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

@@ -141,11 +141,15 @@ 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
-  text form Go's `net/netip` prints
+  (at most 25 lookups for IPv4, 113 for IPv6); prefixes are stored in the
+  text form Go's `net/netip` prints. Routes shorter than /8 for IPv4 or /16
+  for IPv6, such as a default route, are not used.
 
 ### Database Schema
 

internal/database/database.go

@@ -37,6 +37,14 @@ const (
 	ipv6Bits       = 128
 )
 
+// The IP lookup does not use routes shorter than these mask lengths. A default
+// route or a route this short only says where a peer sends traffic it has no
+// better route for; it does not say who holds the address.
+const (
+	minLookupMaskLengthV4 = 8
+	minLookupMaskLengthV6 = 16
+)
+
 // SQLite memory tuning. cache_size and busy_timeout go in the DSN so every
 // pooled connection gets them; the heap limits are process-wide and set once.
 const (
@@ -1876,11 +1884,12 @@ func (d *Database) GetIPInfo(ip string) (*IPInfo, error) {
 // GetIPInfoContext returns comprehensive IP information with context support.
 //
 // It finds the most specific live route for the address with point lookups on
-// the prefix index: for each mask length from the longest (32 or 128) down to 0
-// it looks up the address's network prefix at that length, and the first length
-// with a live route wins. That is at most 33 or 129 indexed lookups, however
-// many routes are live. The prefix is looked up in the text form net/netip
-// prints, which is the form the prefix handler stores.
+// the prefix index: for each mask length from the longest (32 or 128) down to
+// minLookupMaskLengthV4 or minLookupMaskLengthV6 it looks up the address's
+// network prefix at that length, and the first length with a live route wins.
+// That is at most 25 or 113 indexed lookups, however many routes are live. The
+// prefix is looked up in the text form net/netip prints, which is the form the
+// prefix handler stores.
 func (d *Database) GetIPInfoContext(ctx context.Context, ip string) (*IPInfo, error) {
 	addr, err := netip.ParseAddr(ip)
 	if err != nil {
@@ -1891,9 +1900,11 @@ func (d *Database) GetIPInfoContext(ctx context.Context, ip string) (*IPInfo, er
 
 	info := &IPInfo{IP: ip, IPVersion: ipVersionV6}
 	routesTable, prefixesTable := "live_routes_v6", "prefixes_v6"
+	minMaskLength := minLookupMaskLengthV6
 	if addr.Is4() {
 		info.IPVersion = ipVersionV4
 		routesTable, prefixesTable = "live_routes_v4", "prefixes_v4"
+		minMaskLength = minLookupMaskLengthV4
 	}
 
 	// Route info with peer count and prefix first_seen
@@ -1934,7 +1945,7 @@ func (d *Database) GetIPInfoContext(ctx context.Context, ip string) (*IPInfo, er
 	var whoisUpdatedAt sql.NullTime
 
 	found := false
-	for bits := addr.BitLen(); bits >= 0; bits-- {
+	for bits := addr.BitLen(); bits >= minMaskLength; bits-- {
 		prefix := netip.PrefixFrom(addr, bits).Masked().String()
 		err := stmt.QueryRowContext(ctx, prefix).Scan(
 			&info.Netblock,

internal/database/database_test.go

@@ -133,6 +133,78 @@ func TestGetIPInfoFindsMostSpecificLiveRoute(t *testing.T) {
 	lookup("2001:db8:1:2::3", "2001:db8:1::/48")
 }
 
