Identify a live route by prefix and peer (closes #55)

A peer has at most one route for a prefix, but the live route tables were
keyed by prefix, origin AS and peer. An announcement with a new origin AS
added a second row instead of replacing the first, and a withdrawal sent in
the same message as announcements was given their origin AS and so missed
the withdrawn route.

The tables are now unique on prefix and peer. An announcement updates the
peer's row for the prefix, origin AS included, and a withdrawal deletes it
without looking at an origin AS. The origin-only delete statements, the
deletion's origin field and DeleteLiveRoute's origin parameter are removed.

Model: opus-5-5

README.md

@@ -141,10 +141,7 @@ 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**: 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.
+- **Reconnection**: Exponential backoff (5s-320s) with reset after 30s of stable connection
 - **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 25 lookups for IPv4, 113 for IPv6); prefixes are stored in the

internal/database/counts_test.go

@@ -128,7 +128,7 @@ func TestLiveCountsTrackWritesInRealtime(t *testing.T) {
 
 	// A withdrawal removes one route.
 	if err := db.DeleteLiveRouteBatch([]LiveRouteDeletion{
-		{Prefix: "203.0.113.0/24", OriginASN: 64501, PeerIP: "192.0.2.1", IPVersion: ipVersionV4},
+		{Prefix: "203.0.113.0/24", PeerIP: "192.0.2.1", IPVersion: ipVersionV4},
 	}); err != nil {
 		t.Fatalf("DeleteLiveRouteBatch: %v", err)
 	}
@@ -327,13 +327,11 @@ func announceRoutes(db *Database, batch bool, routes []*LiveRoute) error {
 }
 
 // withdrawRoutes removes routes in one DeleteLiveRouteBatch, or with one
-// DeleteLiveRoute each. It names each route by prefix and peer only, with no
-// origin ASN, as a withdrawal from the feed does when its message carries no AS
-// path.
+// DeleteLiveRoute each.
 func withdrawRoutes(db *Database, batch bool, routes []*LiveRoute) error {
 	if !batch {
 		for _, route := range routes {
-			if err := db.DeleteLiveRoute(route.Prefix, 0, route.PeerIP); err != nil {
+			if err := db.DeleteLiveRoute(route.Prefix, route.PeerIP); err != nil {
 				return err
 			}
 		}

internal/database/database.go

@@ -228,21 +228,22 @@ func (d *Database) beginTx() (*loggingTx, error) {
 }
 
 // A live-route upsert is an UPDATE followed, only when no row matched, by an
-// INSERT. The UPDATE's rows-affected count (1 for an existing key, 0 for a new
-// one) is what lets the in-memory route counters stay exact without a COUNT(*).
-// Callers hold the database write lock, so no other writer can insert the same
-// key between the two statements. The id column is set only on INSERT, so an
-// updated route keeps its original id, exactly as the previous ON CONFLICT
-// upsert did.
+// INSERT. A live route is keyed by prefix and peer, so an announcement replaces
+// the peer's route for the prefix whatever its origin AS. The UPDATE's
+// rows-affected count (1 for an existing key, 0 for a new one) is what lets the
+// in-memory route counters stay exact without a COUNT(*). Callers hold the
+// database write lock, so no other writer can insert the same key between the
+// two statements. The id column is set only on INSERT, so an updated route
+// keeps its original id, exactly as the previous ON CONFLICT upsert did.
 const (
-	updateLiveRouteV4SQL = `UPDATE live_routes_v4 SET mask_length = ?, as_path = ?, next_hop = ?,
-		last_updated = ? WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`
+	updateLiveRouteV4SQL = `UPDATE live_routes_v4 SET mask_length = ?, origin_asn = ?, as_path = ?,
+		next_hop = ?, last_updated = ? WHERE prefix = ? AND peer_ip = ?`
 
 	insertLiveRouteV4SQL = `INSERT INTO live_routes_v4 (id, prefix, mask_length, origin_asn, peer_ip,
 		as_path, next_hop, last_updated) VALUES (?, ?, ?, ?, ?, ?, ?, ?)`
 
