Sample the prefix-length pages with bounded lookups, each prefix once (closes #66)

/prefixlength/{length} and /prefixlength6/{length} grouped every live route
of that length by prefix, shuffled all the prefixes and joined the routes
back, so a request grew with the number of routes and listed a prefix once
for each peer that shared its newest time.

Now the first 5001 prefixes of that length are read, one index search each.
With at most 5000, the page lists 500 of them, shuffled; otherwise random
row-id lookups on the mask length index pick them, at most four per prefix
asked for, filled up from the shuffled ones. Each prefix is shown once, with
its newest route, found in one search of a new prefix and last_updated index
that replaces the prefix-only one.

Model: opus-5-5

README.md

@@ -233,6 +233,15 @@ What happens at each limit:
   keeps running.
 - Handler queues: each of the four handler queues holds at most 20,000 messages.
   When a queue fills, the streamer drops messages instead of blocking.
+- Handler batches: while the AS, peer or prefix handler writes a batch to the
+  database, the updates that arrive collect in its next batch, which stops at
+  ten times the batch size: 300,000 updates for the AS handler
+  (`asnBatchMaxSize`, about 10 MiB), 100,000 for the peer handler
+  (`peerBatchMaxSize`, about 10 MiB) and 250,000 for the prefix handler
+  (`prefixBatchMaxSize`, about 60 MiB). At that size the handler takes nothing
+  more off its queue until the write ends, so the queue fills and the streamer
+  drops messages as above. With the batches being written, the three handlers'
+  batches take at most about 0.25 GiB of the Go heap's 1.5 GiB.
 
 With `DEBUG=routewatch` the daemon logs a `System stats` line every 60 seconds
 with the goroutine count and Go memory figures.

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

@@ -240,21 +240,22 @@ 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 (?, ?, ?, ?, ?, ?, ?, ?)`
@@ -273,8 +274,8 @@ 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
 	}
@@ -305,8 +306,8 @@ 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
 	}
@@ -453,32 +454,18 @@ 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 {
@@ -497,38 +484,18 @@ 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)
@@ -1294,9 +1261,8 @@ 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()
 
@@ -1321,22 +1287,12 @@ 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
@@ -1726,9 +1682,10 @@ func (d *Database) GetRandomPrefixesByLength(maskLength, ipVersion, limit int) (
 // GetRandomPrefixesByLengthContext returns limit prefixes with the given mask
 // length picked at random, or all of them when there are fewer, each once with
 // its most recently updated live route, for a limit up to
-// maxPrefixesToShuffle. However many routes there are, it reads at most
-// maxPrefixesToShuffle+1 prefixes and makes at most sampleLookupsPerPrefix
-// lookups for each prefix asked for.
+// maxPrefixesToShuffle. However many routes there are, it makes at most
+// maxPrefixesToShuffle+1 index searches to list prefixes, at most
+// sampleLookupsPerPrefix random lookups for each prefix asked for, and one
+// index search for each listed prefix's newest route.
 func (d *Database) GetRandomPrefixesByLengthContext(
 	ctx context.Context, maskLength, ipVersion, limit int) ([]LiveRoute, error) {
 	tableName := "live_routes_v4"
@@ -1756,35 +1713,39 @@ func (d *Database) GetRandomPrefixesByLengthContext(
 	return d.newestRouteOfEachPrefix(ctx, tableName, ipVersion, prefixes)
 }
 
-// firstPrefixesByLength returns the first n distinct prefixes, in text order,
-// with the given mask length. It reads only their entries in the mask length
-// and prefix index.
+// firstPrefixesByLength returns the first n prefixes, in text order, with the
+// given mask length. Each prefix is one search of the mask length and prefix
+// index for the prefix after the previous one, so the read does not grow with
+// the number of peers that carry a prefix, as a SELECT DISTINCT would.
 func (d *Database) firstPrefixesByLength(
 	ctx context.Context, tableName string, maskLength, n int) ([]string, error) {
 	//nolint:gosec // Table name is hardcoded based on IP version
 	query := fmt.Sprintf(`
-		SELECT DISTINCT prefix FROM %s
-		WHERE mask_length = ?
+		SELECT prefix FROM %s
+		WHERE mask_length = ? AND prefix > ?
 		ORDER BY prefix
-		LIMIT ?
+		LIMIT 1
 	`, tableName)
 
-	rows, err := d.db.QueryContext(ctx, query, maskLength, n)
+	stmt, err := d.db.PrepareContext(ctx, query)
 	if err != nil {
-		return nil, fmt.Errorf("failed to list prefixes: %w", err)
+		return nil, fmt.Errorf("failed to prepare prefix query: %w", err)
 	}
-	defer func() { _ = rows.Close() }()
+	defer func() { _ = stmt.Close() }()
 
 	var prefixes []string
-	for rows.Next() {
+	previous := "" // every prefix sorts after the empty string
+	for len(prefixes) < n {
 		var prefix string
-		if err := rows.Scan(&prefix); err != nil {
-			return nil, fmt.Errorf("failed to scan prefix: %w", err)
+		err := stmt.QueryRowContext(ctx, maskLength, previous).Scan(&prefix)
+		if errors.Is(err, sql.ErrNoRows) {
+			break
+		}
+		if err != nil {
+			return nil, fmt.Errorf("failed to list prefixes: %w", err)
 		}
 		prefixes = append(prefixes, prefix)
-	}
-	if err := rows.Err(); err != nil {
-		return nil, fmt.Errorf("failed to list prefixes: %w", err)
+		previous = prefix
 	}
 
 	return prefixes, nil
@@ -1876,8 +1837,9 @@ func (d *Database) samplePrefixesByLength(
 }
 
 // newestRouteOfEachPrefix returns the most recently updated live route of each
-// prefix, one lookup on the prefix index per prefix. A prefix whose routes were
-// all withdrawn since it was picked is left out.
+// prefix, one search of the prefix and last_updated index per prefix however
+// many peers carry it. A prefix whose routes were all withdrawn since it was
+// picked is left out.
 func (d *Database) newestRouteOfEachPrefix(
 	ctx context.Context, tableName string, ipVersion int, prefixes []string) ([]LiveRoute, error) {
 	//nolint:gosec // Table name is hardcoded based on IP version

internal/database/database_test.go

@@ -14,6 +14,7 @@ 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
@@ -149,8 +150,8 @@ 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)
@@ -232,6 +233,40 @@ 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
@@ -333,7 +368,6 @@ func TestVacuumReturnsFreePages(t *testing.T) {
 		routes = append(routes, route)
 		deletions = append(deletions, LiveRouteDeletion{
 			Prefix:    route.Prefix,
-			OriginASN: asn,
 			PeerIP:    route.PeerIP,
 			IPVersion: ipVersionV4,
 		})
@@ -671,7 +705,7 @@ func TestGetRandomPrefixesByLengthSkipsWithdrawnRoutes(t *testing.T) {
 	if err := db.UpsertLiveRouteBatch([]*LiveRoute{kept, withdrawn}); err != nil {
 		t.Fatalf("UpsertLiveRouteBatch: %v", err)
 	}
-	if err := db.DeleteLiveRoute(withdrawn.Prefix, withdrawn.OriginASN, withdrawn.PeerIP); err != nil {
+	if err := db.DeleteLiveRoute(withdrawn.Prefix, withdrawn.PeerIP); err != nil {
 		t.Fatalf("DeleteLiveRoute: %v", err)
 	}
 

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
@@ -120,11 +120,12 @@ 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
-CREATE INDEX IF NOT EXISTS idx_live_routes_v4_prefix ON live_routes_v4(prefix);
+-- Finds a prefix's newest route in one search, whatever the number of peers.
+CREATE INDEX IF NOT EXISTS idx_live_routes_v4_prefix_last_updated ON live_routes_v4(prefix, last_updated);
 CREATE INDEX IF NOT EXISTS idx_live_routes_v4_mask_length ON live_routes_v4(mask_length);
 CREATE INDEX IF NOT EXISTS idx_live_routes_v4_origin_asn ON live_routes_v4(origin_asn);
 CREATE INDEX IF NOT EXISTS idx_live_routes_v4_last_updated ON live_routes_v4(last_updated);
@@ -132,7 +133,8 @@ CREATE INDEX IF NOT EXISTS idx_live_routes_v4_last_updated ON live_routes_v4(las
 CREATE INDEX IF NOT EXISTS idx_live_routes_v4_mask_prefix ON live_routes_v4(mask_length, prefix);
 
