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 a length with at most 5000 prefixes has them all read and shuffled, and
the page lists 500 of them. A longer one is sampled with random row-id
lookups on the mask length index, at most four per prefix asked for, filled
up from the shuffled prefixes if the lookups keep repeating. Each prefix is
shown once, with its newest route.

Model: opus-5-5

internal/database/database.go

@@ -8,6 +8,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"math/rand/v2"
 	"net"
 	"net/netip"
 	"os"
@@ -53,6 +54,17 @@ const (
 	sqliteHardHeapLimitBytes = 1610612736
 )
 
+// Bounds on the work GetRandomPrefixesByLength does to pick its sample.
+const (
+	// maxPrefixesToShuffle is the most prefixes of one mask length that are
+	// read in full and shuffled to pick the sample. A mask length with more is
+	// sampled with random lookups, which then seldom land on a prefix twice.
+	maxPrefixesToShuffle = 5000
+	// sampleLookupsPerPrefix is the most random lookups made for each prefix
+	// asked for.
+	sampleLookupsPerPrefix = 4
+)
+
 // Common errors
 var (
 	// ErrInvalidIP is returned when an IP address is malformed
@@ -1696,45 +1708,176 @@ func (d *Database) GetRandomPrefixesByLength(maskLength, ipVersion, limit int) (
 	return d.GetRandomPrefixesByLengthContext(context.Background(), maskLength, ipVersion, limit)
 }
 
-// GetRandomPrefixesByLengthContext returns a random sample of prefixes with context support
+// 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. However many routes there are, it
+// reads at most maxPrefixesToShuffle+1 prefixes and makes at most
+// sampleLookupsPerPrefix lookups for each prefix asked for.
 func (d *Database) GetRandomPrefixesByLengthContext(
 	ctx context.Context, maskLength, ipVersion, limit int) ([]LiveRoute, error) {
-	// Select unique prefixes with their most recent route information
 	tableName := "live_routes_v4"
 	if ipVersion == ipVersionV6 {
 		tableName = "live_routes_v6"
 	}
 
+	// Reading one prefix more than maxPrefixesToShuffle tells whether these
+	// are all the prefixes with this mask length.
+	prefixes, err := d.firstPrefixesByLength(ctx, tableName, maskLength, maxPrefixesToShuffle+1)
+	if err != nil {
+		return nil, err
+	}
+	rand.Shuffle(len(prefixes), func(i, j int) {
+		prefixes[i], prefixes[j] = prefixes[j], prefixes[i]
+	})
+	if len(prefixes) > maxPrefixesToShuffle {
+		prefixes, err = d.samplePrefixesByLength(ctx, tableName, maskLength, limit, prefixes)
+		if err != nil {
+			return nil, err
+		}
+	}
+	prefixes = prefixes[:min(limit, len(prefixes))]
+
+	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.
+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(`
-		WITH unique_prefixes AS (
+		SELECT DISTINCT prefix FROM %s
-			SELECT prefix, MAX(last_updated) as max_updated
+		WHERE mask_length = ?
-			FROM %s
+		ORDER BY prefix
-			WHERE mask_length = ?
+		LIMIT ?
-			GROUP BY prefix
+	`, tableName)
-			ORDER BY RANDOM()
-			LIMIT ?
-		)
-		SELECT lr.prefix, lr.mask_length, lr.origin_asn, lr.as_path, 
-			lr.peer_ip, lr.last_updated
-		FROM %s lr
-		INNER JOIN unique_prefixes up ON lr.prefix = up.prefix AND lr.last_updated = up.max_updated
-		WHERE lr.mask_length = ?
-	`, tableName, tableName)
 
-	rows, err := d.db.QueryContext(ctx, query, maskLength, limit, maskLength)
+	rows, err := d.db.QueryContext(ctx, query, maskLength, n)
 	if err != nil {
-		return nil, fmt.Errorf("failed to query random prefixes: %w", err)
+		return nil, fmt.Errorf("failed to list prefixes: %w", err)
 	}
-	defer func() {
+	defer func() { _ = rows.Close() }()
-		_ = rows.Close()
-	}()
 