-	updateLiveRouteV6SQL = `UPDATE live_routes_v6 SET mask_length = ?, as_path = ?, next_hop = ?,
-		last_updated = ? WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`
+	updateLiveRouteV6SQL = `UPDATE live_routes_v6 SET mask_length = ?, origin_asn = ?, as_path = ?,
+		next_hop = ?, last_updated = ? WHERE prefix = ? AND peer_ip = ?`
 
 	insertLiveRouteV6SQL = `INSERT INTO live_routes_v6 (id, prefix, mask_length, origin_asn, peer_ip,
 		as_path, next_hop, last_updated) VALUES (?, ?, ?, ?, ?, ?, ?, ?)`
@@ -261,8 +262,8 @@ const (
 // live route for its prefix.
 func upsertRouteRowV4(upd, ins, has *sql.Stmt, route *LiveRoute, pathJSON string) (
 	inserted, newPrefix bool, err error) {
-	res, err := upd.Exec(route.MaskLength, pathJSON, route.NextHop, route.LastUpdated,
-		route.Prefix, route.OriginASN, route.PeerIP)
+	res, err := upd.Exec(route.MaskLength, route.OriginASN, pathJSON, route.NextHop, route.LastUpdated,
+		route.Prefix, route.PeerIP)
 	if err != nil {
 		return false, false, err
 	}
@@ -293,8 +294,8 @@ func upsertRouteRowV4(upd, ins, has *sql.Stmt, route *LiveRoute, pathJSON string
 // live route for its prefix.
 func upsertRouteRowV6(upd, ins, has *sql.Stmt, route *LiveRoute, pathJSON string) (
 	inserted, newPrefix bool, err error) {
-	res, err := upd.Exec(route.MaskLength, pathJSON, route.NextHop, route.LastUpdated,
-		route.Prefix, route.OriginASN, route.PeerIP)
+	res, err := upd.Exec(route.MaskLength, route.OriginASN, pathJSON, route.NextHop, route.LastUpdated,
+		route.Prefix, route.PeerIP)
 	if err != nil {
 		return false, false, err
 	}
@@ -441,32 +442,18 @@ func (d *Database) DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error {
 		}
 	}()
 
-	// No longer need to separate deletions since we handle them in the loop below
-
 	// Prepare statements for both IPv4 and IPv6 tables
-	stmtV4WithOrigin, err := tx.Prepare(`DELETE FROM live_routes_v4 WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`)
+	deleteV4, err := tx.Prepare(`DELETE FROM live_routes_v4 WHERE prefix = ? AND peer_ip = ?`)
 	if err != nil {
-		return fmt.Errorf("failed to prepare IPv4 delete with origin statement: %w", err)
+		return fmt.Errorf("failed to prepare IPv4 delete statement: %w", err)
 	}
-	defer func() { _ = stmtV4WithOrigin.Close() }()
+	defer func() { _ = deleteV4.Close() }()
 
-	stmtV4WithoutOrigin, err := tx.Prepare(`DELETE FROM live_routes_v4 WHERE prefix = ? AND peer_ip = ?`)
+	deleteV6, err := tx.Prepare(`DELETE FROM live_routes_v6 WHERE prefix = ? AND peer_ip = ?`)
 	if err != nil {
-		return fmt.Errorf("failed to prepare IPv4 delete without origin statement: %w", err)
+		return fmt.Errorf("failed to prepare IPv6 delete statement: %w", err)
 	}
-	defer func() { _ = stmtV4WithoutOrigin.Close() }()
-
-	stmtV6WithOrigin, err := tx.Prepare(`DELETE FROM live_routes_v6 WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`)
-	if err != nil {
-		return fmt.Errorf("failed to prepare IPv6 delete with origin statement: %w", err)
-	}
-	defer func() { _ = stmtV6WithOrigin.Close() }()
-
-	stmtV6WithoutOrigin, err := tx.Prepare(`DELETE FROM live_routes_v6 WHERE prefix = ? AND peer_ip = ?`)
-	if err != nil {
-		return fmt.Errorf("failed to prepare IPv6 delete without origin statement: %w", err)
-	}
-	defer func() { _ = stmtV6WithoutOrigin.Close() }()
+	defer func() { _ = deleteV6.Close() }()
 