 -- Indexes for live_routes_v6 table
-CREATE INDEX IF NOT EXISTS idx_live_routes_v6_prefix ON live_routes_v6(prefix);
+-- Finds a prefix's newest route in one search, whatever the number of peers.
+CREATE INDEX IF NOT EXISTS idx_live_routes_v6_prefix_last_updated ON live_routes_v6(prefix, last_updated);
 CREATE INDEX IF NOT EXISTS idx_live_routes_v6_mask_length ON live_routes_v6(mask_length);
 CREATE INDEX IF NOT EXISTS idx_live_routes_v6_origin_asn ON live_routes_v6(origin_asn);
 CREATE INDEX IF NOT EXISTS idx_live_routes_v6_last_updated ON live_routes_v6(last_updated);

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/ashandler.go

@@ -19,6 +19,12 @@ const (
 	// asnBatchSize is the number of ASN operations to batch together
 	asnBatchSize = 30000
 
+	// asnBatchMaxSize is the most ASN operations the batch holds while the
+	// flush goroutine writes the previous one: ten batches, about 10 MiB
+	// (README "Memory"). At this size HandleMessage waits until the flush
+	// goroutine takes the batch.
+	asnBatchMaxSize = 300000
+
 	// asnBatchTimeout is the maximum time to wait before flushing a batch
 	// DO NOT reduce this timeout - larger batches are more efficient
 	asnBatchTimeout = 2 * time.Second
@@ -32,12 +38,16 @@ type ASHandler struct {
 	db     database.Store
 	logger *logger.Logger
 
-	// Batching
-	mu        sync.Mutex
-	batch     []asnOp
-	lastFlush time.Time
-	stopCh    chan struct{}
-	wg        sync.WaitGroup
+	// Batching. mu guards batch only: the flush goroutine takes the batch
+	// under mu and writes it after releasing mu, so HandleMessage waits for
+	// a database write only while batch is at asnBatchMaxSize.
+	mu          sync.Mutex
+	batch       []asnOp
+	batchTaken  sync.Cond     // signalled when the flush goroutine takes batch; its L is &mu
+	batchFullCh chan struct{} // wakes the flush goroutine when batch is full
+	lastFlush   time.Time     // used only by the flush goroutine
+	stopCh      chan struct{}
+	wg          sync.WaitGroup
 }
 
 type asnOp struct {
@@ -52,12 +62,14 @@ type asnOp struct {
 // background goroutine is terminated.
 func NewASHandler(db database.Store, logger *logger.Logger) *ASHandler {
 	h := &ASHandler{
-		db:        db,
-		logger:    logger,
-		batch:     make([]asnOp, 0, asnBatchSize),
-		lastFlush: time.Now(),
-		stopCh:    make(chan struct{}),
+		db:          db,
+		logger:      logger,
+		batch:       make([]asnOp, 0, asnBatchSize),
+		batchFullCh: make(chan struct{}, 1),
+		lastFlush:   time.Now(),
+		stopCh:      make(chan struct{}),
 	}
+	h.batchTaken.L = &h.mu
 
 	// Start the flush timer goroutine
 	h.wg.Add(1)
@@ -85,8 +97,8 @@ func (h *ASHandler) QueueCapacity() int {
 // HandleMessage processes a RIS Live BGP message by extracting all ASNs from
 // the AS path and queuing them for batch insertion into the database. The
 // origin ASN (last element in the path) and all transit ASNs are recorded
-// with their associated timestamps. The batch is automatically flushed when
-// it reaches the configured size threshold.
+// with their associated timestamps. When the batch reaches the configured size
+// threshold, the flush goroutine is woken to write it.
 func (h *ASHandler) HandleMessage(msg *ristypes.RISMessage) {
 	// Use the pre-parsed timestamp
 	timestamp := msg.ParsedTimestamp
@@ -100,6 +112,13 @@ func (h *ASHandler) HandleMessage(msg *ristypes.RISMessage) {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 
+	// While the batch is at its maximum size, wait until the flush goroutine
+	// takes it. Meanwhile the handler takes nothing off its queue, which
+	// fills, and the streamer drops messages for it.
+	for len(h.batch) >= asnBatchMaxSize {
+		h.batchTaken.Wait()
+	}
+
 	// Queue origin ASN operation
 	if originASN > 0 {
 		h.batch = append(h.batch, asnOp{
@@ -118,13 +137,19 @@ func (h *ASHandler) HandleMessage(msg *ristypes.RISMessage) {
 		}
 	}
 
-	// Check if we need to flush
+	// A full batch wakes the flush goroutine. If it is writing, the wake-up
+	// waits in batchFullCh and the batch keeps growing until the write ends
+	// or the batch reaches asnBatchMaxSize.
 	if len(h.batch) >= asnBatchSize {
-		h.flushBatchLocked()
+		select {
+		case h.batchFullCh <- struct{}{}:
+		default: // already woken
+		}
 	}
 }
 
-// flushLoop runs in a goroutine and periodically flushes batches
+// flushLoop runs in a goroutine, the only one that writes the batch: it
+// flushes when the batch is full, periodically, and once more on Stop.
 func (h *ASHandler) flushLoop() {
 	defer h.wg.Done()
 	ticker := time.NewTicker(asnBatchTimeout)
@@ -133,31 +158,47 @@ func (h *ASHandler) flushLoop() {
 	for {
 		select {
 		case <-ticker.C:
-			h.mu.Lock()
 			if time.Since(h.lastFlush) >= asnBatchTimeout {
-				h.flushBatchLocked()
+				h.flushBatch()
 			}
-			h.mu.Unlock()
+		case <-h.batchFullCh:
+			h.flushBatch()
 		case <-h.stopCh:
 			// Final flush
-			h.mu.Lock()
-			h.flushBatchLocked()
-			h.mu.Unlock()
+			h.flushBatch()
 
 			return
 		}
 	}
 }
 
-// flushBatchLocked flushes the ASN batch to the database (must be called with mutex held)
-func (h *ASHandler) flushBatchLocked() {
+// takeBatch returns the pending operations and leaves an empty batch in their
+// place for HandleMessage to fill.
+func (h *ASHandler) takeBatch() []asnOp {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+
 	if len(h.batch) == 0 {
+		return nil
+	}
+	batch := h.batch
+	h.batch = make([]asnOp, 0, asnBatchSize)
+	h.batchTaken.Broadcast()
+
+	return batch
+}
+
+// flushBatch takes the pending batch and writes it to the database, without
+// holding the mutex during the write.
+func (h *ASHandler) flushBatch() {
+	batch := h.takeBatch()
+	if len(batch) == 0 {
 		return
 	}
 
 	// Process ASNs first (deduped)
 	asnMap := make(map[int]time.Time)
-	for _, op := range h.batch {
+	for _, op := range batch {
 		if existing, ok := asnMap[op.number]; !ok || op.timestamp.After(existing) {
 			asnMap[op.number] = op.timestamp
 		}
@@ -168,8 +209,6 @@ func (h *ASHandler) flushBatchLocked() {
 		h.logger.Error("Failed to process ASN batch", "error", err, "count", len(asnMap))
 	}
 
