feat: jitter dns cache ttl to avoid expiry bursts

client/dns/cache.go

@@ -2,6 +2,7 @@ package dns
 
 import (
 	"fmt"
+	"math/rand/v2"
 	"strings"
 
 	"github.com/coocood/freecache"
@@ -59,6 +60,9 @@ func (c *Cache) Get(name, qtype string, isDirect bool) *dns.Msg {
 
 // Set stores a DNS message in the appropriate cache using DNS TTL.
 // Only A and AAAA records are cached. If isDirect is true, the direct cache is used.
+// The effective cache lifetime is the base TTL plus a random jitter in
+// [0, baseTTL) so that entries with the same base TTL do not expire at the
+// same moment, avoiding bursts of concurrent DNS queries.
 func (c *Cache) Set(msg *dns.Msg, isDirect bool) error {
 	if msg == nil || len(msg.Question) == 0 {
 		return nil
@@ -70,7 +74,7 @@ func (c *Cache) Set(msg *dns.Msg, isDirect bool) error {
 			return err
 		}
 		key := []byte(q.Name + dns.TypeToString[q.Qtype])
-		ttl := dnsCacheTTL(msg, c.serverDomain)
+		ttl := jitterTTL(dnsCacheTTL(msg, c.serverDomain))
 		if isDirect {
 			return c.direct.Set(key, v, ttl)
 		}
@@ -109,6 +113,18 @@ func dnsCacheTTL(msg *dns.Msg, serverDomain string) int {
 	return int(ttl)
 }
 
+// jitterTTL returns the effective cache lifetime in seconds for a base TTL:
+// the base TTL plus a random jitter in [0, ttl). Entries sharing the same
+// base TTL therefore expire at scattered moments instead of all at once,
+// avoiding a burst of concurrent DNS queries. A TTL of 0 (never expire,
+// e.g. the proxy server's own domain) is returned unchanged.
+func jitterTTL(ttl int) int {
+	if ttl <= 0 {
+		return ttl
+	}
+	return ttl + rand.IntN(ttl)
+}
+
 // PrePopulate resolves the domain via the given DNS server and stores the
 // A and AAAA results in both the direct and proxied caches. This is used
 // to pre-seed the cache with the proxy server's IP before TUN routes are

client/dns/cache_test.go

@@ -252,3 +252,31 @@ func TestCache_ServerDomain_NeverExpires(t *testing.T) {
 		t.Fatal("regular domain entry expired before minCacheTTL")
 	}
 }
+
+func TestJitterTTL_NeverExpire(t *testing.T) {
+	// TTL 0（服务器域名永不过期）必须原样返回，不受抖动影响
+	if ttl := jitterTTL(0); ttl != 0 {
+		t.Errorf("jitterTTL(0) expected 0, got %d", ttl)
+	}
+	// 负数属于防御性边界，不应 panic 也不应被改动
+	if ttl := jitterTTL(-1); ttl != -1 {
+		t.Errorf("jitterTTL(-1) expected -1, got %d", ttl)
+	}
+}
+
+func TestJitterTTL_Range(t *testing.T) {
+	const base = 1800
+	seen := make(map[int]bool)
+	for i := 0; i < 1000; i++ {
+		ttl := jitterTTL(base)
+		// 结果必须在 [base, 2*base) 区间内
+		if ttl < base || ttl >= 2*base {
+			t.Fatalf("jitterTTL(%d) = %d, want in [%d, %d)", base, ttl, base, 2*base)
+		}
+		seen[ttl] = true
+	}
+	// 1000 次采样下结果应分散，而非固定同一个值
+	if len(seen) < 2 {
+		t.Errorf("jitterTTL(%d) always returned %d, expected random jitter", base, base)
+	}
+}