resolver: wave DNS lookup with fallback pool and bounded retries

selective-vpn-api/app/resolver.go

@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"hash/fnv"
 	"net"
 	"net/netip"
 	"os"
@@ -140,6 +141,15 @@ type wildcardMatcher struct {
 	suffix []string
 }
 
+var resolverFallbackDNS = []string{
+	"1.1.1.1#53",
+	"1.0.0.1#53",
+	"9.9.9.9#53",
+	"149.112.112.112#53",
+	"8.8.8.8#53",
+	"8.8.4.4#53",
+}
+
 func normalizeWildcardDomain(raw string) string {
 	d := strings.TrimSpace(strings.SplitN(raw, "#", 2)[0])
 	d = strings.ToLower(d)
@@ -248,6 +258,14 @@ func runResolverJob(opts ResolverOpts, logf func(string, ...any)) (resolverResul
 	if workers > 500 {
 		workers = 500
 	}
+	dnsTimeoutMs := envInt("RESOLVE_DNS_TIMEOUT_MS", 1800)
+	if dnsTimeoutMs < 300 {
+		dnsTimeoutMs = 300
+	}
+	if dnsTimeoutMs > 5000 {
+		dnsTimeoutMs = 5000
+	}
+	dnsTimeout := time.Duration(dnsTimeoutMs) * time.Millisecond
 
 	domainCache := loadDomainCacheState(opts.CachePath, logf)
 	ptrCache := loadJSONMap(opts.PtrCachePath)
@@ -266,7 +284,7 @@ func runResolverJob(opts ResolverOpts, logf func(string, ...any)) (resolverResul
 	}
 
 	if logf != nil {
-		logf("resolver start: domains=%d ttl=%ds workers=%d", len(domains), ttl, workers)
+		logf("resolver start: domains=%d ttl=%ds workers=%d dns_timeout_ms=%d", len(domains), ttl, workers, dnsTimeoutMs)
 	}
 	start := time.Now()
 
@@ -309,7 +327,7 @@ func runResolverJob(opts ResolverOpts, logf func(string, ...any)) (resolverResul
 		for i := 0; i < workers; i++ {
 			go func() {
 				for j := range jobs {
-					ips, stats := resolveHostGo(j.host, cfg, metaSpecial, wildcards, logf)
+					ips, stats := resolveHostGo(j.host, cfg, metaSpecial, wildcards, dnsTimeout, logf)
 					results <- struct {
 						host  string
 						ips   []string
@@ -462,7 +480,7 @@ func runResolverJob(opts ResolverOpts, logf func(string, ...any)) (resolverResul
 // DNS resolve helpers
 // ---------------------------------------------------------------------
 
-func resolveHostGo(host string, cfg dnsConfig, metaSpecial []string, wildcards wildcardMatcher, logf func(string, ...any)) ([]string, dnsMetrics) {
+func resolveHostGo(host string, cfg dnsConfig, metaSpecial []string, wildcards wildcardMatcher, timeout time.Duration, logf func(string, ...any)) ([]string, dnsMetrics) {
 	useMeta := false
 	for _, m := range metaSpecial {
 		if host == m {
@@ -484,7 +502,7 @@ func resolveHostGo(host string, cfg dnsConfig, metaSpecial []string, wildcards w
 			dnsList = []string{cfg.SmartDNS}
 		}
 	}
-	ips, stats := digA(host, dnsList, 3*time.Second, logf)
+	ips, stats := digA(host, dnsList, timeout, logf)
 	out := []string{}
 	seen := map[string]struct{}{}
 	for _, ip := range ips {
@@ -504,9 +522,22 @@ func resolveHostGo(host string, cfg dnsConfig, metaSpecial []string, wildcards w
 // RU: `digA` - содержит основную логику для dig a.
 // ---------------------------------------------------------------------
 func digA(host string, dnsList []string, timeout time.Duration, logf func(string, ...any)) ([]string, dnsMetrics) {
-	var ips []string
 	stats := dnsMetrics{}
-	for _, entry := range dnsList {
+	if len(dnsList) == 0 {
+		return nil, stats
+	}
+
+	tryLimit := envInt("RESOLVE_DNS_TRY_LIMIT", 2)
+	if tryLimit < 1 {
+		tryLimit = 1
+	}
+	if tryLimit > len(dnsList) {
+		tryLimit = len(dnsList)
+	}
+
+	start := pickDNSStartIndex(host, len(dnsList))
+	for attempt := 0; attempt < tryLimit; attempt++ {
+		entry := dnsList[(start+attempt)%len(dnsList)]
 		server, port := splitDNS(entry)
 		if server == "" {
 			continue
@@ -531,17 +562,24 @@ func digA(host string, dnsList []string, timeout time.Duration, logf func(string
 			if logf != nil {
 				logf("dns warn %s via %s: kind=%s err=%v", host, addr, kind, err)
 			}
+			// NXDOMAIN usually means authoritative negative answer.
+			// Do not fan out further retries for this host.
+			if kind == dnsErrorNXDomain {
+				break
+			}
 			continue
 		}
 		stats.addSuccess(addr)
+		var ips []string
 		for _, ip := range records {
 			if isPrivateIPv4(ip) {
 				continue
 			}
 			ips = append(ips, ip)
 		}
+		return uniqueStrings(ips), stats
 	}
-	return uniqueStrings(ips), stats
+	return nil, stats
 }
 