 	hasV4, err := tx.Prepare(prefixHasLiveRouteV4SQL)
 	if err != nil {
@@ -485,38 +472,18 @@ func (d *Database) DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error {
 	// Mask lengths of the prefixes this batch leaves with no live route.
 	var gonePrefixMaskLengthsV4, gonePrefixMaskLengthsV6 []int
 	for _, del := range deletions {
-		var stmt *sql.Stmt
-
-		// Select appropriate statement based on IP version and whether we have origin ASN
-		//nolint:nestif // Clear logic for selecting the right statement
-		if del.IPVersion == ipVersionV4 {
-			if del.OriginASN == 0 {
-				stmt = stmtV4WithoutOrigin
-			} else {
-				stmt = stmtV4WithOrigin
-			}
-		} else {
-			if del.OriginASN == 0 {
-				stmt = stmtV6WithoutOrigin
-			} else {
-				stmt = stmtV6WithOrigin
-			}
-		}
-
-		// Execute deletion
-		var res sql.Result
-		if del.OriginASN == 0 {
-			res, err = stmt.Exec(del.Prefix, del.PeerIP)
-		} else {
-			res, err = stmt.Exec(del.Prefix, del.OriginASN, del.PeerIP)
+		stmt := deleteV4
+		if del.IPVersion != ipVersionV4 {
+			stmt = deleteV6
 		}
 
+		res, err := stmt.Exec(del.Prefix, del.PeerIP)
 		if err != nil {
 			return fmt.Errorf("failed to delete route %s: %w", del.Prefix, err)
 		}
 
-		// A deletion with no origin ASN can remove several rows, so use the
-		// exact rows-affected count to keep the in-memory route counters right.
+		// The feed often withdraws a route that is not live, which deletes
+		// nothing, so the in-memory route counters use the rows-affected count.
 		affected, err := res.RowsAffected()
 		if err != nil {
 			return fmt.Errorf("failed to count deleted route %s: %w", del.Prefix, err)
@@ -1282,9 +1249,8 @@ func (d *Database) UpsertLiveRoute(route *LiveRoute) error {
 	return nil
 }
 
-// DeleteLiveRoute deletes a live route
-// If originASN is 0, deletes all routes for the prefix/peer combination
-func (d *Database) DeleteLiveRoute(prefix string, originASN int, peerIP string) error {
+// DeleteLiveRoute deletes the live route the peer has for the prefix
+func (d *Database) DeleteLiveRoute(prefix, peerIP string) error {
 	d.lock("DeleteLiveRoute")
 	defer d.unlock()
 
@@ -1309,22 +1275,12 @@ func (d *Database) DeleteLiveRoute(prefix string, originASN int, peerIP string)
 	}()
 
 	// Literal per-table queries (rather than one formatted with the table name)
-	// so the delete carries no dynamically built SQL. A delete with no origin
-	// ASN can remove several rows.
+	// so the delete carries no dynamically built SQL.
 	var res sql.Result
-	switch {
-	case isV4 && originASN == 0:
+	if isV4 {
 		res, err = tx.Exec(`DELETE FROM live_routes_v4 WHERE prefix = ? AND peer_ip = ?`, prefix, peerIP)
-	case isV4:
-		res, err = tx.Exec(
-			`DELETE FROM live_routes_v4 WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`,
-			prefix, originASN, peerIP)
-	case originASN == 0:
+	} else {
 		res, err = tx.Exec(`DELETE FROM live_routes_v6 WHERE prefix = ? AND peer_ip = ?`, prefix, peerIP)
-	default:
-		res, err = tx.Exec(
-			`DELETE FROM live_routes_v6 WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`,
-			prefix, originASN, peerIP)
 	}
 	if err != nil {
 		return err

internal/database/database_test.go

@@ -13,6 +13,7 @@ import (
 	"git.eeqj.de/sneak/routewatch/internal/config"
 	"git.eeqj.de/sneak/routewatch/internal/logger"
 	"github.com/google/uuid"
+	"github.com/mattn/go-sqlite3"
 )
 
 // tempStoreMemory is the PRAGMA temp_store value meaning "hold temp B-trees in
@@ -122,8 +123,8 @@ func TestGetIPInfoFindsMostSpecificLiveRoute(t *testing.T) {
 	lookup("2001:db9::1", "")
 
 	err = db.DeleteLiveRouteBatch([]LiveRouteDeletion{
-		{Prefix: "10.1.2.0/24", OriginASN: 64502, PeerIP: testPeerIP, IPVersion: ipVersionV4},
-		{Prefix: "2001:db8:1:2::/64", OriginASN: 64502, PeerIP: testPeerIP, IPVersion: ipVersionV6},
+		{Prefix: "10.1.2.0/24", PeerIP: testPeerIP, IPVersion: ipVersionV4},
+		{Prefix: "2001:db8:1:2::/64", PeerIP: testPeerIP, IPVersion: ipVersionV6},
 	})
 	if err != nil {
 		t.Fatalf("DeleteLiveRouteBatch: %v", err)
@@ -205,6 +206,40 @@ func TestGetIPInfoIgnoresShortRoutes(t *testing.T) {
 	lookup("2001:db8::1", "2001:db8::/32", 64502)
 }
 
