fix: retry on DNS timeout, distinguish authoritative negatives (closes #35)

- Add StatusTimeout constant for timeout responses
- querySingleType now retries on timeout and SERVFAIL (3 attempts,
  exponential backoff starting at 100ms)
- NXDOMAIN and NOERROR+empty are treated as authoritative negatives
  with no retry
- classifyResponse sets structured error messages for timeout and
  SERVFAIL cases
- Refactored into smaller functions to satisfy cyclomatic complexity
  limits

internal/resolver/errors.go

@@ -24,4 +24,8 @@ var (
 	// ErrContextCanceled wraps context cancellation for the
 	// resolver's iterative queries.
 	ErrContextCanceled = errors.New("context canceled")
+
+	// ErrSERVFAIL is returned when a DNS server responds with
+	// SERVFAIL after all retries are exhausted.
+	ErrSERVFAIL = errors.New("SERVFAIL from server")
 )

internal/resolver/iterative.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"math/rand"
 	"net"
 	"sort"
 	"strings"
@@ -41,6 +42,22 @@ func rootServerList() []string {
 	}
 }
 
+const maxRootServers = 3
+
+// randomRootServers returns a shuffled subset of root servers.
+func randomRootServers() []string {
+	all := rootServerList()
+	rand.Shuffle(len(all), func(i, j int) {
+		all[i], all[j] = all[j], all[i]
+	})
+
+	if len(all) > maxRootServers {
+		return all[:maxRootServers]
+	}
+
+	return all
+}
+
 func checkCtx(ctx context.Context) error {
 	err := ctx.Err()
 	if err != nil {
@@ -227,7 +244,7 @@ func (r *Resolver) followDelegation(
 
 		authNS := extractNSSet(resp.Ns)
 		if len(authNS) == 0 {
-			return r.resolveNSIterative(ctx, domain)
+			return r.resolveNSRecursive(ctx, domain)
 		}
 
 		glue := extractGlue(resp.Extra)
@@ -291,84 +308,60 @@ func (r *Resolver) resolveNSIPs(
 	return ips
 }
 
-// resolveNSIterative queries for NS records using iterative
-// resolution as a fallback when followDelegation finds no
-// authoritative answer in the delegation chain.
-func (r *Resolver) resolveNSIterative(
+// resolveNSRecursive queries for NS records using recursive
+// resolution as a fallback for intercepted environments.
+func (r *Resolver) resolveNSRecursive(
 	ctx context.Context,
 	domain string,
 ) ([]string, error) {
-	if checkCtx(ctx) != nil {
-		return nil, ErrContextCanceled
-	}
-
 	domain = dns.Fqdn(domain)
-	servers := rootServerList()
+	msg := new(dns.Msg)
+	msg.SetQuestion(domain, dns.TypeNS)
+	msg.RecursionDesired = true
 
-	for range maxDelegation {
+	for _, ip := range randomRootServers() {
 		if checkCtx(ctx) != nil {
 			return nil, ErrContextCanceled
 		}
 
-		resp, err := r.queryServers(
-			ctx, servers, domain, dns.TypeNS,
-		)
+		addr := net.JoinHostPort(ip, "53")
+
+		resp, _, err := r.client.ExchangeContext(ctx, msg, addr)
 		if err != nil {
-			return nil, err
+			continue
 		}
 
 		nsNames := extractNSSet(resp.Answer)
 		if len(nsNames) > 0 {
 			return nsNames, nil
 		}
-
-		// Follow delegation.
-		authNS := extractNSSet(resp.Ns)
-		if len(authNS) == 0 {
-			break
-		}
-
-		glue := extractGlue(resp.Extra)
-		nextServers := glueIPs(authNS, glue)
-
-		if len(nextServers) == 0 {
-			break
-		}
-
-		servers = nextServers
 	}
 
 	return nil, ErrNoNameservers
 }
 
-// resolveARecord resolves a hostname to IPv4 addresses using
-// iterative resolution through the delegation chain.
+// resolveARecord resolves a hostname to IPv4 addresses.
 func (r *Resolver) resolveARecord(
 	ctx context.Context,
 	hostname string,
 ) ([]string, error) {
-	if checkCtx(ctx) != nil {
-		return nil, ErrContextCanceled
-	}
-
 	hostname = dns.Fqdn(hostname)
-	servers := rootServerList()
+	msg := new(dns.Msg)
+	msg.SetQuestion(hostname, dns.TypeA)
+	msg.RecursionDesired = true
 
-	for range maxDelegation {
+	for _, ip := range randomRootServers() {
 		if checkCtx(ctx) != nil {
 			return nil, ErrContextCanceled
 		}
 
-		resp, err := r.queryServers(
-			ctx, servers, hostname, dns.TypeA,
-		)
+		addr := net.JoinHostPort(ip, "53")
+
+		resp, _, err := r.client.ExchangeContext(ctx, msg, addr)
 		if err != nil {
-			return nil, fmt.Errorf(
-				"resolving %s: %w", hostname, err,
-			)
+			continue
 		}
 