 func classifyDNSError(err error) dnsErrorKind {
@@ -1066,6 +1104,8 @@ func loadDNSConfig(path string, logf func(string, ...any)) dnsConfig {
 	if len(meta) > 0 {
 		cfg.Meta = meta
 	}
+	cfg.Default = mergeDNSUpstreamPools(cfg.Default, resolverFallbackPool())
+	cfg.Meta = mergeDNSUpstreamPools(cfg.Meta, resolverFallbackPool())
 	if logf != nil {
 		logf("dns-config: accept %s: mode=%s smartdns=%s default=%v; meta=%v", path, cfg.Mode, cfg.SmartDNS, cfg.Default, cfg.Meta)
 	}
@@ -1130,6 +1170,65 @@ func uniqueStrings(in []string) []string {
 	return out
 }
 
+func pickDNSStartIndex(host string, size int) int {
+	if size <= 1 {
+		return 0
+	}
+	h := fnv.New32a()
+	_, _ = h.Write([]byte(strings.ToLower(strings.TrimSpace(host))))
+	return int(h.Sum32() % uint32(size))
+}
+
+func resolverFallbackPool() []string {
+	raw := strings.TrimSpace(os.Getenv("RESOLVE_DNS_FALLBACKS"))
+	switch strings.ToLower(raw) {
+	case "off", "none", "0":
+		return nil
+	}
+
+	candidates := resolverFallbackDNS
+	if raw != "" {
+		candidates = nil
+		fields := strings.FieldsFunc(raw, func(r rune) bool {
+			return r == ',' || r == ';' || r == ' ' || r == '\n' || r == '\t'
+		})
+		for _, f := range fields {
+			if n := normalizeDNSUpstream(f, "53"); n != "" {
+				candidates = append(candidates, n)
+			}
+		}
+	}
+	return uniqueStrings(candidates)
+}
+
+func mergeDNSUpstreamPools(primary, fallback []string) []string {
+	maxUpstreams := envInt("RESOLVE_DNS_MAX_UPSTREAMS", 12)
+	if maxUpstreams < 1 {
+		maxUpstreams = 1
+	}
+	out := make([]string, 0, len(primary)+len(fallback))
+	seen := map[string]struct{}{}
+	add := func(items []string) {
+		for _, item := range items {
+			if len(out) >= maxUpstreams {
+				return
+			}
+			n := normalizeDNSUpstream(item, "53")
+			if n == "" {
+				continue
+			}
+			if _, ok := seen[n]; ok {
+				continue
+			}
+			seen[n] = struct{}{}
+			out = append(out, n)
+		}
+	}
+	add(primary)
+	add(fallback)
+	return out
+}
+
 // ---------------------------------------------------------------------
 // text cleanup + IP classifiers
 // ---------------------------------------------------------------------