+// TestLiveRouteTablesHoldOneRoutePerPrefixAndPeer checks that each live route
+// table refuses a second route from a peer for a prefix, even with another
+// origin AS. An upsert updates the peer's route before it would insert one, so
+// only an insert straight into the table shows what the table allows.
+func TestLiveRouteTablesHoldOneRoutePerPrefixAndPeer(t *testing.T) {
+	db, err := New(&config.Config{StateDir: t.TempDir()}, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	const otherOriginASN = 64501
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	for _, tt := range []struct {
+		route     *LiveRoute
+		insertSQL string
+	}{
+		{route: mkV4Route(t, "198.51.100.0/24", 64500, ts), insertSQL: insertLiveRouteV4SQL},
+		{route: mkV6Route("2001:db8::/32", 64500, ts), insertSQL: insertLiveRouteV6SQL},
+	} {
+		if err := db.UpsertLiveRoute(tt.route); err != nil {
+			t.Fatalf("UpsertLiveRoute(%s): %v", tt.route.Prefix, err)
+		}
+
+		_, err := db.db.Exec(tt.insertSQL, uuid.NewString(), tt.route.Prefix, tt.route.MaskLength,
+			otherOriginASN, tt.route.PeerIP, "[]", tt.route.NextHop, ts)
+		var sqliteErr sqlite3.Error
+		if !errors.As(err, &sqliteErr) || sqliteErr.ExtendedCode != sqlite3.ErrConstraintUnique {
+			t.Errorf("second route from peer %s for %s: error %v, want a UNIQUE constraint error",
+				tt.route.PeerIP, tt.route.Prefix, err)
+		}
+	}
+}
+
 // 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
@@ -306,7 +341,6 @@ func TestVacuumReturnsFreePages(t *testing.T) {
 		routes = append(routes, route)
 		deletions = append(deletions, LiveRouteDeletion{
 			Prefix:    route.Prefix,
-			OriginASN: asn,
 			PeerIP:    route.PeerIP,
 			IPVersion: ipVersionV4,
 		})

internal/database/interface.go

@@ -55,7 +55,7 @@ type Store interface {
 	// Live route operations
 	UpsertLiveRoute(route *LiveRoute) error
 	UpsertLiveRouteBatch(routes []*LiveRoute) error
-	DeleteLiveRoute(prefix string, originASN int, peerIP string) error
+	DeleteLiveRoute(prefix, peerIP string) error
 	DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error
 	GetPrefixDistribution() (ipv4 []PrefixDistribution, ipv6 []PrefixDistribution, err error)
 	GetPrefixDistributionContext(ctx context.Context) (ipv4 []PrefixDistribution, ipv6 []PrefixDistribution, err error)

internal/database/models.go

@@ -114,7 +114,6 @@ type IPInfo struct {
 // LiveRouteDeletion represents parameters for deleting a live route
 type LiveRouteDeletion struct {
 	Prefix    string
-	OriginASN int
 	PeerIP    string
 	IPVersion int
 }

internal/database/schema.sql

@@ -107,7 +107,7 @@ CREATE TABLE IF NOT EXISTS live_routes_v4 (
 	as_path TEXT NOT NULL, -- JSON array
 	next_hop TEXT NOT NULL,
 	last_updated DATETIME NOT NULL,
-	UNIQUE(prefix, origin_asn, peer_ip)
+	UNIQUE(prefix, peer_ip) -- a peer has at most one route for a prefix
 );
 
 -- IPv6 routing table maintained by PrefixHandler
@@ -120,7 +120,7 @@ CREATE TABLE IF NOT EXISTS live_routes_v6 (
 	as_path TEXT NOT NULL, -- JSON array
 	next_hop TEXT NOT NULL,
 	last_updated DATETIME NOT NULL,
-	UNIQUE(prefix, origin_asn, peer_ip)
+	UNIQUE(prefix, peer_ip) -- a peer has at most one route for a prefix
 );
 
 -- Indexes for live_routes_v4 table

internal/routewatch/app_integration_test.go

@@ -205,7 +205,7 @@ func (m *mockStore) UpsertLiveRoute(route *database.LiveRoute) error {
 }
 