+// TestGetIPInfoIgnoresShortRoutes checks that the IP lookup does not report a
+// default route, or a route shorter than /8 for IPv4 or /16 for IPv6, as the
+// route for an address, and that a route of exactly /8 or /16, or longer, still
+// is. The prefixes are written out so that a wrong limit fails the test.
+func TestGetIPInfoIgnoresShortRoutes(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	announce := func(prefix string, asn int) {
+		t.Helper()
+
+		err := db.UpsertLiveRouteBatch([]*LiveRoute{{
+			ID:          uuid.New(),
+			Prefix:      prefix,
+			MaskLength:  netip.MustParsePrefix(prefix).Bits(),
+			IPVersion:   detectIPVersion(prefix),
+			OriginASN:   asn,
+			PeerIP:      testPeerIP,
+			ASPath:      []int{asn},
+			NextHop:     testPeerIP,
+			LastUpdated: ts,
+		}})
+		if err != nil {
+			t.Fatalf("UpsertLiveRouteBatch(%s): %v", prefix, err)
+		}
+	}
+
+	// lookup checks that ip resolves to the live prefix netblock from asn.
+	lookup := func(ip, netblock string, asn int) {
+		t.Helper()
+
+		info, err := db.GetIPInfo(ip)
+		if err != nil {
+			t.Errorf("GetIPInfo(%s): %v", ip, err)
+
+			return
+		}
+		if info.Netblock != netblock || info.ASN != asn {
+			t.Errorf("GetIPInfo(%s) = %s AS%d, want %s AS%d", ip, info.Netblock, info.ASN, netblock, asn)
+		}
+	}
+
+	// Default and very short routes, and routes one bit shorter than each limit.
+	announce("2000::/3", 64500)
+	announce("0.0.0.0/0", 64500)
+	announce("2000::/15", 64500)
+	announce("192.0.0.0/7", 64500)
+
+	for _, ip := range []string{"2001:db8::1", "192.0.2.1"} {
+		if info, err := db.GetIPInfo(ip); !errors.Is(err, ErrNoRoute) {
+			t.Errorf("GetIPInfo(%s) with only short routes = %+v, %v; want ErrNoRoute", ip, info, err)
+		}
+	}
+
+	// Routes of exactly the minimum length are used.
+	announce("2001::/16", 64501)
+	announce("192.0.0.0/8", 64501)
+
+	lookup("2001:db8::1", "2001::/16", 64501)
+	lookup("192.0.2.1", "192.0.0.0/8", 64501)
+
+	announce("2001:db8::/32", 64502)
+
+	lookup("2001:db8::1", "2001:db8::/32", 64502)
+}
+
 // TestConnectionPoolPragmas holds several pooled connections open at once and
 // checks each one carries the per-connection settings from the DSN, plus the
 // process-wide hard heap limit, and sees the new file with auto_vacuum

internal/metrics/metrics.go

@@ -60,17 +60,22 @@ func New() *Tracker {
 	}
 }
 
-// SetConnected updates the connection status
+// SetConnected updates the connection status. Every connection after the
+// first counts one reconnect.
 func (t *Tracker) SetConnected(connected bool) {
-	wasConnected := t.isConnected.Swap(connected)
-	if connected {
-		t.mu.Lock()
-		t.connectedSince = time.Now()
-		t.mu.Unlock()
-		// Increment reconnect count (but not for the initial connection)
-		if wasConnected || t.reconnectCount.Load() > 0 {
-			t.reconnectCount.Add(1)
-		}
+	t.isConnected.Store(connected)
+	if !connected {
+		return
+	}
+
+	t.mu.Lock()
+	// connectedSince is zero until the first connection.
+	connectedBefore := !t.connectedSince.IsZero()
+	t.connectedSince = time.Now()
+	t.mu.Unlock()
+
+	if connectedBefore {
+		t.reconnectCount.Add(1)
 	}
 }
 

internal/metrics/metrics_test.go

@@ -0,0 +1,21 @@
+package metrics
+
+import "testing"
+
+// TestReconnectCount connects three times with a disconnect between each, the
+// way the streamer does, and expects every connection after the first to count
+// one reconnect.
+func TestReconnectCount(t *testing.T) {
+	tracker := New()
+
+	tracker.SetConnected(true)
+	tracker.SetConnected(false)
+	tracker.SetConnected(true)
+	tracker.SetConnected(false)
+	tracker.SetConnected(true)
+
+	const wantReconnects = 2
+	if got := tracker.GetReconnectCount(); got != wantReconnects {
+		t.Fatalf("reconnect count = %d, want %d", got, wantReconnects)
+	}
+}

