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

@@ -459,9 +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
 }
 
@@ -489,6 +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,
@@ -497,23 +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)
 }
 
@@ -540,8 +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.