-	// Clear batch
-	h.batch = h.batch[:0]
 	h.lastFlush = time.Now()
 }
 

internal/routewatch/batchwrite_test.go

@@ -0,0 +1,374 @@
+package routewatch
+
+import (
+	"slices"
+	"sync"
+	"testing"
+	"time"
+
+	"git.eeqj.de/sneak/routewatch/internal/database"
+	"git.eeqj.de/sneak/routewatch/internal/logger"
+	"git.eeqj.de/sneak/routewatch/internal/ristypes"
+)
+
+const (
+	// handleMessageTimeout is how long a test waits for HandleMessage to
+	// return, or for a write to start, before it fails.
+	handleMessageTimeout = 2 * time.Second
+
+	// stillWaitingAfter is how long a test watches a HandleMessage that must
+	// wait. One that does not wait returns well within it.
+	stillWaitingAfter = 100 * time.Millisecond
+)
+
+// blockingStore is a mockStore whose batch writes for the AS, peer and prefix
+// handlers wait until the test calls releaseWrites. writeStarted is closed when
+// the first of them starts. It records what each write was given.
+type blockingStore struct {
+	*mockStore
+
+	release      chan struct{}
+	releaseOnce  sync.Once
+	writeStarted chan struct{}
+	startOnce    sync.Once
+
+	mu       sync.Mutex
+	prefixes map[string]bool
+	asns     map[int]bool
+	peers    map[string]bool
+}
+
+func newBlockingStore() *blockingStore {
+	return &blockingStore{
+		mockStore:    newMockStore(),
+		release:      make(chan struct{}),
+		writeStarted: make(chan struct{}),
+		prefixes:     make(map[string]bool),
+		asns:         make(map[int]bool),
+		peers:        make(map[string]bool),
+	}
+}
+
+func (s *blockingStore) waitForRelease() {
+	s.startOnce.Do(func() { close(s.writeStarted) })
+	<-s.release
+}
+
+// releaseWrites lets the held writes, and all later ones, finish. A second call
+// does nothing.
+func (s *blockingStore) releaseWrites() {
+	s.releaseOnce.Do(func() { close(s.release) })
+}
+
+func (s *blockingStore) UpsertLiveRouteBatch(routes []*database.LiveRoute) error {
+	s.waitForRelease()
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	for _, route := range routes {
+		s.prefixes[route.Prefix] = true
+	}
+
+	return nil
+}
+
+func (s *blockingStore) GetOrCreateASNBatch(asns map[int]time.Time) error {
+	s.waitForRelease()
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	for asn := range asns {
+		s.asns[asn] = true
+	}
+
+	return nil
+}
+
+func (s *blockingStore) UpdatePeerBatch(peers map[string]database.PeerUpdate) error {
+	s.waitForRelease()
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	for peerIP := range peers {
+		s.peers[peerIP] = true
+	}
+
+	return nil
+}
+
+// returnsInTime reports whether f returns within handleMessageTimeout.
+func returnsInTime(f func()) bool {
+	done := make(chan struct{})
+	go func() {
+		f()
+		close(done)
+	}()
+
+	select {
+	case <-done:
+		return true
+	case <-time.After(handleMessageTimeout):
+		return false
+	}
+}
+
+// startBlockedWrite hands h the messages in fill, which fill its batch so that
+// the handler writes it to store, and waits until that write starts. store
+// holds it until the test calls releaseWrites.
+func startBlockedWrite(
+	t *testing.T,
+	store *blockingStore,
+	h interface{ HandleMessage(*ristypes.RISMessage) },
+	fill []*ristypes.RISMessage,
+) {
+	t.Helper()
+
+	if !returnsInTime(func() {
+		for _, msg := range fill {
+			h.HandleMessage(msg)
+		}
+	}) {
+		t.Fatal("HandleMessage did not return after filling the batch: it waits for the database write")
+	}
+
+	select {
+	case <-store.writeStarted:
+	case <-time.After(handleMessageTimeout):
+		t.Fatal("the handler did not write its full batch")
+	}
+}
+
+// handleWhileAWriteIsBlocked starts a blocked write with the messages in fill
+// and hands h the message extra while store holds the write. HandleMessage must
+// return each time. Then it lets the writes finish and stops h, which writes
+// what it holds.
+func handleWhileAWriteIsBlocked(
+	t *testing.T,
+	store *blockingStore,
+	h interface {
+		HandleMessage(*ristypes.RISMessage)
+		Stop()
+	},
+	fill []*ristypes.RISMessage,
+	extra *ristypes.RISMessage,
+) {
+	t.Helper()
+	defer h.Stop()
+	defer store.releaseWrites()
+
+	startBlockedWrite(t, store, h, fill)
+
+	if !returnsInTime(func() { h.HandleMessage(extra) }) {
+		t.Fatal("HandleMessage did not return while a database write was blocked")
+	}
+}
+
+// handleAtMaxBatchSize starts a blocked write with the messages in fill, then,
+// while store holds the write, hands h the messages in toMax, which bring its
+// next batch to its maximum size, and the message extra. HandleMessage for
+// extra must wait until the test lets the write finish, and return after. Then
+// it stops h, which writes what it holds.
+func handleAtMaxBatchSize(
+	t *testing.T,
+	store *blockingStore,
+	h interface {
+		HandleMessage(*ristypes.RISMessage)
+		Stop()
+	},
+	fill, toMax []*ristypes.RISMessage,
+	extra *ristypes.RISMessage,
+) {
+	t.Helper()
+	defer h.Stop()
+	defer store.releaseWrites()
+
+	startBlockedWrite(t, store, h, fill)
+	for _, msg := range toMax {
+		h.HandleMessage(msg)
+	}
+
+	returned := make(chan struct{})
+	go func() {
+		h.HandleMessage(extra)
+		close(returned)
+	}()
+
+	select {
+	case <-returned:
+		t.Fatal("HandleMessage returned while the batch was at its maximum size and a write was blocked")
+	case <-time.After(stillWaitingAfter):
+	}
+
+	store.releaseWrites()
+	select {
+	case <-returned:
+	case <-time.After(handleMessageTimeout):
+		t.Fatal("HandleMessage did not return after the database write finished")
+	}
+}
+
+// prefixAnnouncement returns a message in which testPeerIP announces prefixes
+// with origin AS testASNA.
+func prefixAnnouncement(prefixes []string) *ristypes.RISMessage {
+	return &ristypes.RISMessage{
+		Peer:          testPeerIP,
+		Path:          ristypes.ASPath{testASNA},
+		Announcements: []ristypes.RISAnnouncement{{NextHop: testPeerIP, Prefixes: prefixes}},
+	}
+}
+
+func TestPrefixHandlerHandlesMessagesWhileAWriteIsBlocked(t *testing.T) {
+	const (
+		fillPrefix  = "192.0.2.0/24"
+		extraPrefix = "198.51.100.0/24"
+	)
+
+	store := newBlockingStore()
+	h := NewPrefixHandler(store, logger.New())
+	handleWhileAWriteIsBlocked(t, store, h,
+		[]*ristypes.RISMessage{prefixAnnouncement(slices.Repeat([]string{fillPrefix}, prefixBatchSize))},
+		prefixAnnouncement([]string{extraPrefix}))
+
+	store.mu.Lock()
+	defer store.mu.Unlock()
+	if !store.prefixes[extraPrefix] {
+		t.Errorf("Stop did not write %s, handled while a write was blocked", extraPrefix)
+	}
+}
+
+func TestASHandlerHandlesMessagesWhileAWriteIsBlocked(t *testing.T) {
+	store := newBlockingStore()
+	h := NewASHandler(store, logger.New())
+	handleWhileAWriteIsBlocked(t, store, h,
+		[]*ristypes.RISMessage{{Path: slices.Repeat(ristypes.ASPath{testASNA}, asnBatchSize)}},
+		&ristypes.RISMessage{Path: ristypes.ASPath{testASNB}})
+
+	store.mu.Lock()
+	defer store.mu.Unlock()
+	if !store.asns[testASNB] {
+		t.Errorf("Stop did not write AS%d, handled while a write was blocked", testASNB)
+	}
+}
+
+func TestPeerHandlerHandlesMessagesWhileAWriteIsBlocked(t *testing.T) {
+	const otherPeerIP = "2001:db8:ffff::2"
+
+	store := newBlockingStore()
+	h := NewPeerHandler(store, logger.New())
+	handleWhileAWriteIsBlocked(t, store, h,
+		slices.Repeat([]*ristypes.RISMessage{{Peer: testPeerIP}}, peerBatchSize),
+		&ristypes.RISMessage{Peer: otherPeerIP})
+
+	store.mu.Lock()
+	defer store.mu.Unlock()
+	if !store.peers[otherPeerIP] {
+		t.Errorf("Stop did not write peer %s, handled while a write was blocked", otherPeerIP)
+	}
+}
+
+func TestPrefixHandlerWaitsAtTheMaximumBatchSize(t *testing.T) {
+	const (
+		fillPrefix  = "192.0.2.0/24"
+		extraPrefix = "198.51.100.0/24"
+	)
+
+	store := newBlockingStore()
+	h := NewPrefixHandler(store, logger.New())
+	handleAtMaxBatchSize(t, store, h,
+		[]*ristypes.RISMessage{prefixAnnouncement(slices.Repeat([]string{fillPrefix}, prefixBatchSize))},
+		[]*ristypes.RISMessage{prefixAnnouncement(slices.Repeat([]string{fillPrefix}, prefixBatchMaxSize))},
+		prefixAnnouncement([]string{extraPrefix}))
+
+	store.mu.Lock()
+	defer store.mu.Unlock()
+	if !store.prefixes[extraPrefix] {
+		t.Errorf("Stop did not write %s, handled after the batch reached its maximum size", extraPrefix)
+	}
+}
+
+func TestASHandlerWaitsAtTheMaximumBatchSize(t *testing.T) {
+	store := newBlockingStore()
+	h := NewASHandler(store, logger.New())
+	handleAtMaxBatchSize(t, store, h,
+		[]*ristypes.RISMessage{{Path: slices.Repeat(ristypes.ASPath{testASNA}, asnBatchSize)}},
+		[]*ristypes.RISMessage{{Path: slices.Repeat(ristypes.ASPath{testASNA}, asnBatchMaxSize)}},
+		&ristypes.RISMessage{Path: ristypes.ASPath{testASNB}})
+
+	store.mu.Lock()
+	defer store.mu.Unlock()
+	if !store.asns[testASNB] {
+		t.Errorf("Stop did not write AS%d, handled after the batch reached its maximum size", testASNB)
+	}
+}
+
+func TestPeerHandlerWaitsAtTheMaximumBatchSize(t *testing.T) {
+	const otherPeerIP = "2001:db8:ffff::2"
+
+	store := newBlockingStore()
+	h := NewPeerHandler(store, logger.New())
+	handleAtMaxBatchSize(t, store, h,
+		slices.Repeat([]*ristypes.RISMessage{{Peer: testPeerIP}}, peerBatchSize),
+		slices.Repeat([]*ristypes.RISMessage{{Peer: testPeerIP}}, peerBatchMaxSize),
+		&ristypes.RISMessage{Peer: otherPeerIP})
+
+	store.mu.Lock()
+	defer store.mu.Unlock()
+	if !store.peers[otherPeerIP] {
+		t.Errorf("Stop did not write peer %s, handled after the batch reached its maximum size", otherPeerIP)
+	}
+}
+
+// TestHandlersWakeTheFlushGoroutineWhenTheBatchIsFull hands each handler, built
+// without its flush goroutine so that nothing takes a wake-up, one update short
+// of a full batch, and a second one exactly a full batch. Only the full batch
+// leaves a wake-up in batchFullCh.
+func TestHandlersWakeTheFlushGoroutineWhenTheBatchIsFull(t *testing.T) {
+	tests := []struct {
+		name      string
+		batchSize int
+		// handle hands a new handler n updates and returns its batchFullCh.
+		handle func(n int) chan struct{}
+	}{
+		{
+			name:      "prefix",
+			batchSize: prefixBatchSize,
+			handle: func(n int) chan struct{} {
+				h := &PrefixHandler{batchFullCh: make(chan struct{}, 1)}
+				h.HandleMessage(prefixAnnouncement(slices.Repeat([]string{"192.0.2.0/24"}, n)))
+
+				return h.batchFullCh
+			},
+		},
+		{
+			name:      "AS",
+			batchSize: asnBatchSize,
+			handle: func(n int) chan struct{} {
+				h := &ASHandler{batchFullCh: make(chan struct{}, 1)}
+				// The origin AS and each AS in the path are one update each.
+				h.HandleMessage(&ristypes.RISMessage{Path: slices.Repeat(ristypes.ASPath{testASNA}, n-1)})
+
+				return h.batchFullCh
+			},
+		},
+		{
+			name:      "peer",
+			batchSize: peerBatchSize,
+			handle: func(n int) chan struct{} {
+				h := &PeerHandler{batchFullCh: make(chan struct{}, 1)}
+				for range n {
+					h.HandleMessage(&ristypes.RISMessage{Peer: testPeerIP})
+				}
+
+				return h.batchFullCh
+			},
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			if got := len(tt.handle(tt.batchSize - 1)); got != 0 {
+				t.Errorf("one update short of a full batch left %d wake-ups, want none", got)
+			}
+			if got := len(tt.handle(tt.batchSize)); got != 1 {
+				t.Errorf("a full batch left %d wake-ups, want 1", got)
+			}
+		})
+	}
+}