-	var routes []LiveRoute
+	var prefixes []string
 	for rows.Next() {
-		var route LiveRoute
+		var prefix string
+		if err := rows.Scan(&prefix); err != nil {
+			return nil, fmt.Errorf("failed to scan prefix: %w", err)
+		}
+		prefixes = append(prefixes, prefix)
+	}
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("failed to list prefixes: %w", err)
+	}
+
+	return prefixes, nil
+}
+
+// samplePrefixesByLength picks limit distinct prefixes with the given mask
+// length at random, in at most sampleLookupsPerPrefix lookups for each prefix
+// asked for. ORDER BY RANDOM() would read every route with that mask length.
+// Instead each lookup draws a row id between those of the first and last route
+// with that mask length and takes the prefix of the first such route at or
+// after it. SQLite keeps the row id at the end of every index entry, so the
+// mask length index is ordered by mask length, then row id, and the lookup is
+// one search of it. A prefix with more routes, or with more rows of other mask
+// lengths just before its own, comes up more often: the page needs a varied
+// sample, not an even one. Should the lookups keep landing on prefixes already
+// picked, the rest are taken from shuffled, prefixes of that length in random
+// order.
+func (d *Database) samplePrefixesByLength(
+	ctx context.Context, tableName string, maskLength, limit int, shuffled []string) ([]string, error) {
+	//nolint:gosec // Table name is hardcoded based on IP version
+	rangeQuery := fmt.Sprintf(`
+		SELECT
+			(SELECT MIN(rowid) FROM %[1]s WHERE mask_length = ?),
+			(SELECT MAX(rowid) FROM %[1]s WHERE mask_length = ?)
+	`, tableName)
+
+	var first, last sql.NullInt64
+	if err := d.db.QueryRowContext(ctx, rangeQuery, maskLength, maskLength).Scan(&first, &last); err != nil {
+		return nil, fmt.Errorf("failed to read row id range: %w", err)
+	}
+	if !first.Valid {
+		// Every route with this mask length was withdrawn since they were listed.
+		return nil, nil
+	}
+
+	//nolint:gosec // Table name is hardcoded based on IP version
+	lookupQuery := fmt.Sprintf(`
+		SELECT prefix FROM %s
+		WHERE mask_length = ? AND rowid >= ?
+		ORDER BY rowid
+		LIMIT 1
+	`, tableName)
+
+	stmt, err := d.db.PrepareContext(ctx, lookupQuery)
+	if err != nil {
+		return nil, fmt.Errorf("failed to prepare prefix lookup: %w", err)
+	}
+	defer func() { _ = stmt.Close() }()
+
+	picked := make(map[string]bool, limit)
+	prefixes := make([]string, 0, limit)
+	pick := func(prefix string) {
+		if !picked[prefix] {
+			picked[prefix] = true
+			prefixes = append(prefixes, prefix)
+		}
+	}
+
+	for lookups := 0; lookups < limit*sampleLookupsPerPrefix && len(prefixes) < limit; lookups++ {
+		//nolint:gosec // The sample needs variety, not unpredictability.
+		rowID := first.Int64 + rand.Int64N(last.Int64-first.Int64+1)
+
+		var prefix string
+		err := stmt.QueryRowContext(ctx, maskLength, rowID).Scan(&prefix)
+		if errors.Is(err, sql.ErrNoRows) {
+			// The routes from rowID on were withdrawn since the range was read.
+			continue
+		}
+		if err != nil {
+			return nil, fmt.Errorf("failed to look up prefix: %w", err)
+		}
+		pick(prefix)
+	}
+
+	for _, prefix := range shuffled {
+		if len(prefixes) == limit {
+			break
+		}
+		pick(prefix)
+	}
+
+	return prefixes, nil
+}
+
+// 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.
+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
+	query := fmt.Sprintf(`
+		SELECT prefix, mask_length, origin_asn, as_path, peer_ip, last_updated
+		FROM %s
+		WHERE prefix = ?
+		ORDER BY last_updated DESC
+		LIMIT 1
+	`, tableName)
+
+	stmt, err := d.db.PrepareContext(ctx, query)
+	if err != nil {
+		return nil, fmt.Errorf("failed to prepare route query: %w", err)
+	}
+	defer func() { _ = stmt.Close() }()
+
+	routes := make([]LiveRoute, 0, len(prefixes))
+	for _, prefix := range prefixes {
+		route := LiveRoute{IPVersion: ipVersion}
 		var pathJSON string
-		err := rows.Scan(
+		err := stmt.QueryRowContext(ctx, prefix).Scan(
 			&route.Prefix,
 			&route.MaskLength,
 			&route.OriginASN,
@@ -1742,17 +1885,16 @@ func (d *Database) GetRandomPrefixesByLengthContext(
 			&route.PeerIP,
 			&route.LastUpdated,
 		)
-		// Set IP version based on which table we queried
+		if errors.Is(err, sql.ErrNoRows) {
-		route.IPVersion = ipVersion
-		if err != nil {
 			continue
 		}
+		if err != nil {
+			return nil, fmt.Errorf("failed to query route of %s: %w", prefix, err)
+		}
 