 // DeleteLiveRoute mock implementation
-func (m *mockStore) DeleteLiveRoute(prefix string, originASN int, peerIP string) error {
+func (m *mockStore) DeleteLiveRoute(prefix, peerIP string) error {
 	// Simple mock - just return nil
 	return nil
 }

internal/routewatch/cli.go

@@ -20,6 +20,8 @@ const (
 	debugInterval = 60 * time.Second
 	// bytesPerMB is bytes per megabyte
 	bytesPerMB = 1024 * 1024
+	// failureExitCode is the exit status when RouteWatch stops on an error
+	failureExitCode = 1
 )
 
 // logDebugStats logs goroutine count and memory usage
@@ -83,6 +85,9 @@ func CLIEntry() {
 					go func() {
 						if err := rw.Run(ctx); err != nil {
 							logger.Error("RouteWatch error", "error", err)
+							if err := shutdowner.Shutdown(fx.ExitCode(failureExitCode)); err != nil {
+								logger.Error("Failed to shutdown gracefully", "error", err)
+							}
 						}
 					}()
 

internal/routewatch/prefixhandler.go

@@ -108,14 +108,14 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
 	// 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).
+	// A withdrawal names only the prefix and peer: the message's AS path, and
+	// so its origin AS, belongs to the announced prefixes.
 	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 {
@@ -221,7 +221,6 @@ func (h *PrefixHandler) flushBatchLocked() {
 			// Create deletion record for batch delete
 			routesToDelete = append(routesToDelete, database.LiveRouteDeletion{
 				Prefix:    update.prefix,
-				OriginASN: update.originASN,
 				PeerIP:    update.peer,
 				IPVersion: ipVersion,
 			})
@@ -425,51 +424,24 @@ func (h *PrefixHandler) processAnnouncementDirect(update prefixUpdate) {
 // processWithdrawalDirect handles removing a route directly without prefix table
 // nolint:unused // kept for potential future use
 func (h *PrefixHandler) processWithdrawalDirect(update prefixUpdate) {
-	// For withdrawals, we need to delete the route from live_routes
-	if update.originASN > 0 {
-		if err := h.db.DeleteLiveRoute(update.prefix, update.originASN, update.peer); err != nil {
-			h.logger.Error("Failed to delete live route",
-				"prefix", update.prefix,
-				"origin_asn", update.originASN,
-				"peer", update.peer,
-				"error", err,
-			)
-		}
-	} else {
-		// If no origin ASN, just delete all routes for this prefix from this peer
-		if err := h.db.DeleteLiveRoute(update.prefix, 0, update.peer); err != nil {
-			h.logger.Error("Failed to delete live route (no origin ASN)",
-				"prefix", update.prefix,
-				"peer", update.peer,
-				"error", err,
-			)
-		}
+	if err := h.db.DeleteLiveRoute(update.prefix, update.peer); err != nil {
+		h.logger.Error("Failed to delete live route",
+			"prefix", update.prefix,
+			"peer", update.peer,
+			"error", err,
+		)
 	}
 }
 
 // processWithdrawal handles removing a route from the live routing table
 // nolint:unused // kept for potential future use
 func (h *PrefixHandler) processWithdrawal(_ *database.Prefix, update prefixUpdate) {
-	// For withdrawals, we need to delete the route from live_routes
-	// Since we have the origin ASN from the update, we can delete the specific route
-	if update.originASN > 0 {
-		if err := h.db.DeleteLiveRoute(update.prefix, update.originASN, update.peer); err != nil {
-			h.logger.Error("Failed to delete live route",
-				"prefix", update.prefix,
-				"origin_asn", update.originASN,
-				"peer", update.peer,
-				"error", err,
-			)
-		}
-	} else {
-		// If no origin ASN, just delete all routes for this prefix from this peer
-		if err := h.db.DeleteLiveRoute(update.prefix, 0, update.peer); err != nil {
-			h.logger.Error("Failed to delete live route (no origin ASN)",
-				"prefix", update.prefix,
-				"peer", update.peer,
-				"error", err,
-			)
-		}
+	if err := h.db.DeleteLiveRoute(update.prefix, update.peer); err != nil {
+		h.logger.Error("Failed to delete live route",
+			"prefix", update.prefix,
+			"peer", update.peer,
+			"error", err,
+		)
 	}
 }
 