internal/routewatch/peerhandler.go

@@ -23,6 +23,12 @@ const (
 	// peerBatchSize is the number of peer updates to batch together
 	peerBatchSize = 10000
 
+	// peerBatchMaxSize is the most peer updates the batch holds while the
+	// flush goroutine writes the previous one: ten batches, about 10 MiB
+	// (README "Memory"). At this size HandleMessage waits until the flush
+	// goroutine takes the batch.
+	peerBatchMaxSize = 100000
+
 	// peerBatchTimeout is the maximum time to wait before flushing a batch
 	peerBatchTimeout = 2 * time.Second
 )
@@ -35,12 +41,16 @@ type PeerHandler struct {
 	db     database.Store
 	logger *logger.Logger
 
-	// Batching
-	mu        sync.Mutex
-	peerBatch []peerUpdate
-	lastFlush time.Time
-	stopCh    chan struct{}
-	wg        sync.WaitGroup
+	// Batching. mu guards peerBatch only: the flush goroutine takes the batch
+	// under mu and writes it after releasing mu, so HandleMessage waits for
+	// a database write only while peerBatch is at peerBatchMaxSize.
+	mu          sync.Mutex
+	peerBatch   []peerUpdate
+	batchTaken  sync.Cond     // signalled when the flush goroutine takes peerBatch; its L is &mu
+	batchFullCh chan struct{} // wakes the flush goroutine when peerBatch is full
+	lastFlush   time.Time     // used only by the flush goroutine
+	stopCh      chan struct{}
+	wg          sync.WaitGroup
 }
 
 type peerUpdate struct {
@@ -56,12 +66,14 @@ type peerUpdate struct {
 // should be stopped by calling Stop when it is no longer needed.
 func NewPeerHandler(db database.Store, logger *logger.Logger) *PeerHandler {
 	h := &PeerHandler{
-		db:        db,
-		logger:    logger,
-		peerBatch: make([]peerUpdate, 0, peerBatchSize),
-		lastFlush: time.Now(),
-		stopCh:    make(chan struct{}),
+		db:          db,
+		logger:      logger,
+		peerBatch:   make([]peerUpdate, 0, peerBatchSize),
+		batchFullCh: make(chan struct{}, 1),
+		lastFlush:   time.Now(),
+		stopCh:      make(chan struct{}),
 	}
+	h.batchTaken.L = &h.mu
 
 	// Start the flush timer goroutine
 	h.wg.Add(1)
@@ -87,8 +99,8 @@ func (h *PeerHandler) QueueCapacity() int {
 
 // HandleMessage processes a RIS message to track peer information.
 // It extracts the peer IP address and ASN from the message and adds
-// the update to an internal batch. When the batch reaches peerBatchSize
-// or the batch timeout expires, the batch is flushed to the database.
+// the update to an internal batch. The flush goroutine writes the batch to the
+// database when it reaches peerBatchSize or the batch timeout expires.
 func (h *PeerHandler) HandleMessage(msg *ristypes.RISMessage) {
 	// Parse peer ASN from string
 	peerASN := 0
@@ -101,6 +113,13 @@ func (h *PeerHandler) HandleMessage(msg *ristypes.RISMessage) {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 
+	// While the batch is at its maximum size, wait until the flush goroutine
+	// takes it. Meanwhile the handler takes nothing off its queue, which
+	// fills, and the streamer drops messages for it.
+	for len(h.peerBatch) >= peerBatchMaxSize {
+		h.batchTaken.Wait()
+	}
+
 	// Add to batch
 	h.peerBatch = append(h.peerBatch, peerUpdate{
 		peerIP:      msg.Peer,
@@ -109,13 +128,19 @@ func (h *PeerHandler) HandleMessage(msg *ristypes.RISMessage) {
 		timestamp:   msg.ParsedTimestamp,
 	})
 
-	// Check if we need to flush
+	// A full batch wakes the flush goroutine. If it is writing, the wake-up
+	// waits in batchFullCh and the batch keeps growing until the write ends
+	// or the batch reaches peerBatchMaxSize.
 	if len(h.peerBatch) >= peerBatchSize {
-		h.flushBatchLocked()
+		select {
+		case h.batchFullCh <- struct{}{}:
+		default: // already woken
+		}
 	}
 }
 
-// flushLoop runs in a goroutine and periodically flushes batches
+// flushLoop runs in a goroutine, the only one that writes the batch: it
+// flushes when the batch is full, periodically, and once more on Stop.
 func (h *PeerHandler) flushLoop() {
 	defer h.wg.Done()
 	ticker := time.NewTicker(peerBatchTimeout)
@@ -124,31 +149,47 @@ func (h *PeerHandler) flushLoop() {
 	for {
 		select {
 		case <-ticker.C:
-			h.mu.Lock()
 			if time.Since(h.lastFlush) >= peerBatchTimeout {
-				h.flushBatchLocked()
+				h.flushBatch()
 			}
-			h.mu.Unlock()
+		case <-h.batchFullCh:
+			h.flushBatch()
 		case <-h.stopCh:
 			// Final flush
-			h.mu.Lock()
-			h.flushBatchLocked()
-			h.mu.Unlock()
+			h.flushBatch()
 
 			return
 		}
 	}
 }
 
-// flushBatchLocked flushes the peer batch to the database (must be called with mutex held)
-func (h *PeerHandler) flushBatchLocked() {
+// takeBatch returns the pending updates and leaves an empty batch in their
+// place for HandleMessage to fill.
+func (h *PeerHandler) takeBatch() []peerUpdate {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+
 	if len(h.peerBatch) == 0 {
+		return nil
+	}
+	batch := h.peerBatch
+	h.peerBatch = make([]peerUpdate, 0, peerBatchSize)
+	h.batchTaken.Broadcast()
+
+	return batch
+}
+
+// flushBatch takes the pending batch and writes it to the database, without
+// holding the mutex during the write.
+func (h *PeerHandler) flushBatch() {
+	batch := h.takeBatch()
+	if len(batch) == 0 {
 		return
 	}
 
