fix: restore tracing endpoint validation; honor per-side client timeouts; cache eviction target; redis close

- tracing: validateEndpoint used fmt.Sscanf("%s:%d") which treats %s as
  greedy, so every valid host:port endpoint failed and traced==nil disabled
  tracing entirely. Validate with net.SplitHostPort (pure helper, port range
  0-65535); test TestValidateEndpoint.
- proxy: createFasthttpClient used clientTimeout for every timeout slot,
  ignoring the parsed CLIENT_READ_TIMEOUT/CLIENT_WRITE_TIMEOUT (silent
  config loss, documented separately in README). Honour per-side read/write
  timeouts, falling back to client timeout when unset.
- cache/memory: evictToFreeMemory capped candidates at maxCacheSize/5 and
  stopped at the subset end, so a large write could leave the cache over
  its memory limit. Collect all entries and sort.Slice (correct + O(n log n)
  vs O(cap^2)), evict until the target is met or the cache is drained.
- cache/redis: the Redis cache client was never closed at shutdown (unlike
  the metrics aggregator). Add RedisConfig.Close, CacheWrapper.Close, a
  cache.Shutdown and register it with the ShutdownManager.

cache/cache.go

@@ -168,6 +168,16 @@ func EnableCache(cfg *CacheConfig) {
 	config = cfg
 }
 
+// Shutdown closes the underlying cache client if it owns external resources
+// (the Redis connection pool). No-op for in-memory caches.
+func Shutdown() {
+	if config != nil && config.Client != nil {
+		if closer, ok := config.Client.(io.Closer); ok {
+			_ = closer.Close()
+		}
+	}
+}
+
 func CacheLookup(hash string) []byte {
 	if !IsCacheInitialized() {
 		return nil

cache/memory/eviction_test.go

@@ -183,3 +183,32 @@ func TestSetMaxMemorySize(t *testing.T) {
 func generateKey(index int) string {
 	return "test-key-" + fmt.Sprintf("%d", index)
 }
+
+// TestEvictToFreeMemoryWithSmallMaxEntriesReachesTarget is a regression for
+// under-eviction: evictToFreeMemory used to cap its candidate set at
+// maxCacheSize/5 and stop at the end of that subset. With a small
+// maxCacheSize the cap (maxCacheSize/5) could be lower than the number of
+// entries needed to satisfy bytesToFree, so the cache stayed over its
+// memory limit.
+func TestEvictToFreeMemoryWithSmallMaxEntriesReachesTarget(t *testing.T) {
+	// maxCacheSize=10 -> old candidate cap was maxCacheSize/5 = 2 entries.
+	cache := NewWithSize(5*time.Second, DefaultMaxMemorySize, 10)
+	val := make([]byte, 1024) // ~1KB + overhead each
+	const count = 8
+	for i := 0; i < count; i++ {
+		cache.Set(fmt.Sprintf("e%d", i), val, time.Second)
+	}
+	before := cache.GetMemoryUsage()
+	// Request freeing an amount equal to ~half the entries (well beyond the
+	// old 2-entry cap).
+	need := (count / 2) * (int64(len(val)) + approxEntryOverhead)
+	cache.evictToFreeMemory(need)
+	after := cache.GetMemoryUsage()
+	assert.LessOrEqual(t, after, before-need,
+		"evictToFreeMemory must free at least the requested bytes even across more than maxCacheSize/5 entries")
+
+	// Requesting more than the total must drain the cache.
+	cache.evictToFreeMemory(1 << 30)
+	assert.Equal(t, int64(0), cache.GetMemoryUsage(),
+		"an oversized request should evict every entry")
+}

cache/memory/memory.go

@@ -8,6 +8,7 @@ import (
 	"compress/gzip"
 	"context"
 	"io"
+	"sort"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -322,32 +323,29 @@ func (c *Cache) evictToFreeMemory(bytesToFree int64) {
 		memorySize int64
 	}
 
-	// Collect entries to consider for eviction
-	entries := make([]keyMemorySize, 0, int(c.maxCacheSize/5))
+	// Collect ALL entries so eviction can actually reach the target. The
+	// previous code capped the candidate set at maxCacheSize/5 and stopped at
+	// the end of that subset, so a large write could leave the cache over its
+	// memory limit.
+	entries := make([]keyMemorySize, 0, int(c.maxCacheSize))
 	c.entries.Range(func(k, v any) bool {
 		key := k.(string)
 		entry := v.(CacheEntry)
 		entries = append(entries, keyMemorySize{entry.ExpiresAt, key, entry.MemorySize})
-		return len(entries) < cap(entries)
+		return true
 	})
 
-	// Sort entries by expiry time (oldest first)
-	// Simple selection sort since we only need to find the oldest entries
-	var freedBytes int64
-	for i := 0; i < len(entries) && freedBytes < bytesToFree; i++ {
-		oldest := i
-		for j := i + 1; j < len(entries); j++ {
-			if entries[j].expiresAt.Before(entries[oldest].expiresAt) {
-				oldest = j
-			}
-		}
-		// Swap
-		if oldest != i {
-			entries[i], entries[oldest] = entries[oldest], entries[i]
-		}
+	// Sort by expiry time (oldest first)
+	sort.Slice(entries, func(i, j int) bool {
+		return entries[i].expiresAt.Before(entries[j].expiresAt)
+	})
 
-		// Delete this entry
-		if entry, exists := c.entries.LoadAndDelete(entries[i].key); exists {
+	var freedBytes int64
+	for _, e := range entries {
+		if freedBytes >= bytesToFree {
+			break
+		}
+		if entry, exists := c.entries.LoadAndDelete(e.key); exists {
 			cacheEntry := entry.(CacheEntry)
 			atomic.AddInt64(&c.entryCount, -1)
 			atomic.AddInt64(&c.memoryUsage, -cacheEntry.MemorySize)

cache/redis/redis.go

@@ -19,6 +19,16 @@ type RedisConfig struct {
 	prefix      string
 }
 
+// Close releases the underlying Redis connection pool. Called at process
+// shutdown so the pooled connections and go-redis background goroutines are
+// terminated instead of relying on process exit.
+func (c *RedisConfig) Close() error {
+	if c == nil || c.client == nil {
+		return nil
+	}
+	return c.client.Close()
+}
+
 func (c *RedisConfig) prependKeyName(key string) string {
 	builder := c.builderPool.Get().(*strings.Builder)
 	defer c.builderPool.Put(builder)

cache/redis/wrapper.go

@@ -102,3 +102,12 @@ func (w *CacheWrapper) GetMemoryUsage() int64 {
 func (w *CacheWrapper) GetMaxMemorySize() int64 {
 	return 0
 }
+
+// Close releases the underlying Redis client's connection pool at shutdown.
+// Makes CacheWrapper satisfy io.Closer so the cache layer can close it.
+func (w *CacheWrapper) Close() error {
+	if w == nil || w.redis == nil {
+		return nil
+	}
+	return w.redis.Close()
+}