internal/routewatch/prefixhandler_test.go

@@ -3,6 +3,7 @@ package routewatch
 import (
 	"context"
 	"errors"
+	"fmt"
 	"slices"
 	"testing"
 	"time"
@@ -182,32 +183,155 @@ func TestPrefixHandlerAppliesEachPeersLastUpdateInABatch(t *testing.T) {
 				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)
-			}
+			assertCounts(t, db, ipv4Version, len(tt.wantPeers), wantDistribution)
 		})
 	}
 }
+
+// TestPrefixHandlerKeepsOneRoutePerPrefixAndPeer runs updates from one peer
+// through the prefix handler into a real database, flushing each message as its
+// own batch, once with IPv4 prefixes and once with IPv6 prefixes. A peer has at
+// most one route for a prefix: a later announcement replaces it whatever its
+// origin AS, and a withdrawal removes it whatever AS the message's AS path ends
+// in. The in-memory route count and prefix distribution must match the route
+// table.
+func TestPrefixHandlerKeepsOneRoutePerPrefixAndPeer(t *testing.T) {
+	families := []struct {
+		ipVersion           int
+		prefix, otherPrefix string
+		// The mask length of both prefixes.
+		maskLength int
+	}{
+		{ipVersion: ipv4Version, prefix: "198.51.100.0/24", otherPrefix: "203.0.113.0/24", maskLength: 24},
+		{ipVersion: ipv6Version, prefix: "2001:db8:1::/48", otherPrefix: "2001:db8:2::/48", maskLength: 48},
+	}
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	announce := func(path ristypes.ASPath, announced string) *ristypes.RISMessage {
+		return &ristypes.RISMessage{
+			Peer:            testPeerIP,
+			Path:            path,
+			ParsedTimestamp: ts,
+			Announcements:   []ristypes.RISAnnouncement{{NextHop: testPeerIP, Prefixes: []string{announced}}},
+		}
+	}
+
+	for _, family := range families {
+		prefix, otherPrefix := family.prefix, family.otherPrefix
+
+		// One message that announces otherPrefix with origin AS testASNB and
+		// withdraws prefix.
+		announceOtherAndWithdraw := announce(ristypes.ASPath{testASNA, testASNB}, otherPrefix)
+		announceOtherAndWithdraw.Withdrawals = []string{prefix}
+
+		tests := []struct {
+			name     string
+			messages []*ristypes.RISMessage
+			// The origin AS of the peer's live route for each prefix. A prefix
+			// left out must have no live route.
+			wantOrigins map[string]int
+		}{
+			{
+				name: "the peer announces the prefix with a new origin AS",
+				messages: []*ristypes.RISMessage{
+					announce(ristypes.ASPath{testASNA}, prefix),
+					announce(ristypes.ASPath{testASNA, testASNB}, prefix),
+				},
+				wantOrigins: map[string]int{prefix: testASNB},
+			},
+			{
+				name: "one message announces a prefix with another origin AS and withdraws the prefix",
+				messages: []*ristypes.RISMessage{
+					announce(ristypes.ASPath{testASNA}, prefix),
+					announceOtherAndWithdraw,
+				},
+				wantOrigins: map[string]int{otherPrefix: testASNB},
+			},
+		}
+
+		for _, tt := range tests {
+			t.Run(fmt.Sprintf("IPv%d %s", family.ipVersion, 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 _, msg := range tt.messages {
+					h.HandleMessage(msg)
+					h.mu.Lock()
+					h.flushBatchLocked()
+					h.mu.Unlock()
+				}
+
+				for _, p := range []string{prefix, otherPrefix} {
+					routes, err := db.GetPrefixDetailsContext(context.Background(), p)
+					if err != nil && !errors.Is(err, database.ErrNoRoute) {
+						t.Fatalf("GetPrefixDetailsContext(%s): %v", p, err)
+					}
+					var gotOrigins []int
+					for _, route := range routes {
+						gotOrigins = append(gotOrigins, route.OriginASN)
+					}
+					var wantOrigins []int
+					if origin, ok := tt.wantOrigins[p]; ok {
+						wantOrigins = []int{origin}
+					}
+					if !slices.Equal(gotOrigins, wantOrigins) {
+						t.Errorf("origin ASes of the live routes for %s = %v, want %v", p, gotOrigins, wantOrigins)
+					}
+				}
+
+				wantDistribution := []database.PrefixDistribution{
+					{MaskLength: family.maskLength, Count: len(tt.wantOrigins)},
+				}
+				assertCounts(t, db, family.ipVersion, len(tt.wantOrigins), wantDistribution)
+			})
+		}
+	}
+}
+
+// assertCounts checks that the live route count and prefix distribution of one
+// IP version are wantRoutes and wantDistribution, both as held in memory and as
+// read from the route table.
+func assertCounts(t *testing.T, db *database.Database, ipVersion, wantRoutes int,
+	wantDistribution []database.PrefixDistribution) {
+	t.Helper()
+
+	ctx := context.Background()
+	stats, err := db.GetStatsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetStatsContext: %v", err)
+	}
+	tableRoutesV4, tableRoutesV6, err := db.GetLiveRouteCountsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetLiveRouteCountsContext: %v", err)
+	}
+	queryDistributionV4, queryDistributionV6, err := db.GetPrefixDistributionContext(ctx)
+	if err != nil {
+		t.Fatalf("GetPrefixDistributionContext: %v", err)
+	}
+
+	memoryRoutes, tableRoutes := stats.IPv4Routes, tableRoutesV4
+	memoryDistribution, queryDistribution := stats.IPv4PrefixDistribution, queryDistributionV4
+	if ipVersion == ipv6Version {
+		memoryRoutes, tableRoutes = stats.IPv6Routes, tableRoutesV6
+		memoryDistribution, queryDistribution = stats.IPv6PrefixDistribution, queryDistributionV6
+	}
+
+	if memoryRoutes != wantRoutes || tableRoutes != wantRoutes {
+		t.Errorf("IPv%d live routes: %d in memory, %d in the table, want %d",
+			ipVersion, memoryRoutes, tableRoutes, wantRoutes)
+	}
+	if !slices.Equal(memoryDistribution, wantDistribution) ||
+		!slices.Equal(queryDistribution, wantDistribution) {
+		t.Errorf("IPv%d prefix distribution: %v in memory, %v from the table, want %v",
+			ipVersion, memoryDistribution, queryDistribution, wantDistribution)
+	}
+}

