.CompareAndSwap(current, newValue) {
-					break
-				}
+			if rb.currentTokens.CompareAndSwap(current, newValue) {
+				break
 			}
 		}
 	}
 }
 
+// refillStep returns the whole tokens earned between last and now (ns) and the
+// clock to store. The clock advances only by the time those tokens pay for, so
+// the fractional remainder carries into the next tick.
+func refillStep(tokensPerSecond float64, last, now int64) (int64, int64) {
+	if tokensPerSecond <= 0 || now <= last {
+		return 0, last
+	}
+	tokens := int64(float64(now-last) / float64(time.Second) * tokensPerSecond)
+	if tokens <= 0 {
+		return 0, last
+	}
+	return tokens, last + int64(float64(tokens)/tokensPerSecond*float64(time.Second))
+}
+
 // GetStats returns current statistics
 func (rb *RetryBudget) GetStats() map[string]any {
 	totalAttempts := rb.totalAttempts.Load()

retry_budget_refill_test.go

@@ -0,0 +1,47 @@
+package main
+
+import (
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// Pins theorems/Theorems/RetryRefill.lean: total_added(15 tps, 100ms ticks, 10s) = 150.
+func TestRefillStep_KeepsFractionalTokens(t *testing.T) {
+	tests := []struct {
+		name string
+		tps  float64
+		want int64
+	}{
+		{"15 tps", 15, 150},
+		{"25 tps", 25, 250},
+		{"5 tps (sub-token per tick)", 5, 50},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			tick := 100 * time.Millisecond
+			var last, total int64
+			now := int64(tick)
+			for i := 0; i < 100; i++ {
+				added, newLast := refillStep(tt.tps, last, now)
+				total += added
+				last = newLast
+				now += int64(tick)
+			}
+			assert.Equal(t, tt.want, total)
+		})
+	}
+}
+
+func TestRefillStep_NoElapsedNoChange(t *testing.T) {
+	added, last := refillStep(10, 1000, 1000)
+	assert.Zero(t, added)
+	assert.Equal(t, int64(1000), last)
+}
+
+func TestRefillStep_ZeroRateNeverAdds(t *testing.T) {
+	added, last := refillStep(0, 0, int64(time.Hour))
+	assert.Zero(t, added)
+	assert.Zero(t, last)
+}

rps_tracker.go

@@ -57,17 +57,23 @@ func (r *RPSTracker) Shutdown() {
 
 // sample calculates RPS since last sample
 func (r *RPSTracker) sample() {
-	now := time.Now()
-	nowNano := now.UnixNano()
+	r.sampleCount()
+}
+
+// sampleCount performs one sample and returns the requests it consumed.
+func (r *RPSTracker) sampleCount() int64 {
+	nowNano := time.Now().UnixNano()
 
-	currentCount := r.lastCount.Load()
 	lastSampleNano := r.lastSampleTime.Load()
 
 	if lastSampleNano == 0 {
 		r.lastSampleTime.Store(nowNano)
-		return
+		return 0
 	}
 
+	// Swap, not Load+Store(0): requests recorded between the two would be lost.
+	currentCount := r.lastCount.Swap(0)
+
 	elapsed := float64(nowNano-lastSampleNano) / float64(time.Second)
 	if elapsed > 0 {
 		rps := float64(currentCount) / elapsed
@@ -75,9 +81,8 @@ func (r *RPSTracker) sample() {
 		r.currentRPS.Store(uint64(rps * 100))
 	}
 
-	// Reset for next sample
-	r.lastCount.Store(0)
 	r.lastSampleTime.Store(nowNano)
+	return currentCount
 }
 
 // GetCurrentRPS returns the current requests per second

rps_tracker_sample_test.go

@@ -0,0 +1,43 @@
+package main
+
+import (
+	"context"
+	"sync"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+// Pins theorems/Theorems/RpsSample.lean: no recorded request is dropped by sample().
+func TestRPSTracker_SampleKeepsConcurrentRequests(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	tr := NewRPSTracker(ctx)
+	tr.Shutdown() // stop the background sampler; sample() is driven by the test
+
+	const writers, perWriter = 8, 20000
+	var wg sync.WaitGroup
+	for i := 0; i < writers; i++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for j := 0; j < perWriter; j++ {
+				tr.RecordRequest()
+			}
+		}()
+	}
+	var sampled int64
+	done := make(chan struct{})
+	go func() { wg.Wait(); close(done) }()
+loop:
+	for {
+		select {
+		case <-done:
+			break loop
+		default:
+			sampled += tr.sampleCount()
+		}
+	}
+	sampled += tr.sampleCount()
+	assert.Equal(t, int64(writers*perWriter), sampled)
+}