 	// Deduplicate by peer IP, keeping the latest update for each peer
 	peerMap := make(map[string]peerUpdate)
-	for _, update := range h.peerBatch {
+	for _, update := range batch {
 		if existing, ok := peerMap[update.peerIP]; !ok || update.timestamp.After(existing.timestamp) {
 			peerMap[update.peerIP] = update
 		}
@@ -170,8 +211,6 @@ func (h *PeerHandler) flushBatchLocked() {
 		h.logger.Error("Failed to process peer batch", "error", err, "count", len(dbPeerMap))
 	}
 
-	// Clear batch
-	h.peerBatch = h.peerBatch[:0]
 	h.lastFlush = time.Now()
 }
 

internal/routewatch/prefixhandler.go

@@ -24,6 +24,12 @@ const (
 	// prefixBatchSize is the number of prefix updates to batch together
 	prefixBatchSize = 25000
 
+	// prefixBatchMaxSize is the most prefix updates the batch holds while the
+	// flush goroutine writes the previous one: ten batches, about 60 MiB
+	// (README "Memory"). At this size HandleMessage waits until the flush
+	// goroutine takes the batch.
+	prefixBatchMaxSize = 250000
+
 	// prefixBatchTimeout is the maximum time to wait before flushing a batch
 	// DO NOT reduce this timeout - larger batches are more efficient
 	prefixBatchTimeout = 1 * time.Second
@@ -40,12 +46,16 @@ type PrefixHandler struct {
 	logger  *logger.Logger
 	metrics *metrics.Tracker
 
-	// Batching
-	mu        sync.Mutex
-	batch     []prefixUpdate
-	lastFlush time.Time
-	stopCh    chan struct{}
-	wg        sync.WaitGroup
+	// Batching. mu guards batch only: the flush goroutine takes the batch
+	// under mu and writes it after releasing mu, so HandleMessage waits for
+	// a database write only while batch is at prefixBatchMaxSize.
+	mu          sync.Mutex
+	batch       []prefixUpdate
+	batchTaken  sync.Cond     // signalled when the flush goroutine takes batch; its L is &mu
+	batchFullCh chan struct{} // wakes the flush goroutine when batch is full
+	lastFlush   time.Time     // used only by the flush goroutine
+	stopCh      chan struct{}
+	wg          sync.WaitGroup
 }
 
 type prefixUpdate struct {
@@ -60,12 +70,14 @@ type prefixUpdate struct {
 // NewPrefixHandler creates a new batched prefix tracking handler
 func NewPrefixHandler(db database.Store, logger *logger.Logger) *PrefixHandler {
 	h := &PrefixHandler{
-		db:        db,
-		logger:    logger,
-		batch:     make([]prefixUpdate, 0, prefixBatchSize),
-		lastFlush: time.Now(),
-		stopCh:    make(chan struct{}),
+		db:          db,
+		logger:      logger,
+		batch:       make([]prefixUpdate, 0, prefixBatchSize),
+		batchFullCh: make(chan struct{}, 1),
+		lastFlush:   time.Now(),
+		stopCh:      make(chan struct{}),
 	}
+	h.batchTaken.L = &h.mu
 
 	// Start the flush timer goroutine
 	h.wg.Add(1)
@@ -105,17 +117,24 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
 	h.mu.Lock()
 	defer h.mu.Unlock()
 
+	// While the batch is at its maximum size, wait until the flush goroutine
+	// takes it. Meanwhile the handler takes nothing off its queue, which
+	// fills, and the streamer drops messages for it.
+	for len(h.batch) >= prefixBatchMaxSize {
+		h.batchTaken.Wait()
+	}
+
 	// 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 {
@@ -140,13 +159,19 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
 		}
 	}
 
-	// Check if we need to flush
+	// A full batch wakes the flush goroutine. If it is writing, the wake-up
+	// waits in batchFullCh and the batch keeps growing until the write ends
+	// or the batch reaches prefixBatchMaxSize.
 	if len(h.batch) >= prefixBatchSize {
-		h.flushBatchLocked()
+		select {
+		case h.batchFullCh <- struct{}{}:
+		default: // already woken
+		}
 	}
 }
 
-// flushLoop runs in a goroutine and periodically flushes batches
+// flushLoop runs in a goroutine, the only one that writes the batch: it
+// flushes when the batch is full, periodically, and once more on Stop.
 func (h *PrefixHandler) flushLoop() {
 	defer h.wg.Done()
 	ticker := time.NewTicker(prefixBatchTimeout)
@@ -155,30 +180,46 @@ func (h *PrefixHandler) flushLoop() {
 	for {
 		select {
 		case <-ticker.C:
-			h.mu.Lock()
 			if time.Since(h.lastFlush) >= prefixBatchTimeout {
-				h.flushBatchLocked()
+				h.flushBatch()
 			}
-			h.mu.Unlock()
+		case <-h.batchFullCh:
+			h.flushBatch()
 		case <-h.stopCh:
 			// Final flush
-			h.mu.Lock()
-			h.flushBatchLocked()
-			h.mu.Unlock()
+			h.flushBatch()
 
 			return
 		}
 	}
 }
 
-// flushBatchLocked flushes the prefix batch to the database (must be called with mutex held)
-func (h *PrefixHandler) flushBatchLocked() {
+// takeBatch returns the pending updates and leaves an empty batch in their
+// place for HandleMessage to fill.
+func (h *PrefixHandler) takeBatch() []prefixUpdate {
+	h.mu.Lock()
+	defer h.mu.Unlock()
+
 	if len(h.batch) == 0 {
+		return nil
+	}
+	batch := h.batch
+	h.batch = make([]prefixUpdate, 0, prefixBatchSize)
+	h.batchTaken.Broadcast()
+
+	return batch
+}
+
+// flushBatch takes the pending batch and writes it to the database, without
+// holding the mutex during the write.
+func (h *PrefixHandler) flushBatch() {
+	batch := h.takeBatch()
+	if len(batch) == 0 {
 		return
 	}
 
 	startTime := time.Now()
-	batchSize := len(h.batch)
+	batchSize := len(batch)
 
 	// 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
@@ -186,7 +227,7 @@ func (h *PrefixHandler) flushBatchLocked() {
 	// updates within one second carry the same one.
 	type prefixAndPeer struct{ prefix, peer string }
 	lastUpdates := make(map[prefixAndPeer]prefixUpdate)
-	for _, update := range h.batch {
+	for _, update := range batch {
 		lastUpdates[prefixAndPeer{update.prefix, update.peer}] = update
 	}
 
@@ -221,7 +262,6 @@ 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,
 			})
@@ -261,8 +301,6 @@ func (h *PrefixHandler) flushBatchLocked() {
 		"duration_ms", elapsed.Milliseconds(),
 	)
 
-	// Clear batch
-	h.batch = h.batch[:0]
 	h.lastFlush = time.Now()
 }
 
@@ -425,51 +463,24 @@ 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"
@@ -30,11 +31,6 @@ func TestPrefixHandlerStoresPrefixesTheIPLookupFinds(t *testing.T) {
 
 	// Built without NewPrefixHandler's flush timer, so the test flushes itself.
 	h := &PrefixHandler{db: db, logger: logger.New()}
-	flush := func() {
-		h.mu.Lock()
-		defer h.mu.Unlock()
-		h.flushBatchLocked()
-	}
 
 	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
 	h.HandleMessage(&ristypes.RISMessage{
@@ -46,7 +42,7 @@ func TestPrefixHandlerStoresPrefixesTheIPLookupFinds(t *testing.T) {
 			Prefixes: []string{"2001:db8:1::/48", "192.0.2.0/24"},
 		}},
 	})
-	flush()
+	h.flushBatch()
 
 	for ip, want := range map[string]string{
 		"2001:db8:1::1": "2001:db8:1::/48",
@@ -66,7 +62,7 @@ func TestPrefixHandlerStoresPrefixesTheIPLookupFinds(t *testing.T) {
 		ParsedTimestamp: ts.Add(time.Minute),
 		Withdrawals:     []string{"2001:db8:1:0:0:0:0:0/48", "192.0.2.0/24"},
 	})
-	flush()
+	h.flushBatch()
 
 	for _, ip := range []string{"2001:db8:1::1", "192.0.2.1"} {
 		if info, err := db.GetIPInfo(ip); !errors.Is(err, database.ErrNoRoute) {
@@ -164,9 +160,7 @@ func TestPrefixHandlerAppliesEachPeersLastUpdateInABatch(t *testing.T) {
 				for _, msg := range batch {
 					h.HandleMessage(msg)
 				}
-				h.mu.Lock()
-				h.flushBatchLocked()
-				h.mu.Unlock()
+				h.flushBatch()
 			}
 
 			ctx := context.Background()
@@ -182,32 +176,153 @@ 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.flushBatch()
+				}
+
+				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/handlers.go

