Reconnect when RIS Live ends or stalls the stream (closes #53)

Only Stop now ends the reconnect loop. A stream that ends any other way,
with or without an error, marks the feed disconnected and reconnects with
the existing backoff. Before, a response RIS Live ended normally stopped
the loop for good while the health check kept reporting ok.

The streamer also drops the connection when no line arrives for 60
seconds; each line starts that limit over, and it also bounds the wait
for RIS Live to answer. The idle limit and the first backoff delay are
fields so the tests, against a local server, run in well under a second.
The README's Reconnection note describes this behaviour.

Model: opus-5-5

README.md

@@ -141,7 +141,10 @@ routewatch/
 - **Batched Writes**: All database operations are batched for performance
 - **Backpressure**: Probabilistic message dropping when queues exceed 50% capacity
 - **Graceful Shutdown**: 60-second timeout, flushes all pending batches
-- **Reconnection**: Exponential backoff (5s-320s) with reset after 30s of stable connection
+- **Reconnection**: Any end of the stream other than shutdown, including RIS Live
+  ending the response normally, marks the feed disconnected and reconnects with
+  exponential backoff (5s-320s). The backoff resets after a stream that lasted
+  more than 30s. A stream with no line for 60s is dropped and reconnected.
 - **IP Lookup**: the most specific live route is found by looking up the
   address's prefix at each mask length, longest first, in the prefix index
   (at most 33 lookups for IPv4, 129 for IPv6); prefixes are stored in the

internal/streamer/streamer.go

@@ -7,6 +7,7 @@ import (
 	"compress/gzip"
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"io"
 	"math"
@@ -48,6 +49,7 @@ const (
 	metricsUpdateRate     = time.Second
 	minBackoffDelay       = 5 * time.Second
 	maxBackoffDelay       = 320 * time.Second
+	streamIdleTimeout     = 60 * time.Second // Reconnect when no line arrives for this long
 	metricsLogInterval    = 10 * time.Second
 	bytesPerKB            = 1024
 	bytesPerMB            = 1024 * 1024
@@ -58,6 +60,9 @@ const (
 	backpressureSlope     = 2.0 // Slope for linear drop probability increase
 )
 
+// errStreamEnded is returned by stream when RIS Live ends the response.
+var errStreamEnded = errors.New("RIS Live ended the stream")
+
 // MessageHandler defines the interface for processing RIS messages.
 // Implementations must specify which message types they want to receive,
 // how to process messages, and their desired queue capacity.
@@ -107,6 +112,8 @@ type Streamer struct {
 	logger       *logger.Logger
 	client       *http.Client
 	url          string
+	idleTimeout  time.Duration // Drop the connection when no line arrives for this long
+	minBackoff   time.Duration // First wait before reconnecting; doubles up to maxBackoffDelay
 	handlers     []*handlerInfo
 	rawHandler   RawMessageHandler
 	mu           sync.RWMutex
@@ -126,6 +133,8 @@ func New(logger *logger.Logger, metrics *metrics.Tracker) *Streamer {
 	return &Streamer{
 		logger:      logger,
 		url:         risLiveURL,
+		idleTimeout: streamIdleTimeout,
+		minBackoff:  minBackoffDelay,
 		client: &http.Client{
 			Timeout: 0, // No timeout for streaming
 			Transport: &http.Transport{
@@ -405,7 +414,7 @@ func (s *Streamer) updateMetrics(messageBytes int) {
 
 // streamWithReconnect handles streaming with automatic reconnection and exponential backoff
 func (s *Streamer) streamWithReconnect(ctx context.Context) {
-	backoffDelay := minBackoffDelay
+	backoffDelay := s.minBackoff
 	consecutiveFailures := 0
 
 	for {
@@ -422,8 +431,9 @@ func (s *Streamer) streamWithReconnect(ctx context.Context) {
 		err := s.stream(ctx)
 		streamDuration := time.Since(startTime)
 
-		if err == nil {
+		// Only Stop ends the streamer. A stream that ends any other way,
-			// Clean exit (context cancelled)
+		// with or without an error, is a disconnect and is reconnected.
+		if ctx.Err() != nil {
 			return
 		}
 
@@ -434,17 +444,13 @@ func (s *Streamer) streamWithReconnect(ctx context.Context) {
 			"stream_duration", streamDuration)
 		s.metrics.SetConnected(false)
 
