Remove a peer's live routes when its session goes down (closes #56)

RIS sends no withdrawals for the routes of a peer whose BGP session goes
down, so those routes stayed in the live tables for good.

The streamer now hands STATE messages to the prefix handler and never drops
one for backpressure: one that finds the queue full is held, and queued in
feed order ahead of anything else for that handler once there is room. On a
down, the prefix handler drops the peer's updates still in its batch and
notes the peer; the next flush deletes all of the peer's live routes, IPv4
and IPv6, before writing the batch, so updates sent after the down are kept.
The live route tables' unique key now leads with peer_ip, so it also finds a
peer's routes.

Model: opus-5-5

internal/database/counts_test.go

@@ -349,6 +349,89 @@ func withdrawRoutes(db *Database, batch bool, routes []*LiveRoute) error {
 	return db.DeleteLiveRouteBatch(deletions)
 }
 
+// TestDeleteLiveRoutesOfPeer gives two peers IPv4 and IPv6 live routes, some
+// for the same prefixes, and deletes all of one peer's routes. The other peer's
+// routes stay, and the route counts and prefix distribution held in memory
+// equal what is read from the route tables.
+func TestDeleteLiveRoutesOfPeer(t *testing.T) {
+	const (
+		peerIP      = "192.0.2.1"
+		otherPeerIP = "192.0.2.2"
+		sharedV4    = "198.51.100.0/24"
+		sharedV6    = "2001:db8::/32"
+	)
+
+	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() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	ofPeer := func(peer string, route *LiveRoute) *LiveRoute {
+		route.PeerIP = peer
+
+		return route
+	}
+	onlyV6 := mkV6Route("2001:db8:1::/48", 64503, ts)
+	onlyV6.MaskLength = 48
+	if err := db.UpsertLiveRouteBatch([]*LiveRoute{
+		ofPeer(peerIP, mkV4Route(t, sharedV4, 64500, ts)),
+		ofPeer(otherPeerIP, mkV4Route(t, sharedV4, 64500, ts)),
+		ofPeer(peerIP, mkV4Route(t, "203.0.113.0/24", 64501, ts)),
+		ofPeer(peerIP, mkV6Route(sharedV6, 64502, ts)),
+		ofPeer(otherPeerIP, mkV6Route(sharedV6, 64502, ts)),
+		ofPeer(peerIP, onlyV6),
+	}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	// peerIP has two IPv4 and two IPv6 routes; a second delete finds none.
+	for _, want := range []int{4, 0} {
+		deleted, err := db.DeleteLiveRoutesOfPeer(peerIP)
+		if err != nil {
+			t.Fatalf("DeleteLiveRoutesOfPeer: %v", err)
+		}
+		if deleted != want {
+			t.Errorf("DeleteLiveRoutesOfPeer deleted %d routes, want %d", deleted, want)
+		}
+	}
+
+	ctx := context.Background()
+	for _, prefix := range []string{sharedV4, sharedV6} {
+		routes, _, err := db.GetPrefixDetailsContext(ctx, prefix)
+		if err != nil {
+			t.Fatalf("GetPrefixDetailsContext(%s): %v", prefix, err)
+		}
+		if len(routes) != 1 || routes[0].PeerIP != otherPeerIP {
+			t.Errorf("live routes for %s = %+v, want only the one from %s", prefix, routes, otherPeerIP)
+		}
+	}
+
+	stats, err := db.GetStatsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetStatsContext: %v", err)
+	}
+	tableV4, tableV6, err := db.GetLiveRouteCountsContext(ctx)
+	if err != nil {
+		t.Fatalf("GetLiveRouteCountsContext: %v", err)
+	}
+	if stats.IPv4Routes != 1 || tableV4 != 1 || stats.IPv6Routes != 1 || tableV6 != 1 {
+		t.Errorf("live routes (IPv4, IPv6): (%d, %d) in memory, (%d, %d) in the tables, want (1, 1)",
+			stats.IPv4Routes, stats.IPv6Routes, tableV4, tableV6)
+	}
+	queryV4, queryV6, err := db.GetPrefixDistributionContext(ctx)
+	if err != nil {
+		t.Fatalf("GetPrefixDistributionContext: %v", err)
+	}
+	wantV4 := []PrefixDistribution{{MaskLength: 24, Count: 1}}
+	wantV6 := []PrefixDistribution{{MaskLength: 32, Count: 1}}
+	assertDistribution(t, "IPv4 distribution", stats.IPv4PrefixDistribution, wantV4)
+	assertDistribution(t, "IPv6 distribution", stats.IPv6PrefixDistribution, wantV6)
+	assertDistribution(t, "IPv4 distribution query", queryV4, wantV4)
+	assertDistribution(t, "IPv6 distribution query", queryV6, wantV6)
+}
+
 // TestStatsRouteTimestamps checks the oldest/newest route timestamps are read
 // from the right rows across both tables and parse into time.Time. The old
 // MIN/MAX union query read its result into *time.Time, which the driver could

internal/database/database.go

@@ -542,6 +542,91 @@ func (d *Database) DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error {
 	return nil
 }
 