internal/routewatch/prefixhandler.go

@@ -105,7 +105,24 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 
-	// Process announcements
+	// Withdrawals are queued before announcements, because the batch keeps
+	// each peer's last update for a prefix: a message that lists a prefix as
+	// both withdrawn and announced keeps the route (RFC 4271 section 4.3).
+	for _, prefix := range msg.Withdrawals {
+		h.batch = append(h.batch, prefixUpdate{
+			prefix:      canonicalPrefix(prefix),
+			originASN:   originASN, // Use the originASN from path if available
+			peer:        msg.Peer,
+			messageType: "withdrawal",
+			timestamp:   timestamp,
+			path:        msg.Path,
+		})
+		// Record withdrawal in metrics
+		if h.metrics != nil {
+			h.metrics.RecordWithdrawal()
+		}
+	}
+
 	for _, announcement := range msg.Announcements {
 		for _, prefix := range announcement.Prefixes {
 			h.batch = append(h.batch, prefixUpdate{
@@ -123,22 +140,6 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
 		}
 	}
 
-	// Process withdrawals
-	for _, prefix := range msg.Withdrawals {
-		h.batch = append(h.batch, prefixUpdate{
-			prefix:      canonicalPrefix(prefix),
-			originASN:   originASN, // Use the originASN from path if available
-			peer:        msg.Peer,
-			messageType: "withdrawal",
-			timestamp:   timestamp,
-			path:        msg.Path,
-		})
-		// Record withdrawal in metrics
-		if h.metrics != nil {
-			h.metrics.RecordWithdrawal()
-		}
-	}
-
 	// Check if we need to flush
 	if len(h.batch) >= prefixBatchSize {
 		h.flushBatchLocked()
@@ -179,14 +180,14 @@ func (h *PrefixHandler) flushBatchLocked() {
 	startTime := time.Now()
 	batchSize := len(h.batch)
 
-	// Group updates by prefix to deduplicate
-	// For each prefix, keep the latest update
-	prefixMap := make(map[string]prefixUpdate)
+	// For each prefix and peer, keep only the last update the feed delivered.
+	// Updates from different peers for the same prefix are all kept. Arrival
+	// order decides, not the timestamp: timestamps are whole seconds, so
+	// updates within one second carry the same one.
+	type prefixAndPeer struct{ prefix, peer string }
+	lastUpdates := make(map[prefixAndPeer]prefixUpdate)
 	for _, update := range h.batch {
-		key := update.prefix
-		if existing, ok := prefixMap[key]; !ok || update.timestamp.After(existing.timestamp) {
-			prefixMap[key] = update
-		}
+		lastUpdates[prefixAndPeer{update.prefix, update.peer}] = update
 	}
 
 	// Collect routes to upsert and delete
@@ -196,7 +197,7 @@ func (h *PrefixHandler) flushBatchLocked() {
 	// Collect unique prefixes to update
 	prefixesToUpdate := make(map[string]time.Time)
 
-	for _, update := range prefixMap {
+	for _, update := range lastUpdates {
 		// Track prefix for both announcements and withdrawals
 		if _, exists := prefixesToUpdate[update.prefix]; !exists || update.timestamp.After(prefixesToUpdate[update.prefix]) {
 			prefixesToUpdate[update.prefix] = update.timestamp
@@ -255,7 +256,7 @@ func (h *PrefixHandler) flushBatchLocked() {
 	elapsed := time.Since(startTime)
 	h.logger.Debug("Flushed prefix batch",
 		"batch_size", batchSize,
-		"unique_prefixes", len(prefixMap),
+		"unique_prefixes", len(prefixesToUpdate),
 		"success", successCount,
 		"duration_ms", elapsed.Milliseconds(),
 	)

internal/routewatch/prefixhandler_test.go

@@ -1,7 +1,9 @@
 package routewatch
 
 import (
+	"context"
 	"errors"
+	"slices"
 	"testing"
 	"time"
 