-		// Check if context is cancelled
+		// A stream that lasted more than 30 seconds resets the backoff. That
-		if ctx.Err() != nil {
+		// includes a stream dropped after streamIdleTimeout with no line, so
-			return
+		// a silent feed is retried after the first backoff delay every time.
-		}
-
-		// If we streamed for more than 30 seconds, reset the backoff
-		// This indicates we had a successful connection that received data
 		if streamDuration > 30*time.Second {
 			s.logger.Info("Resetting backoff delay due to successful connection",
 				"stream_duration", streamDuration)
-			backoffDelay = minBackoffDelay
+			backoffDelay = s.minBackoff
 			consecutiveFailures = 0
 		} else {
 			// Increment consecutive failures
@@ -477,7 +483,13 @@ func (s *Streamer) stream(ctx context.Context) error {
 	connCtx, connCancel := context.WithCancel(ctx)
 	defer connCancel()
 
-	req, err := http.NewRequestWithContext(ctx, "GET", s.url, nil)
+	// The idle timer drops the connection, by cancelling connCtx, when no
+	// line arrives for s.idleTimeout. Each line read starts it over. It
+	// also covers the wait for the server to answer.
+	idleTimer := time.AfterFunc(s.idleTimeout, connCancel)
+	defer idleTimer.Stop()
+
+	req, err := http.NewRequestWithContext(connCtx, "GET", s.url, nil)
 	if err != nil {
 		return fmt.Errorf("failed to create request: %w", err)
 	}
@@ -559,6 +571,8 @@ func (s *Streamer) stream(ctx context.Context) error {
 	scanner := bufio.NewScanner(reader)
 
 	for scanner.Scan() {
+		idleTimer.Reset(s.idleTimeout)
+
 		select {
 		case <-ctx.Done():
 			s.logger.Info("Stream stopped by context")
@@ -707,11 +721,19 @@ func (s *Streamer) stream(ctx context.Context) error {
 		s.mu.RUnlock()
 	}
 
+	if ctx.Err() != nil {
+		return ctx.Err()
+	}
+
+	if connCtx.Err() != nil {
+		return fmt.Errorf("no line received for %s", s.idleTimeout)
+	}
+
 	if err := scanner.Err(); err != nil {
 		return fmt.Errorf("scanner error: %w", err)
 	}
 
-	return nil
+	return errStreamEnded
 }
 
 // shouldDropForBackpressure determines if a message should be dropped based on queue utilization

internal/streamer/streamer_test.go

@@ -7,6 +7,7 @@ import (
 	"net/http"
 	"net/http/httptest"
 	"runtime"
+	"sync/atomic"
 	"testing"
 	"time"
 
@@ -57,8 +58,8 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 
 	// One warm-up connection so any persistent HTTP transport goroutine exists
 	// before we take the baseline.
-	if err := s.stream(context.Background()); err != nil {
+	if err := s.stream(context.Background()); !errors.Is(err, errStreamEnded) {
-		t.Fatalf("warm-up stream returned error: %v", err)
+		t.Fatalf("warm-up stream returned %v, want %v", err, errStreamEnded)
 	}
 	s.client.CloseIdleConnections()
 
@@ -66,8 +67,8 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 
 	const reconnects = 20
 	for range reconnects {
-		if err := s.stream(context.Background()); err != nil {
+		if err := s.stream(context.Background()); !errors.Is(err, errStreamEnded) {
-			t.Fatalf("stream returned error: %v", err)
+			t.Fatalf("stream returned %v, want %v", err, errStreamEnded)
 		}
 	}
 	s.client.CloseIdleConnections()
@@ -78,23 +79,35 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 	}
 }
 
-// updateHandler wants UPDATE messages and does nothing with them.
+// updateLine is one UPDATE message as RIS Live sends it.
-type updateHandler struct{}
+const updateLine = `{"type":"ris_message","data":{"type":"UPDATE","peer":"192.0.2.1",` +
+	`"peer_asn":"64496","timestamp":1700000000}}` + "\n"
+
+// updateHandler wants UPDATE messages. For each one it handles it sends on
+// handled if there is room, and otherwise does nothing; a nil handled is
+// never sent on.
+type updateHandler struct {
+	handled chan struct{}
+}
 
 func (updateHandler) WantsMessage(messageType string) bool { return messageType == "UPDATE" }
-func (updateHandler) HandleMessage(*ristypes.RISMessage)   {}
 func (updateHandler) QueueCapacity() int                   { return 10 }
 