internal/server/server.go

@@ -3,6 +3,8 @@ package server
 
 import (
 	"context"
+	"fmt"
+	"net"
 	"net/http"
 	"strconv"
 	"time"
@@ -53,9 +55,12 @@ func New(db database.Store, streamer *streamer.Streamer, logger *logger.Logger,
 	return s
 }
 
-// Start starts the HTTP server
+// Start listens on the configured port and serves HTTP in the background.
+// It returns an error if it cannot listen, for example because the port is
+// already in use.
 func (s *Server) Start() error {
 	const (
+		network           = "tcp"
 		readHeaderTimeout = 40 * time.Second
 		readTimeout       = 60 * time.Second
 		writeTimeout      = 60 * time.Second
@@ -73,10 +78,15 @@ func (s *Server) Start() error {
 
 	s.logger.Info("Starting HTTP server", "port", s.port, "addr", s.srv.Addr)
 
-	// Start in goroutine but log when actually listening
+	listener, err := net.Listen(network, s.srv.Addr)
+	if err != nil {
+		return fmt.Errorf("starting HTTP server: %w", err)
+	}
+
+	s.logger.Info("HTTP server listening", "addr", listener.Addr().String())
+
 	go func() {
-		s.logger.Info("HTTP server listening", "addr", s.srv.Addr)
-		if err := s.srv.ListenAndServe(); err != nil && err != http.ErrServerClosed {
+		if err := s.srv.Serve(listener); err != nil && err != http.ErrServerClosed {
 			s.logger.Error("HTTP server error", "error", err)
 		}
 	}()

internal/server/server_test.go

@@ -0,0 +1,33 @@
+package server
+
+import (
+	"context"
+	"net"
+	"testing"
+
+	"git.eeqj.de/sneak/routewatch/internal/config"
+	"git.eeqj.de/sneak/routewatch/internal/logger"
+	"git.eeqj.de/sneak/routewatch/internal/metrics"
+	"git.eeqj.de/sneak/routewatch/internal/streamer"
+)
+
+// TestStartFailsWhenPortIsHeld holds a port with a listener of its own and
+// checks that Start on that port returns an error. Before the fix Start
+// listened in a goroutine, only logged the failure and returned nil, so the
+// daemon kept running without its web server.
+func TestStartFailsWhenPortIsHeld(t *testing.T) {
+	held, err := net.Listen("tcp", ":0")
+	if err != nil {
+		t.Fatalf("failed to open the test listener: %v", err)
+	}
+	defer func() { _ = held.Close() }()
+
+	port := held.Addr().(*net.TCPAddr).Port
+	log := logger.New()
+	s := New(nil, streamer.New(log, metrics.New()), log, &config.Config{Port: port})
+	defer func() { _ = s.Stop(context.Background()) }()
+
+	if err := s.Start(); err == nil {
+		t.Fatalf("Start returned nil while port %d was held by another listener", port)
+	}
+}