@@ -72,3 +74,140 @@ func TestPrefixHandlerStoresPrefixesTheIPLookupFinds(t *testing.T) {
 		}
 	}
 }
+
+// TestPrefixHandlerAppliesEachPeersLastUpdateInABatch runs updates for one
+// prefix through the prefix handler into a real database. All updates carry the
+// same timestamp, as updates within one second of the feed do. Within a batch,
+// the last update each peer sent for the prefix decides whether that peer has a
+// live route for it, and updates from different peers are all applied. The
+// in-memory route count and prefix distribution must match the route table.
+func TestPrefixHandlerAppliesEachPeersLastUpdateInABatch(t *testing.T) {
+	const (
+		prefix           = "198.51.100.0/24"
+		prefixMaskLength = 24
+		otherPeerIP      = "2001:db8:ffff::2"
+	)
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	announce := func(peer string) *ristypes.RISMessage {
+		return &ristypes.RISMessage{
+			Peer:            peer,
+			Path:            ristypes.ASPath{testASNA, testASNB},
+			ParsedTimestamp: ts,
+			Announcements:   []ristypes.RISAnnouncement{{NextHop: peer, Prefixes: []string{prefix}}},
+		}
+	}
+	withdraw := func(peer string) *ristypes.RISMessage {
+		return &ristypes.RISMessage{Peer: peer, ParsedTimestamp: ts, Withdrawals: []string{prefix}}
+	}
+	announceAndWithdraw := func(peer string) *ristypes.RISMessage {
+		msg := announce(peer)
+		msg.Withdrawals = []string{prefix}
+
+		return msg
+	}
+
+	tests := []struct {
+		name string
+		// Each inner list is handled and then flushed as one batch.
+		batches [][]*ristypes.RISMessage
+		// The peers left with a live route for the prefix, by peer IP.
+		wantPeers []string
+	}{
+		{
+			name:      "two peers announce the prefix",
+			batches:   [][]*ristypes.RISMessage{{announce(testPeerIP), announce(otherPeerIP)}},
+			wantPeers: []string{testPeerIP, otherPeerIP},
+		},
+		{
+			name: "two peers withdraw the prefix",
+			batches: [][]*ristypes.RISMessage{
+				{announce(testPeerIP)},
+				{announce(otherPeerIP)},
+				{withdraw(testPeerIP), withdraw(otherPeerIP)},
+			},
+		},
+		{
+			name:    "a peer announces and then withdraws the prefix",
+			batches: [][]*ristypes.RISMessage{{announce(testPeerIP), withdraw(testPeerIP)}},
+		},
+		{
+			name: "a peer withdraws and then re-announces the prefix",
+			batches: [][]*ristypes.RISMessage{
+				{announce(testPeerIP)},
+				{withdraw(testPeerIP), announce(testPeerIP)},
+			},
+			wantPeers: []string{testPeerIP},
+		},
+		{
+			// RFC 4271 section 4.3: the announcement counts, not the withdrawal.
+			name: "one message announces and withdraws the prefix",
+			batches: [][]*ristypes.RISMessage{
+				{announce(testPeerIP)},
+				{announceAndWithdraw(testPeerIP)},
+			},
+			wantPeers: []string{testPeerIP},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+			if err != nil {
+				t.Fatalf("failed to create database: %v", err)
+			}
+			defer func() { _ = db.Close() }()
+
+			// Built without NewPrefixHandler's flush timer, so the test flushes itself.
+			h := &PrefixHandler{db: db, logger: logger.New()}
+			for _, batch := range tt.batches {
+				for _, msg := range batch {
+					h.HandleMessage(msg)
+				}
+				h.mu.Lock()
+				h.flushBatchLocked()
+				h.mu.Unlock()
+			}
+
+			ctx := context.Background()
+			routes, err := db.GetPrefixDetailsContext(ctx, prefix)
+			if err != nil && !errors.Is(err, database.ErrNoRoute) {
+				t.Fatalf("GetPrefixDetailsContext: %v", err)
+			}
+			var gotPeers []string
+			for _, route := range routes {
+				gotPeers = append(gotPeers, route.PeerIP)
+			}
+			if !slices.Equal(gotPeers, tt.wantPeers) {
+				t.Errorf("peers with a live route for %s = %v, want %v", prefix, gotPeers, tt.wantPeers)
+			}
+
+			stats, err := db.GetStatsContext(ctx)
+			if err != nil {
+				t.Fatalf("GetStatsContext: %v", err)
+			}
+			tableRoutesV4, _, err := db.GetLiveRouteCountsContext(ctx)
+			if err != nil {
+				t.Fatalf("GetLiveRouteCountsContext: %v", err)
+			}
+			if stats.IPv4Routes != len(tt.wantPeers) || tableRoutesV4 != len(tt.wantPeers) {
+				t.Errorf("IPv4 live routes: %d in memory, %d in the table, want %d",
+					stats.IPv4Routes, tableRoutesV4, len(tt.wantPeers))
+			}
+
+			var wantDistribution []database.PrefixDistribution
+			if len(tt.wantPeers) > 0 {
+				wantDistribution = []database.PrefixDistribution{{MaskLength: prefixMaskLength, Count: 1}}
+			}
+			queryDistributionV4, _, err := db.GetPrefixDistributionContext(ctx)
+			if err != nil {
+				t.Fatalf("GetPrefixDistributionContext: %v", err)
+			}
+			if !slices.Equal(stats.IPv4PrefixDistribution, wantDistribution) ||
+				!slices.Equal(queryDistributionV4, wantDistribution) {
+				t.Errorf("IPv4 prefix distribution: %v in memory, %v from the table, want %v",
+					stats.IPv4PrefixDistribution, queryDistributionV4, wantDistribution)
+			}
+		})
+	}
+}