server.go

@@ -442,6 +442,11 @@ func processGraphQLRequest(c fiber.Ctx) error {
 // never an error; the request stays anonymous, exactly like before
 // verification existed.
 func extractUserInfo(c fiber.Ctx) (string, string, error) {
+	return extractUserInfoFromHeaders(func(key string) string { return c.Get(key) })
+}
+
+// extractUserInfoFromHeaders resolves user id and role from request headers via get.
+func extractUserInfoFromHeaders(get func(string) string) (string, string, error) {
 	extractedUserID := defaultValue
 	extractedRoleName := defaultValue
 
@@ -450,7 +455,7 @@ func extractUserInfo(c fiber.Ctx) (string, string, error) {
 	// Authorization header must still be checked even with no claim path
 	// set, otherwise JWT_VERIFY_SIGNATURE=true alone would never call
 	// Verify and the forged-token rejection below would never trigger.
-	if authorization := c.Get("Authorization"); authorization != "" &&
+	if authorization := get("Authorization"); authorization != "" &&
 		(len(cfg.Client.JWTUserClaimPath) > 0 || len(cfg.Client.JWTRoleClaimPath) > 0 || cfg.Client.JWTVerifier != nil) {
 		usr, role, err := extractClaimsFromJWTHeader(authorization)
 		if err != nil {
@@ -461,7 +466,7 @@ func extractUserInfo(c fiber.Ctx) (string, string, error) {
 
 	// Override role from header if configured
 	if cfg.Client.RoleFromHeader != "" {
-		if role := c.Get(cfg.Client.RoleFromHeader); role != "" {
+		if role := get(cfg.Client.RoleFromHeader); role != "" {
 			extractedRoleName = role
 		}
 	}

websocket.go

@@ -2,6 +2,7 @@ package main
 
 import (
 	"context"
+	"errors"
 	"fmt"
 	"net/http"
 	"sync"
@@ -231,6 +232,22 @@ func (wsp *WebSocketProxy) handleConnection(ctx context.Context, clientConn *web
 	// Try to extract headers from connection_init payload (for GraphQL WebSocket protocols)
 	enrichedHeaders := wsp.extractAuthFromPayload(message, headers)
 
+	// The upgrade request was gated on its own headers; the backend acts as the
+	// identity in enrichedHeaders, so gate that one too.
+	if authErr := authorizePayloadIdentity(headers, enrichedHeaders); authErr != nil {
+		wsp.errors.Add(1)
+		if wsp.logger != nil {
+			wsp.logger.Warning(&libpack_logger.LogMessage{
+				Message: "WebSocket connection_init identity rejected",
+				Pairs:   map[string]any{"connection_id": connectionID, "reason": authErr.Error()},
+			})
+		}
+		_ = clientConn.WriteMessage(websocket.CloseMessage,
+			websocket.FormatCloseMessage(websocket.ClosePolicyViolation, authErr.Error()))
+		_ = clientConn.Close() // Best-effort cleanup
+		return
+	}
+
 	// Connect to backend WebSocket with enriched headers
 	backendConn, err := wsp.dialBackend(ctx, enrichedHeaders)
 	if err != nil {
@@ -586,6 +603,36 @@ func (wsp *WebSocketProxy) extractAuthFromPayload(message []byte, originalHeader
 	return enrichedHeaders
 }
 