@@ -198,19 +198,24 @@ func (s *Server) handleStatusJSON() http.HandlerFunc {
 		}()
 
 		var dbStats database.Stats
+		var err error
 		select {
 		case <-ctx.Done():
+			err = ctx.Err()
+		case err = <-errChan:
+		case dbStats = <-statsChan:
+		}
+		if deadlinePassed(ctx) {
 			s.logger.Error("Database stats timeout in status.json")
-			writeJSONError(w, http.StatusRequestTimeout, "Database timeout")
+			writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
 
 			return
-		case err := <-errChan:
+		}
+		if err != nil {
 			s.logger.Error("Failed to get database stats", "error", err)
 			writeJSONError(w, http.StatusInternalServerError, err.Error())
 
 			return
-		case dbStats = <-statsChan:
-			// Success
 		}
 
 		uptime := time.Since(metrics.ConnectedSince).Truncate(time.Second).String()
@@ -231,6 +236,11 @@ func (s *Server) handleStatusJSON() http.HandlerFunc {
 		var whoisStats *WHOISStatsInfo
 		if s.asnFetcher != nil {
 			whoisStats = s.getWHOISStats(ctx)
+			if deadlinePassed(ctx) {
+				writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
+
+				return
+			}
 		}
 
 		stats := Stats{
@@ -386,7 +396,7 @@ func (s *Server) handleStats() http.HandlerFunc {
 		// Check if context is already cancelled
 		select {
 		case <-ctx.Done():
-			http.Error(w, "Request timeout", http.StatusRequestTimeout)
+			writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
 
 			return
 		default:
@@ -412,18 +422,24 @@ func (s *Server) handleStats() http.HandlerFunc {
 		}()
 
 		var dbStats database.Stats
+		var err error
 		select {
 		case <-ctx.Done():
+			err = ctx.Err()
+		case err = <-errChan:
+		case dbStats = <-statsChan:
+		}
+		if deadlinePassed(ctx) {
 			s.logger.Error("Database stats timeout")
-			// Don't write response here - timeout middleware already handles it
+			writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
+
 			return
-		case err := <-errChan:
+		}
+		if err != nil {
 			s.logger.Error("Failed to get database stats", "error", err)
 			writeJSONError(w, http.StatusInternalServerError, err.Error())
 
 			return
-		case dbStats = <-statsChan:
-			// Success
 		}
 
 		uptime := time.Since(metrics.ConnectedSince).Truncate(time.Second).String()
@@ -462,6 +478,11 @@ func (s *Server) handleStats() http.HandlerFunc {
 		var whoisStats *WHOISStatsInfo
 		if s.asnFetcher != nil {
 			whoisStats = s.getWHOISStats(ctx)
+			if deadlinePassed(ctx) {
+				writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
+
+				return
+			}
 		}
 
 		// Calculate connection duration
@@ -613,6 +634,11 @@ func (s *Server) handleIPInfo() http.HandlerFunc {
 			} else {
 				// It's a hostname - resolve it
 				resolved, err := net.DefaultResolver.LookupHost(ctx, t)
+				if deadlinePassed(ctx) {
+					writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
+
+					return
+				}
 				if err != nil {
 					response.Errors = append(response.Errors, t+": "+err.Error())
 
@@ -634,6 +660,11 @@ func (s *Server) handleIPInfo() http.HandlerFunc {
 		// Look up each IP
 		for _, ip := range ipsToLookup {
 			ipInfo, err := s.db.GetIPInfoContext(ctx, ip)
+			if deadlinePassed(ctx) {
+				writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
+
+				return
+			}
 			if err != nil {
 				response.Errors = append(response.Errors, ip+": "+err.Error())
 
@@ -642,6 +673,11 @@ func (s *Server) handleIPInfo() http.HandlerFunc {
 
 			// Do PTR lookup for this IP
 			ptrs, err := net.DefaultResolver.LookupAddr(ctx, ip)
+			if deadlinePassed(ctx) {
+				writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
+
+				return
+			}
 			if err == nil && len(ptrs) > 0 {
 				// Remove trailing dots from PTR records
 				for i, ptr := range ptrs {
@@ -702,6 +738,11 @@ func (s *Server) handleASDetailJSON() http.HandlerFunc {
 		}
 
 		asInfo, prefixes, err := s.db.GetASDetailsContext(r.Context(), asn)
+		if deadlinePassed(r.Context()) {
+			writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
+
+			return
+		}
 		if err != nil {
 			if errors.Is(err, database.ErrNoRoute) {
 				writeJSONError(w, http.StatusNotFound, err.Error())
@@ -760,6 +801,11 @@ func (s *Server) handlePrefixDetailJSON() http.HandlerFunc {
 		prefix := parsed.Masked().String()
 
 		routes, _, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
+		if deadlinePassed(r.Context()) {
+			writeJSONError(w, http.StatusGatewayTimeout, requestTimeoutMessage)
+
+			return
+		}
 		if err != nil {
 			if errors.Is(err, database.ErrNoRoute) {
 				writeJSONError(w, http.StatusNotFound, err.Error())
@@ -802,6 +848,11 @@ func (s *Server) handleASDetail() http.HandlerFunc {
 		}
 
 		asInfo, prefixes, err := s.db.GetASDetailsContext(r.Context(), asn)
+		if deadlinePassed(r.Context()) {
+			http.Error(w, requestTimeoutMessage, http.StatusGatewayTimeout)
+
+			return
+		}
 		if err != nil {
 			if errors.Is(err, database.ErrNoRoute) {
 				http.Error(w, "AS not found", http.StatusNotFound)
@@ -815,6 +866,11 @@ func (s *Server) handleASDetail() http.HandlerFunc {
 
 		// Get peers
 		peers, err := s.db.GetASPeersContext(r.Context(), asn)
+		if deadlinePassed(r.Context()) {
+			http.Error(w, requestTimeoutMessage, http.StatusGatewayTimeout)
+
+			return
+		}
 		if err != nil {
 			s.logger.Error("Failed to get AS peers", "error", err)
 			// Continue without peers rather than failing the whole request
@@ -931,6 +987,11 @@ func (s *Server) handlePrefixDetail() http.HandlerFunc {
 		prefix := parsed.Masked().String()
 
 		routes, originASNs, err := s.db.GetPrefixDetailsContext(r.Context(), prefix)
+		if deadlinePassed(r.Context()) {
+			http.Error(w, requestTimeoutMessage, http.StatusGatewayTimeout)
+
+			return
+		}
 		if err != nil {
 			if errors.Is(err, database.ErrNoRoute) {
 				http.Error(w, "Prefix not found", http.StatusNotFound)
@@ -1079,6 +1140,11 @@ func (s *Server) handlePrefixLength() http.HandlerFunc {
 		// Get random sample of prefixes
 		const maxPrefixes = 500
 		prefixes, err := s.db.GetRandomPrefixesByLengthContext(r.Context(), maskLength, ipVersion, maxPrefixes)
+		if deadlinePassed(r.Context()) {
+			http.Error(w, requestTimeoutMessage, http.StatusGatewayTimeout)
+
+			return
+		}
 		if err != nil {
 			s.logger.Error("Failed to get prefixes by length", "error", err)
 			http.Error(w, "Internal server error", http.StatusInternalServerError)
@@ -1170,6 +1236,11 @@ func (s *Server) handlePrefixLength6() http.HandlerFunc {
 		// Get random sample of prefixes
 		const maxPrefixes = 500
 		prefixes, err := s.db.GetRandomPrefixesByLengthContext(r.Context(), maskLength, ipVersion, maxPrefixes)
+		if deadlinePassed(r.Context()) {
+			http.Error(w, requestTimeoutMessage, http.StatusGatewayTimeout)
+
+			return
+		}
 		if err != nil {
 			s.logger.Error("Failed to get prefixes by length", "error", err)
 			http.Error(w, "Internal server error", http.StatusInternalServerError)

internal/server/handlers_test.go

@@ -270,6 +270,257 @@ func TestPrefixDetailShowsOriginNames(t *testing.T) {
 	}
 }
 