internal/server/handlers.go

@@ -7,6 +7,7 @@ import (
 	"errors"
 	"net"
 	"net/http"
+	"net/netip"
 	"runtime"
 	"sort"
 	"strconv"
@@ -748,8 +749,15 @@ func (s *Server) handlePrefixDetailJSON() http.HandlerFunc {
 			return
 		}
 
-		// Combine prefix and length into CIDR notation
-		prefix := prefixParam + "/" + lenParam
+		// Look the prefix up in the text form net/netip prints, the form the
+		// tables store, so that any valid way of writing it finds the route.
+		parsed, err := netip.ParsePrefix(prefixParam + "/" + lenParam)
+		if err != nil {
+			writeJSONError(w, http.StatusBadRequest, err.Error())
+
+			return
+		}
+		prefix := parsed.Masked().String()
 
 		routes, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
 		if err != nil {
@@ -912,8 +920,15 @@ func (s *Server) handlePrefixDetail() http.HandlerFunc {
 			return
 		}
 
-		// Combine prefix and length into CIDR notation
-		prefix := prefixParam + "/" + lenParam
+		// Look the prefix up in the text form net/netip prints, the form the
+		// tables store, so that any valid way of writing it finds the route.
+		parsed, err := netip.ParsePrefix(prefixParam + "/" + lenParam)
+		if err != nil {
+			http.Error(w, "Invalid prefix", http.StatusBadRequest)
+
+			return
+		}
+		prefix := parsed.Masked().String()
 
 		routes, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
 		if err != nil {

internal/server/handlers_test.go

@@ -148,6 +148,58 @@ func TestStatsHandlersAnswerFromMemory(t *testing.T) {
 	}
 }
 
+// TestPrefixDetailHandlersParsePrefix checks that both prefix pages find a
+// stored route whatever valid text form the URL writes its prefix in, and
+// answer 400 for a prefix that does not parse. The tables store the form
+// net/netip prints. Before the fix the URL text was looked up as typed, so
+// "2001:DB8:0:0::/32" was not found although "2001:db8::/32" was stored, and
+// "foo/24" answered 500 (https://git.eeqj.de/sneak/routewatch/issues/71).
+func TestPrefixDetailHandlersParsePrefix(t *testing.T) {
+	db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
+	if err != nil {
+		t.Fatalf("database.New: %v", err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	if err := db.UpsertLiveRouteBatch([]*database.LiveRoute{
+		{
+			ID: uuid.New(), Prefix: "192.0.2.0/24", MaskLength: 24, IPVersion: 4,
+			OriginASN: 64500, PeerIP: "198.51.100.1", ASPath: []int{64500}, NextHop: "198.51.100.1",
+			LastUpdated: ts,
+		},
+		{
+			ID: uuid.New(), Prefix: "2001:db8::/32", MaskLength: 32, IPVersion: 6,
+			OriginASN: 64501, PeerIP: "2001:db8::1", ASPath: []int{64501}, NextHop: "2001:db8::1",
+			LastUpdated: ts,
+		},
+	}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	s := New(db, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+
+	tests := []struct {
+		path string
+		want int
+	}{
+		{"/api/v1/prefix/foo/24", http.StatusBadRequest},
+		{"/prefix/foo/24", http.StatusBadRequest},
+		{"/api/v1/prefix/2001:DB8:0:0::/32", http.StatusOK},
+		{"/prefix/2001:DB8:0:0::/32", http.StatusOK},
+		{"/api/v1/prefix/192.0.2.1/24", http.StatusOK},
+		{"/prefix/192.0.2.1/24", http.StatusOK},
+	}
+
+	for _, tt := range tests {
+		rec := httptest.NewRecorder()
+		s.router.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, tt.path, nil))
+		if rec.Code != tt.want {
+			t.Errorf("%s: status %d, want %d; body %s", tt.path, rec.Code, tt.want, rec.Body)
+		}
+	}
+}
+
 // settledGoroutineCount lets transient goroutines finish, then reports the
 // current count.
 func settledGoroutineCount() int {