internal/streamer/streamer.go

@@ -7,7 +7,6 @@ import (
 	"compress/gzip"
 	"context"
 	"encoding/json"
-	"errors"
 	"fmt"
 	"io"
 	"math"
@@ -49,7 +48,6 @@ 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
@@ -60,9 +58,6 @@ 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.
@@ -112,8 +107,6 @@ 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
@@ -131,10 +124,8 @@ 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,
-		idleTimeout: streamIdleTimeout,
-		minBackoff:  minBackoffDelay,
+		logger: logger,
+		url:    risLiveURL,
 		client: &http.Client{
 			Timeout: 0, // No timeout for streaming
 			Transport: &http.Transport{
@@ -414,7 +405,7 @@ func (s *Streamer) updateMetrics(messageBytes int) {
 
 // streamWithReconnect handles streaming with automatic reconnection and exponential backoff
 func (s *Streamer) streamWithReconnect(ctx context.Context) {
-	backoffDelay := s.minBackoff
+	backoffDelay := minBackoffDelay
 	consecutiveFailures := 0
 
 	for {
@@ -431,9 +422,8 @@ func (s *Streamer) streamWithReconnect(ctx context.Context) {
 		err := s.stream(ctx)
 		streamDuration := time.Since(startTime)
 
-		// 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 {
+		if err == nil {
+			// Clean exit (context cancelled)
 			return
 		}
 
@@ -444,13 +434,17 @@ func (s *Streamer) streamWithReconnect(ctx context.Context) {
 			"stream_duration", streamDuration)
 		s.metrics.SetConnected(false)
 
-		// 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.
+		// 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
 		if streamDuration > 30*time.Second {
 			s.logger.Info("Resetting backoff delay due to successful connection",
 				"stream_duration", streamDuration)
-			backoffDelay = s.minBackoff
+			backoffDelay = minBackoffDelay
 			consecutiveFailures = 0
 		} else {
 			// Increment consecutive failures
@@ -483,13 +477,7 @@ func (s *Streamer) stream(ctx context.Context) error {
 	connCtx, connCancel := context.WithCancel(ctx)
 	defer connCancel()
 
-	// 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)
+	req, err := http.NewRequestWithContext(ctx, "GET", s.url, nil)
 	if err != nil {
 		return fmt.Errorf("failed to create request: %w", err)
 	}
@@ -571,8 +559,6 @@ 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")
@@ -721,19 +707,11 @@ 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 errStreamEnded
+	return nil
 }
 
 // shouldDropForBackpressure determines if a message should be dropped based on queue utilization

internal/streamer/streamer_test.go

@@ -7,7 +7,6 @@ import (
 	"net/http"
 	"net/http/httptest"
 	"runtime"
-	"sync/atomic"
 	"testing"
 	"time"
 
@@ -40,14 +39,6 @@ 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
@@ -66,8 +57,8 @@ 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()); !errors.Is(err, errStreamEnded) {
-		t.Fatalf("warm-up stream returned %v, want %v", err, errStreamEnded)
+	if err := s.stream(context.Background()); err != nil {
+		t.Fatalf("warm-up stream returned error: %v", err)
 	}
 	s.client.CloseIdleConnections()
 
@@ -75,8 +66,8 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 
 	const reconnects = 20
 	for range reconnects {
-		if err := s.stream(context.Background()); !errors.Is(err, errStreamEnded) {
-			t.Fatalf("stream returned %v, want %v", err, errStreamEnded)
+		if err := s.stream(context.Background()); err != nil {
+			t.Fatalf("stream returned error: %v", err)
 		}
 	}
 	s.client.CloseIdleConnections()
@@ -87,35 +78,23 @@ func TestStreamDoesNotLeakTickersAcrossReconnects(t *testing.T) {
 	}
 }
 
-// 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{}
-}
+// updateHandler wants UPDATE messages and does nothing with them.
+type updateHandler 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, updateLine)
+		_, _ = io.WriteString(w, line)
 	}))
 	defer srv.Close()
 
@@ -138,117 +117,6 @@ 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 {