+// Deleting all live routes of a peer. The first query of each table selects the
+// mask length of each of the peer's routes whose prefix no other peer has a
+// live route for: the delete takes those prefixes out of the prefix
+// distribution.
+const (
+	peerOnlyRouteMaskLengthsV4SQL = `SELECT r.mask_length FROM live_routes_v4 AS r WHERE r.peer_ip = ?
+		AND NOT EXISTS (SELECT 1 FROM live_routes_v4 AS o WHERE o.prefix = r.prefix AND o.peer_ip != r.peer_ip)`
+	deletePeerLiveRoutesV4SQL = `DELETE FROM live_routes_v4 WHERE peer_ip = ?`
+
+	peerOnlyRouteMaskLengthsV6SQL = `SELECT r.mask_length FROM live_routes_v6 AS r WHERE r.peer_ip = ?
+		AND NOT EXISTS (SELECT 1 FROM live_routes_v6 AS o WHERE o.prefix = r.prefix AND o.peer_ip != r.peer_ip)`
+	deletePeerLiveRoutesV6SQL = `DELETE FROM live_routes_v6 WHERE peer_ip = ?`
+)
+
+// deletePeerRoutes deletes all live routes of peerIP from one table, given that
+// table's two statements above. It returns how many routes it deleted and the
+// mask length of each prefix it left with no live route.
+func deletePeerRoutes(tx *loggingTx, onlyRouteMaskLengthsSQL, deleteSQL, peerIP string) (
+	deleted int64, goneMaskLengths []int, err error) {
+	rows, err := tx.Query(onlyRouteMaskLengthsSQL, peerIP)
+	if err != nil {
+		return 0, nil, err
+	}
+	defer func() { _ = rows.Close() }()
+	for rows.Next() {
+		var maskLength int
+		if err := rows.Scan(&maskLength); err != nil {
+			return 0, nil, err
+		}
+		goneMaskLengths = append(goneMaskLengths, maskLength)
+	}
+	if err := rows.Err(); err != nil {
+		return 0, nil, err
+	}
+
+	res, err := tx.Exec(deleteSQL, peerIP)
+	if err != nil {
+		return 0, nil, err
+	}
+	deleted, err = res.RowsAffected()
+	if err != nil {
+		return 0, nil, err
+	}
+
+	return deleted, goneMaskLengths, nil
+}
+
+// DeleteLiveRoutesOfPeer deletes every live route of the peer, IPv4 and IPv6,
+// in one transaction, as when the peer's BGP session goes down. It returns how
+// many routes it deleted.
+func (d *Database) DeleteLiveRoutesOfPeer(peerIP string) (int, error) {
+	d.lock("DeleteLiveRoutesOfPeer")
+	defer d.unlock()
+
+	tx, err := d.beginTx()
+	if err != nil {
+		return 0, fmt.Errorf("failed to begin transaction: %w", err)
+	}
+	defer func() {
+		if err := tx.Rollback(); err != nil && err != sql.ErrTxDone {
+			d.logger.Error("Failed to rollback transaction", "error", err)
+		}
+	}()
+
+	deletedV4, goneMaskLengthsV4, err := deletePeerRoutes(tx,
+		peerOnlyRouteMaskLengthsV4SQL, deletePeerLiveRoutesV4SQL, peerIP)
+	if err != nil {
+		return 0, fmt.Errorf("failed to delete the IPv4 routes of peer %s: %w", peerIP, err)
+	}
+	deletedV6, goneMaskLengthsV6, err := deletePeerRoutes(tx,
+		peerOnlyRouteMaskLengthsV6SQL, deletePeerLiveRoutesV6SQL, peerIP)
+	if err != nil {
+		return 0, fmt.Errorf("failed to delete the IPv6 routes of peer %s: %w", peerIP, err)
+	}
+
+	if err := tx.Commit(); err != nil {
+		return 0, fmt.Errorf("failed to commit transaction: %w", err)
+	}
+
+	d.counts.addRoutes(-int(deletedV4), -int(deletedV6))
+	d.counts.addToDistribution(goneMaskLengthsV4, goneMaskLengthsV6, -1)
+
+	return int(deletedV4 + deletedV6), nil
+}
+
 // UpdatePrefixesBatch updates the last_seen time for multiple prefixes in a single transaction
 func (d *Database) UpdatePrefixesBatch(prefixes map[string]time.Time) error {
 	if len(prefixes) == 0 {

internal/database/interface.go

@@ -57,6 +57,7 @@ type Store interface {
 	UpsertLiveRouteBatch(routes []*LiveRoute) error
 	DeleteLiveRoute(prefix, peerIP string) error
 	DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error
+	DeleteLiveRoutesOfPeer(peerIP string) (deleted int, err error)
 	GetPrefixDistribution() (ipv4 []PrefixDistribution, ipv6 []PrefixDistribution, err error)
 	GetPrefixDistributionContext(ctx context.Context) (ipv4 []PrefixDistribution, ipv6 []PrefixDistribution, err error)
 	GetLiveRouteCounts() (ipv4Count, ipv6Count int, err error)

internal/database/schema.sql

@@ -107,7 +107,10 @@ 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, peer_ip) -- a peer has at most one route for a prefix
+	-- A peer has at most one route for a prefix. peer_ip comes first so that
+	-- this key also finds all of a peer's routes, when its BGP session goes
+	-- down; lookups by prefix use the prefix index below.
+	UNIQUE(peer_ip, prefix)
 );
 
 -- IPv6 routing table maintained by PrefixHandler
@@ -120,7 +123,8 @@ 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, peer_ip) -- a peer has at most one route for a prefix
+	-- As in live_routes_v4: peer_ip first, to find all of a peer's routes.
+	UNIQUE(peer_ip, prefix)
 );
 
 -- Indexes for live_routes_v4 table