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
@@ -124,8 +131,10 @@ type Streamer struct {
 // The metrics tracker is used to record message counts, bytes received, and connection status.
 func New(logger *logger.Logger, metrics *metrics.Tracker) *Streamer {
 	return &Streamer{
-		logger: logger,
-		url:    risLiveURL,
+		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 {
-			// Clean exit (context cancelled)
+		// Only Stop ends the streamer. A stream that ends any other way,
+		// 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
-		if ctx.Err() != nil {
-			return
-		}
-
-		// If we streamed for more than 30 seconds, reset the backoff
-		// This indicates we had a successful connection that received data
+		// A stream that lasted more than 30 seconds resets the backoff. That
+		// includes a stream dropped after streamIdleTimeout with no line, so
+		// a silent feed is retried after the first backoff delay every time.
 		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"
 
@@ -39,6 +40,14 @@ func TestNewStreamer(t *testing.T) {
 	if s.metrics != metricsTracker {
 		t.Error("metrics tracker not set correctly")
 	}
+
+	if s.idleTimeout != streamIdleTimeout {
+		t.Errorf("idleTimeout = %s, want %s", s.idleTimeout, streamIdleTimeout)
+	}
+
+	if s.minBackoff != minBackoffDelay {
+		t.Errorf("minBackoff = %s, want %s", s.minBackoff, minBackoffDelay)
+	}
 }
 
 // TestStreamDoesNotLeakTickersAcrossReconnects drives many short-lived
@@ -57,8 +66,8 @@ 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 {
-		t.Fatalf("warm-up stream returned error: %v", err)
+	if err := s.stream(context.Background()); !errors.Is(err, errStreamEnded) {
+		t.Fatalf("warm-up stream returned %v, want %v", err, errStreamEnded)
 	}
 	s.client.CloseIdleConnections()
 
@@ -66,8 +75,8 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 
 	const reconnects = 20
 	for range reconnects {
-		if err := s.stream(context.Background()); err != nil {
-			t.Fatalf("stream returned error: %v", err)
+		if err := s.stream(context.Background()); !errors.Is(err, errStreamEnded) {
+			t.Fatalf("stream returned %v, want %v", err, errStreamEnded)
 		}
 	}
 	s.client.CloseIdleConnections()
@@ -78,23 +87,35 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 	}
 }
 
-// updateHandler wants UPDATE messages and does nothing with them.
-type updateHandler struct{}
+// updateLine is one UPDATE message as RIS Live sends it.
+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 +138,117 @@ 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)
+}
+
+// TestReconnectWhenServerNeverAnswers has the server accept the first
+// request and never answer it, not even with headers. The streamer must give
+// up on it once the idle timeout passes and connect again.
+func TestReconnectWhenServerNeverAnswers(t *testing.T) {
+	const idleTimeout = 50 * time.Millisecond
+	neverAnswer := func(_ http.ResponseWriter, r *http.Request) {
+		<-r.Context().Done()
+	}
+	checkReconnect(t, neverAnswer, 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 {