+var (
+	errWSUnauthorized = errors.New("invalid or expired authorization token")
+	errWSBanned       = errors.New("user is banned")
+	errWSRateLimited  = errors.New("rate limit exceeded")
+)
+
+// authorizePayloadIdentity applies the ban and rate-limit checks of the upgrade
+// path to the identity that connection_init supplied. It does nothing when the
+// payload left Authorization and the role header unchanged, because the upgrade
+// request already passed those checks (and counted against the rate limit).
+func authorizePayloadIdentity(upgrade, enriched http.Header) error {
+	roleHeader := cfg.Client.RoleFromHeader
+	if upgrade.Get("Authorization") == enriched.Get("Authorization") &&
+		(roleHeader == "" || upgrade.Get(roleHeader) == enriched.Get(roleHeader)) {
+		return nil
+	}
+
+	userID, role, err := extractUserInfoFromHeaders(enriched.Get)
+	if err != nil {
+		return errWSUnauthorized
+	}
+	if isUserBanned(userID) {
+		return errWSBanned
+	}
+	if cfg.Client.RoleRateLimit && !rateLimitedRequest(userID, role) {
+		return errWSRateLimited
+	}
+	return nil
+}
+
 // dialBackend establishes a WebSocket connection to the backend
 func (wsp *WebSocketProxy) dialBackend(ctx context.Context, headers http.Header) (*gorillaws.Conn, error) {
 	// Convert http:// to ws:// or https:// to wss://

websocket_identity_test.go

@@ -0,0 +1,43 @@
+package main
+
+import (
+	"encoding/base64"
+	"net/http"
+	"testing"
+
+	libpack_logger "github.com/lukaszraczylo/graphql-monitoring-proxy/logging"
+	"github.com/stretchr/testify/assert"
+)
+
+func unsignedJWT(sub, role string) string {
+	enc := base64.RawURLEncoding.EncodeToString
+	return "Bearer " + enc([]byte(`{"alg":"none"}`)) + "." +
+		enc([]byte(`{"sub":"`+sub+`","role":"`+role+`"}`)) + "."
+}
+
+// Pins theorems/Theorems/WsIdentityGate.lean: the identity the backend acts as is the one that is gated.
+func TestAuthorizePayloadIdentity(t *testing.T) {
+	cfg = &config{Logger: libpack_logger.New()}
+	cfg.Client.JWTUserClaimPath = "sub"
+	cfg.Client.JWTRoleClaimPath = "role"
+	bannedUsersIDs.Store("banned-user", "test")
+	defer bannedUsersIDs.Delete("banned-user")
+
+	tests := []struct {
+		name     string
+		upgrade  http.Header
+		enriched http.Header
+		wantErr  error
+	}{
+		{"payload adds banned identity", http.Header{}, http.Header{"Authorization": {unsignedJWT("banned-user", "user")}}, errWSBanned},
+		{"payload replaces clean identity with banned", http.Header{"Authorization": {unsignedJWT("ok-user", "user")}}, http.Header{"Authorization": {unsignedJWT("banned-user", "user")}}, errWSBanned},
+		{"payload identity clean", http.Header{}, http.Header{"Authorization": {unsignedJWT("ok-user", "user")}}, nil},
+		{"payload does not change identity", http.Header{"Authorization": {unsignedJWT("ok-user", "user")}}, http.Header{"Authorization": {unsignedJWT("ok-user", "user")}}, nil},
+		{"no authorization anywhere", http.Header{}, http.Header{}, nil},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			assert.ErrorIs(t, authorizePayloadIdentity(tt.upgrade, tt.enriched), tt.wantErr)
+		})
+	}
+}

websocket_test.go

@@ -11,7 +11,6 @@ import (
 
 	fiber "github.com/gofiber/fiber/v3"
 	gorillaws "github.com/gorilla/websocket"
-	goratecounter "github.com/lukaszraczylo/go-ratecounter"
 	libpack_logger "github.com/lukaszraczylo/graphql-monitoring-proxy/logging"
 	libpack_monitoring "github.com/lukaszraczylo/graphql-monitoring-proxy/monitoring"
 	"github.com/stretchr/testify/assert"
@@ -652,11 +651,9 @@ func TestHandleWebSocketOrDefault_RateLimitedUserRejected(t *testing.T) {
 	rateLimitMu.Lock()
 	rateLimits = map[string]RateLimitConfig{
 		defaultValue: {
-			RateCounterTicker: goratecounter.NewRateCounter().WithConfig(goratecounter.RateCounterConfig{
-				Interval: 1 * time.Second,
-			}),
-			Interval: 1 * time.Second,
-			Req:      0,
+			sustained: newAttemptWindow(1*time.Second, 0),
+			Interval:  1 * time.Second,
+			Req:       0,
 		},
 	}
 	rateLimitConfigAtomic.Store(rateLimits)