+// shortRequestDeadline is the deadline the tests below give a request. The
+// server's 30-second limit and the stats handlers' 4-second limit inherit it,
+// so the tests do not wait for either.
+const shortRequestDeadline = 50 * time.Millisecond
+
+// TestRequestPastTimeLimitAnswers504 sends a request through the server's
+// router to a page and to a JSON route whose handlers wait for the time limit
+// to pass and return without answering. The page answers 504; the JSON route
+// answers 504 with the JSON error body. Before the fix both answered 408,
+// written by the time limit middleware from its own goroutine while the
+// handler could still be writing.
+func TestRequestPastTimeLimitAnswers504(t *testing.T) {
+	s := New(nil, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+
+	waitForTimeLimit := func(_ http.ResponseWriter, r *http.Request) {
+		<-r.Context().Done()
+	}
+	// Registered the way setupRoutes registers pages and JSON routes.
+	s.router.Get("/slow-page", waitForTimeLimit)
+	s.router.Get("/slow.json", JSONValidationMiddleware(http.HandlerFunc(waitForTimeLimit)).ServeHTTP)
+
+	tests := []struct {
+		path         string
+		wantJSONBody bool
+	}{
+		{path: "/slow-page"},
+		{path: "/slow.json", wantJSONBody: true},
+	}
+
+	for _, tt := range tests {
+		ctx, cancel := context.WithTimeout(context.Background(), shortRequestDeadline)
+		rec := httptest.NewRecorder()
+		s.router.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, tt.path, nil).WithContext(ctx))
+		cancel()
+
+		if rec.Code != http.StatusGatewayTimeout {
+			t.Errorf("%s: status %d, want %d; body %s", tt.path, rec.Code, http.StatusGatewayTimeout, rec.Body)
+		}
+		if tt.wantJSONBody {
+			checkJSONErrorCode(t, tt.path, rec, http.StatusGatewayTimeout)
+		}
+	}
+}
+
+// TestStatsHandlersAnswer504WhenDatabaseReadRunsOut calls each stats handler
+// with a store that does not answer before the handler's 4-second limit runs
+// out, and checks that the handler itself answers 504 with the JSON error
+// body. Before the fix status.json answered 408, and /api/v1/stats wrote
+// nothing, which the JSON check on its route turned into 500.
+func TestStatsHandlersAnswer504WhenDatabaseReadRunsOut(t *testing.T) {
+	release := make(chan struct{})
+	defer close(release)
+	db := blockingStatsDB{release: release}
+	s := New(db, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+
+	handlers := map[string]http.HandlerFunc{
+		"status.json": s.handleStatusJSON(),
+		"stats":       s.handleStats(),
+	}
+
+	for name, handler := range handlers {
+		ctx, cancel := context.WithTimeout(context.Background(), shortRequestDeadline)
+		rec := httptest.NewRecorder()
+		handler(rec, httptest.NewRequest(http.MethodGet, "/", nil).WithContext(ctx))
+		cancel()
+
+		if rec.Code != http.StatusGatewayTimeout {
+			t.Errorf("%s: status %d, want %d; body %s", name, rec.Code, http.StatusGatewayTimeout, rec.Body)
+		}
+		checkJSONErrorCode(t, name, rec, http.StatusGatewayTimeout)
+	}
+}
+
+// slowDB embeds database.Store (left nil) and overrides the reads the routes
+// in TestRoutesAnswer504WhenDatabaseReadIsCutOff make. Each read waits until
+// its context ends and then answers as the real store does when the time
+// limit cuts it off: the IP read with the context's error, the prefix read
+// that there is no route, and the others with what they had read, here
+// nothing, and no error. Every other Store method is unused and would panic
+// if called.
+type slowDB struct {
+	database.Store
+}
+
+func (slowDB) GetStatsContext(ctx context.Context) (database.Stats, error) {
+	<-ctx.Done()
+
+	return database.Stats{}, nil
+}
+
+func (slowDB) GetIPInfoContext(ctx context.Context, _ string) (*database.IPInfo, error) {
+	<-ctx.Done()
+
+	return nil, ctx.Err()
+}
+
+func (slowDB) GetASDetailsContext(ctx context.Context, asn int) (*database.ASN, []database.LiveRoute, error) {
+	<-ctx.Done()
+
+	return &database.ASN{ASN: asn}, nil, nil
+}
+
+func (slowDB) GetPrefixDetailsContext(ctx context.Context, _ string) ([]database.LiveRoute, map[int]database.ASN, error) {
+	<-ctx.Done()
+
+	return nil, nil, database.ErrNoRoute
+}
+
+func (slowDB) GetRandomPrefixesByLengthContext(ctx context.Context, _, _, _ int) ([]database.LiveRoute, error) {
+	<-ctx.Done()
+
+	return nil, nil
+}
+
+// TestRoutesAnswer504WhenDatabaseReadIsCutOff sends a request through the
+// server's router for each route that reads the database, with a store whose
+// reads are cut off when the request's time limit passes. Each route answers
+// 504 with the time limit message. Without its check after the read, a route
+// answers with what the cut-off read returned: 500, 404, 200, or, on a page
+// that stops before rendering once the request has ended, 504 with an empty
+// body. The health check is left out: it answers 503 when its own 2-second
+// ping runs out.
+func TestRoutesAnswer504WhenDatabaseReadIsCutOff(t *testing.T) {
+	s := New(slowDB{}, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+
+	tests := []struct {
+		path         string
+		wantJSONBody bool
+	}{
+		{path: "/as/64500"},
+		{path: "/prefix/192.0.2.0/24"},
+		{path: "/prefixlength/24"},
+		{path: "/prefixlength6/48"},
+		{path: "/status.json", wantJSONBody: true},
+		{path: "/ip/192.0.2.1", wantJSONBody: true},
+		{path: "/api/v1/stats", wantJSONBody: true},
+		{path: "/api/v1/ip/192.0.2.1", wantJSONBody: true},
+		{path: "/api/v1/as/64500", wantJSONBody: true},
+		{path: "/api/v1/prefix/192.0.2.0/24", wantJSONBody: true},
+	}
+
+	for _, tt := range tests {
+		checkRouteAnswers504(t, s, tt.path, tt.wantJSONBody)
+	}
+}
+
+// slowSecondReadDB embeds database.Store (left nil) and overrides the reads
+// of the routes that read the database twice. The first read answers at once;
+// the second, the AS page's peers read or the stats routes' WHOIS read, waits
+// until its context ends and then answers with the context's error, as the
+// real store does.
+type slowSecondReadDB struct {
+	database.Store
+}
+
+func (slowSecondReadDB) GetASDetailsContext(_ context.Context, asn int) (*database.ASN, []database.LiveRoute, error) {
+	return &database.ASN{ASN: asn}, nil, nil
+}
+
+func (slowSecondReadDB) GetASPeersContext(ctx context.Context, _ int) ([]database.ASPeer, error) {
+	<-ctx.Done()
+
+	return nil, ctx.Err()
+}
+
+func (slowSecondReadDB) GetStatsContext(_ context.Context) (database.Stats, error) {
+	return database.Stats{}, nil
+}
+
+func (slowSecondReadDB) GetWHOISStats(ctx context.Context, _ time.Duration) (*database.WHOISStats, error) {
+	<-ctx.Done()
+
+	return nil, ctx.Err()
+}
+
+// idleASNFetcher is a WHOIS fetcher that does nothing. The stats handlers
+// read the WHOIS stats only when the server has a fetcher.
+type idleASNFetcher struct{}
+
+func (idleASNFetcher) QueueImmediate(int) {}
+
+func (idleASNFetcher) GetStats() ASNFetcherStats {
+	return ASNFetcherStats{}
+}
+
+// TestRoutesAnswer504WhenSecondDatabaseReadIsCutOff sends a request through
+// the server's router for each route that reads the database twice, with a
+// store that answers the first read and cuts off the second when the
+// request's time limit passes. Each route answers 504 with the time limit
+// message. Without the check after the second read, the AS page answers 504
+// with an empty body and the stats routes answer 200 without the WHOIS stats.
+func TestRoutesAnswer504WhenSecondDatabaseReadIsCutOff(t *testing.T) {
+	s := New(slowSecondReadDB{}, streamer.New(logger.New(), metrics.New()), logger.New(), &config.Config{})
+	s.SetASNFetcher(idleASNFetcher{})
+
+	tests := []struct {
+		path         string
+		wantJSONBody bool
+	}{
+		{path: "/as/64500"},
+		{path: "/status.json", wantJSONBody: true},
+		{path: "/api/v1/stats", wantJSONBody: true},
+	}
+
+	for _, tt := range tests {
+		checkRouteAnswers504(t, s, tt.path, tt.wantJSONBody)
+	}
+}
+
+// checkRouteAnswers504 sends a request for path through the server's router
+// with a short deadline and checks that it answers 504 with the time limit
+// message: the JSON error body on JSON routes, the plain message on pages.
+func checkRouteAnswers504(t *testing.T, s *Server, path string, wantJSONBody bool) {
+	t.Helper()
+
+	ctx, cancel := context.WithTimeout(context.Background(), shortRequestDeadline)
+	defer cancel()
+	rec := httptest.NewRecorder()
+	s.router.ServeHTTP(rec, httptest.NewRequest(http.MethodGet, path, nil).WithContext(ctx))
+
+	if rec.Code != http.StatusGatewayTimeout {
+		t.Errorf("%s: status %d, want %d; body %s", path, rec.Code, http.StatusGatewayTimeout, rec.Body)
+	}
+	if wantJSONBody {
+		checkJSONErrorCode(t, path, rec, http.StatusGatewayTimeout)
+	} else if body := strings.TrimSpace(rec.Body.String()); body != requestTimeoutMessage {
+		t.Errorf("%s: body %q, want %q", path, body, requestTimeoutMessage)
+	}
+}
+
+// checkJSONErrorCode checks that rec holds the JSON error body with the given
+// code.
+func checkJSONErrorCode(t *testing.T, name string, rec *httptest.ResponseRecorder, want int) {
+	t.Helper()
+
+	var body struct {
+		Status string `json:"status"`
+		Error  struct {
+			Code int `json:"code"`
+		} `json:"error"`
+	}
+	if err := json.Unmarshal(rec.Body.Bytes(), &body); err != nil {
+		t.Errorf("%s: decoding the answer %q: %v", name, rec.Body, err)
+
+		return
+	}
+	if body.Status != "error" || body.Error.Code != want {
+		t.Errorf("%s: answer %s, want status \"error\" and error code %d", name, rec.Body, want)
+	}
+}
+
 // settledGoroutineCount lets transient goroutines finish, then reports the
 // current count.
 func settledGoroutineCount() int {