+func (h updateHandler) HandleMessage(*ristypes.RISMessage) {
+	select {
+	case h.handled <- struct{}{}:
+	default:
+	}
+}
+
 // TestStopBeforeMessageReachesQueues stops the streamer after the read loop
 // has checked for cancellation but before it hands the message to the handler
 // queues. That is the gap a stop from another goroutine can land in, and it
 // used to end in "send on closed channel". The raw handler runs in that gap on
 // the read loop itself, so calling Stop from it hits the gap every time.
 func TestStopBeforeMessageReachesQueues(t *testing.T) {
-	const line = `{"type":"ris_message","data":{"type":"UPDATE","peer":"192.0.2.1",` +
-		`"peer_asn":"64496","timestamp":1700000000}}` + "\n"
 	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
-		_, _ = io.WriteString(w, line)
+		_, _ = io.WriteString(w, updateLine)
 	}))
 	defer srv.Close()
 
@@ -117,6 +130,106 @@ func TestStopBeforeMessageReachesQueues(t *testing.T) {
 	s.Stop()
 }
 
+// TestReconnectWhenStreamEnds has the server end the first response
+// normally, as when RIS Live restarts. The streamer must treat that as a
+// disconnect and connect again.
+func TestReconnectWhenStreamEnds(t *testing.T) {
+	endResponse := func(http.ResponseWriter, *http.Request) {}
+	checkReconnect(t, endResponse, streamIdleTimeout)
+}
+
+// TestReconnectWhenStreamGoesIdle has the server hold the first response
+// open without sending anything. The streamer must drop it once the idle
+// timeout passes and connect again.
+func TestReconnectWhenStreamGoesIdle(t *testing.T) {
+	const idleTimeout = 50 * time.Millisecond
+	holdResponse := func(w http.ResponseWriter, r *http.Request) {
+		w.(http.Flusher).Flush()
+		<-r.Context().Done()
+	}
+	checkReconnect(t, holdResponse, idleTimeout)
+}
+
+// TestStreamNotDroppedWhileLinesArrive has the server send lines well within
+// the idle timeout for several times the timeout, then end the response. Each
+// line must start the idle timeout over, so the stream lasts until the server
+// ends it instead of being dropped as idle.
+func TestStreamNotDroppedWhileLinesArrive(t *testing.T) {
+	const (
+		idleTimeout  = 100 * time.Millisecond
+		lineInterval = idleTimeout / 10
+		sendFor      = 5 * idleTimeout
+	)
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
+		deadline := time.Now().Add(sendFor)
+		for time.Now().Before(deadline) {
+			_, _ = io.WriteString(w, updateLine)
+			w.(http.Flusher).Flush()
+			time.Sleep(lineInterval)
+		}
+	}))
+	defer srv.Close()
+
+	s := New(logger.New(), metrics.New())
+	s.url = srv.URL
+	s.idleTimeout = idleTimeout
+
+	if err := s.stream(context.Background()); !errors.Is(err, errStreamEnded) {
+		t.Fatalf("stream returned %v, want %v", err, errStreamEnded)
+	}
+}
+
+// checkReconnect runs a streamer against a fake RIS Live server. The first
+// response is served by firstResponse; every later one carries one UPDATE
+// line and then stays open. The streamer must come back for a second
+// response, must not be marked connected when it does, and must handle the
+// UPDATE in it.
+func checkReconnect(t *testing.T, firstResponse http.HandlerFunc, idleTimeout time.Duration) {
+	t.Helper()
+
+	const maxWait = 5 * time.Second
+
+	tracker := metrics.New()
+	connectedAtReconnect := make(chan bool, 1)
+	var requests atomic.Int32
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if requests.Add(1) == 1 {
+			firstResponse(w, r)
+
+			return
+		}
+		select {
+		case connectedAtReconnect <- tracker.IsConnected():
+		default:
+		}
+		_, _ = io.WriteString(w, updateLine)
+		w.(http.Flusher).Flush()
+		<-r.Context().Done()
+	}))
+	defer srv.Close()
+
+	s := New(logger.New(), tracker)
+	s.url = srv.URL
+	s.idleTimeout = idleTimeout
+	s.minBackoff = time.Millisecond
+	handler := updateHandler{handled: make(chan struct{}, 1)}
+	s.RegisterHandler(handler)
+	if err := s.Start(); err != nil {
+		t.Fatalf("Start returned error: %v", err)
+	}
+	defer s.Stop()
+
+	select {
+	case <-handler.handled:
+	case <-time.After(maxWait):
+		t.Fatalf("no message from a second response handled within %s", maxWait)
+	}
+
+	if <-connectedAtReconnect {
+		t.Error("streamer was still marked connected when it came back for a second response")
+	}
+}
+
 // settledGoroutineCount lets transient goroutines finish, then reports the
 // current count.
 func settledGoroutineCount() int {