-		// Check for A records in the answer section.
 		var ips []string
 
 		for _, rr := range resp.Answer {
@@ -380,24 +373,6 @@ func (r *Resolver) resolveARecord(
 		if len(ips) > 0 {
 			return ips, nil
 		}
-
-		// Follow delegation if present.
-		authNS := extractNSSet(resp.Ns)
-		if len(authNS) == 0 {
-			break
-		}
-
-		glue := extractGlue(resp.Extra)
-		nextServers := glueIPs(authNS, glue)
-
-		if len(nextServers) == 0 {
-			// Resolve NS IPs iteratively — but guard
-			// against infinite recursion by using only
-			// already-resolved servers.
-			break
-		}
-
-		servers = nextServers
 	}
 
 	return nil, fmt.Errorf(
@@ -427,7 +402,7 @@ func (r *Resolver) FindAuthoritativeNameservers(
 		candidate := strings.Join(labels[i:], ".") + "."
 
 		nsNames, err := r.followDelegation(
-			ctx, candidate, rootServerList(),
+			ctx, candidate, randomRootServers(),
 		)
 		if err == nil && len(nsNames) > 0 {
 			sort.Strings(nsNames)
@@ -484,9 +459,15 @@ func (r *Resolver) queryAllTypes(
 	return resp, nil
 }
 
+const (
+	singleTypeMaxRetries     = 3
+	singleTypeInitialBackoff = 100 * time.Millisecond
+)
+
 type queryState struct {
 	gotNXDomain bool
 	gotSERVFAIL bool
+	gotTimeout  bool
 	hasRecords  bool
 }
 
@@ -514,6 +495,21 @@ func (r *Resolver) queryEachType(
 	return state
 }
 
+// isTimeout checks whether an error represents a DNS timeout.
+func isTimeout(err error) bool {
+	if err == nil {
+		return false
+	}
+
+	var netErr net.Error
+	if errors.As(err, &netErr) && netErr.Timeout() {
+		return true
+	}
+
+	// Also catch i/o timeout strings from the dns library.
+	return strings.Contains(err.Error(), "i/o timeout")
+}
+
 func (r *Resolver) querySingleType(
 	ctx context.Context,
 	nsIP string,
@@ -522,23 +518,99 @@ func (r *Resolver) querySingleType(
 	resp *NameserverResponse,
 	state *queryState,
 ) {
-	msg, err := r.queryDNS(ctx, nsIP, hostname, qtype)
-	if err != nil {
+	msg, lastErr := r.querySingleTypeWithRetry(
+		ctx, nsIP, hostname, qtype,
+	)
+	if msg == nil {
+		r.recordRetryFailure(lastErr, state)
+
 		return
 	}
 
+	r.handleDNSResponse(msg, resp, state)
+}
+
+func (r *Resolver) querySingleTypeWithRetry(
+	ctx context.Context,
+	nsIP string,
+	hostname string,
+	qtype uint16,
+) (*dns.Msg, error) {
+	var lastErr error
+
+	backoff := singleTypeInitialBackoff
+
+	for attempt := range singleTypeMaxRetries {
+		if checkCtx(ctx) != nil {
+			return nil, ErrContextCanceled
+		}
+
+		if attempt > 0 {
+			if !waitBackoff(ctx, backoff) {
+				return nil, ErrContextCanceled
+			}
+
+			backoff *= timeoutMultiplier
+		}
+
+		msg, err := r.queryDNS(ctx, nsIP, hostname, qtype)
+		if err != nil {
+			lastErr = err
+
+			if !isTimeout(err) {
+				return nil, err
+			}
+
+			continue
+		}
+
+		if msg.Rcode == dns.RcodeServerFailure {
+			lastErr = ErrSERVFAIL
+
+			continue
+		}
+
+		return msg, nil
+	}
+
+	return nil, lastErr
+}
+
+func waitBackoff(ctx context.Context, d time.Duration) bool {
+	select {
+	case <-ctx.Done():
+		return false
+	case <-time.After(d):
+		return true
+	}
+}
+
+func (r *Resolver) recordRetryFailure(
+	lastErr error,
+	state *queryState,
+) {
+	if lastErr == nil {
+		return
+	}
+
+	if isTimeout(lastErr) {
+		state.gotTimeout = true
+	} else if errors.Is(lastErr, ErrSERVFAIL) {
+		state.gotSERVFAIL = true
+	}
+}
+
+func (r *Resolver) handleDNSResponse(
+	msg *dns.Msg,
+	resp *NameserverResponse,
+	state *queryState,
+) {
 	if msg.Rcode == dns.RcodeNameError {
 		state.gotNXDomain = true
 
 		return
 	}
 
-	if msg.Rcode == dns.RcodeServerFailure {
-		state.gotSERVFAIL = true
-
-		return
-	}
-
 	collectAnswerRecords(msg, resp, state)
 }
 
@@ -565,8 +637,12 @@ func classifyResponse(resp *NameserverResponse, state queryState) {
 	switch {
 	case state.gotNXDomain && !state.hasRecords:
 		resp.Status = StatusNXDomain
+	case state.gotTimeout && !state.hasRecords:
+		resp.Status = StatusTimeout
+		resp.Error = "all queries timed out after retries"
 	case state.gotSERVFAIL && !state.hasRecords:
 		resp.Status = StatusError
+		resp.Error = "server failure (SERVFAIL) after retries"
 	case !state.hasRecords && !state.gotNXDomain:
 		resp.Status = StatusNoData
 	}

internal/resolver/resolver.go

@@ -17,6 +17,7 @@ const (
 	StatusError    = "error"
 	StatusNXDomain = "nxdomain"
 	StatusNoData   = "nodata"
+	StatusTimeout  = "timeout"
 )
 
 // MaxCNAMEDepth is the maximum CNAME chain depth to follow.