internal/server/middleware.go

@@ -4,6 +4,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"log/slog"
 	"net/http"
@@ -109,142 +110,28 @@ func JSONResponseMiddleware(next http.Handler) http.Handler {
 	})
 }
 
-// timeoutWriter wraps ResponseWriter to prevent concurrent writes after timeout
-type timeoutWriter struct {
-	http.ResponseWriter
-	mu      sync.Mutex
-	written bool
-	header  http.Header // cached header to prevent concurrent access
-}
+// requestTimeoutMessage is the error message sent with 504 Gateway Timeout.
+const requestTimeoutMessage = "Request timeout"
 
-func (tw *timeoutWriter) Write(b []byte) (int, error) {
-	tw.mu.Lock()
-	defer tw.mu.Unlock()
-
-	if tw.written {
-		return 0, nil // Discard writes after timeout
-	}
-
-	return tw.ResponseWriter.Write(b)
-}
-
-func (tw *timeoutWriter) WriteHeader(statusCode int) {
-	tw.mu.Lock()
-	defer tw.mu.Unlock()
-
-	if tw.written {
-		return // Discard writes after timeout
-	}
-
-	tw.ResponseWriter.WriteHeader(statusCode)
-}
-
-func (tw *timeoutWriter) Header() http.Header {
-	tw.mu.Lock()
-	defer tw.mu.Unlock()
-
-	if tw.written {
-		// Return a copy to prevent modifications after timeout
-		if tw.header == nil {
-			tw.header = make(http.Header)
-		}
-
-		return tw.header
-	}
-
-	return tw.ResponseWriter.Header()
-}
-
-func (tw *timeoutWriter) markWritten() {
-	tw.mu.Lock()
-	defer tw.mu.Unlock()
-	tw.written = true
-}
-
-// TimeoutMiddleware creates an HTTP middleware that enforces a request timeout.
-// If the handler does not complete within the specified duration, the middleware
-// returns a JSON error response with HTTP status 408 (Request Timeout).
+// deadlinePassed reports whether ctx ended because its deadline passed.
 //
-// The timeout parameter specifies the maximum duration allowed for request processing.
-// The returned middleware handles panics from the wrapped handler by re-panicking
-// after cleanup, and prevents concurrent writes to the response after timeout occurs.
-func TimeoutMiddleware(timeout time.Duration, logger *slog.Logger) func(http.Handler) http.Handler {
-	return func(next http.Handler) http.Handler {
-		return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
-			startTime := time.Now()
-
-			ctx, cancel := context.WithTimeout(r.Context(), timeout)
-			defer cancel()
-
-			tw := &timeoutWriter{
-				ResponseWriter: w,
-				header:         make(http.Header),
-			}
-
-			done := make(chan struct{})
-			panicChan := make(chan interface{}, 1)
-
-			go func() {
-				defer func() {
-					if p := recover(); p != nil {
-						panicChan <- p
-					}
-				}()
-
-				next.ServeHTTP(tw, r.WithContext(ctx))
-				close(done)
-			}()
-
-			select {
-			case p := <-panicChan:
-				panic(p)
-			case <-done:
-				return
-			case <-ctx.Done():
-				tw.markWritten() // Prevent the handler from writing after timeout
-				execTime := time.Since(startTime)
-
-				// Log the timeout as a warning
-				logger.Warn("Request timeout",
-					"method", r.Method,
-					"path", r.URL.Path,
-					"duration_ms", execTime.Milliseconds(),
-					"remote_addr", r.RemoteAddr,
-				)
-
-				// Write directly to the underlying writer since we've marked tw as written
-				// This is safe because markWritten() prevents the handler from writing
-				tw.mu.Lock()
-				w.Header().Set("Content-Type", "application/json")
-				w.WriteHeader(http.StatusRequestTimeout)
-				tw.mu.Unlock()
-
-				response := map[string]interface{}{
-					"status": "error",
-					"error": map[string]interface{}{
-						"msg":  "Request timeout",
-						"code": http.StatusRequestTimeout,
-					},
-					"@meta": map[string]interface{}{
-						"time_zone":      "UTC",
-						"api_version":    1,
-						"execution_time": fmt.Sprintf("%d ms", execTime.Milliseconds()),
-					},
-				}
-
-				_ = json.NewEncoder(w).Encode(response)
-			}
-		})
-	}
+// A handler calls it right after each database read or host name lookup,
+// before it looks at what came back, error or not, and answers 504 if it
+// reports true. A read cut off by the time limit can return an error, or part
+// of its rows or none of them without one, so only this check tells a cut-off
+// read from a complete one.
+func deadlinePassed(ctx context.Context) bool {
+	return errors.Is(ctx.Err(), context.DeadlineExceeded)
 }
 
 // JSONValidationMiddleware is an HTTP middleware that validates JSON API responses.
 // It ensures that responses with Content-Type "application/json" contain valid JSON.
 //
 // If a response is not valid JSON or is empty when JSON is expected, the middleware
-// returns a properly formatted JSON error response. For timeout errors (status 408),
-// the error message will be "Request timeout". For other errors, it returns
-// "Internal server error" with status 500 if the original status was 200.
+// returns a properly formatted JSON error response. If the request's time limit
+// has passed, it answers 504 with the error message "Request timeout". For other
+// errors, it returns "Internal server error" with status 500 if the original
+// status was 200.
 //
 // Non-JSON responses are passed through unchanged.
 func JSONValidationMiddleware(next http.Handler) http.Handler {
@@ -293,11 +180,14 @@ func JSONValidationMiddleware(next http.Handler) http.Handler {
 		if statusCode == http.StatusOK {
 			statusCode = http.StatusInternalServerError
 		}
+		if deadlinePassed(r.Context()) {
+			statusCode = http.StatusGatewayTimeout
+		}
 		w.WriteHeader(statusCode)
 
 		errorMsg := "Internal server error"
-		if statusCode == http.StatusRequestTimeout {
-			errorMsg = "Request timeout"
+		if statusCode == http.StatusGatewayTimeout {
+			errorMsg = requestTimeoutMessage
 		}
 
 		response := map[string]interface{}{

internal/server/routes.go

@@ -16,9 +16,12 @@ func (s *Server) setupRoutes() {
 	r.Use(middleware.RealIP)
 	r.Use(RequestLoggerMiddleware(s.logger.Logger)) // Structured request logging
 	r.Use(middleware.Recoverer)
-	const requestTimeout = 30 * time.Second // Increased from 8s for slow queries
-	r.Use(TimeoutMiddleware(requestTimeout, s.logger.Logger))
 	r.Use(JSONResponseMiddleware)
+	// Timeout must come after JSONResponseMiddleware, which sends its answer
+	// when the handler returns and would send a 200 ahead of Timeout's 504 for
+	// a handler that wrote nothing.
+	const requestTimeout = 30 * time.Second // Increased from 8s for slow queries
+	r.Use(middleware.Timeout(requestTimeout))
 
 	// Routes
 	r.Get("/", s.handleIndex())