-		// Decode AS path
 		if err := json.Unmarshal([]byte(pathJSON), &route.ASPath); err != nil {
 			route.ASPath = []int{}
 		}
-
 		routes = append(routes, route)
 	}
 

internal/database/database_test.go

@@ -6,6 +6,7 @@ import (
 	"errors"
 	"fmt"
 	"net/netip"
+	"slices"
 	"sync"
 	"testing"
 	"time"
@@ -45,6 +46,19 @@ const (
 	asnSecondBand = 100
 )
 
+// Parameters for the prefix-length sample tests.
+const (
+	// sampleLimit is how many prefixes the tests ask the sample for.
+	sampleLimit = 10
+	// sampledPrefixCount is how many prefixes the sampling test stores, more
+	// than sampleLimit so the sample has to choose among them.
+	sampledPrefixCount = 30
+	// secondV4PeerIP and secondV6PeerIP are the second peer each prefix is
+	// learned from, besides the one mkV4Route and mkV6Route set.
+	secondV4PeerIP = "192.0.2.2"
+	secondV6PeerIP = "2001:db8::2"
+)
+
 // TestGetIPInfoFindsMostSpecificLiveRoute stores nested live prefixes for both
 // families and checks that a lookup returns the most specific one covering the
 // address, ErrNoRoute when none covers it, and the next less specific prefix
@@ -317,3 +331,141 @@ func TestBatchWriteDuringCheckpoint(t *testing.T) {
 	cancel()
 	wg.Wait()
 }
+
+// TestGetRandomPrefixesByLengthListsPrefixOnce stores one prefix of each
+// family learned from two peers in the same second and checks that the
+// prefix-length sample lists each of them once.
+func TestGetRandomPrefixesByLengthListsPrefixOnce(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	v4First := mkV4Route(t, "198.51.100.0/24", 64500, ts)
+	v4Second := mkV4Route(t, "198.51.100.0/24", 64500, ts)
+	v4Second.PeerIP = secondV4PeerIP
+	v6First := mkV6Route("2001:db8::/32", 64500, ts)
+	v6Second := mkV6Route("2001:db8::/32", 64500, ts)
+	v6Second.PeerIP = secondV6PeerIP
+	if err := db.UpsertLiveRouteBatch([]*LiveRoute{v4First, v4Second, v6First, v6Second}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	for _, want := range []*LiveRoute{v4First, v6First} {
+		routes, err := db.GetRandomPrefixesByLength(want.MaskLength, want.IPVersion, sampleLimit)
+		if err != nil {
+			t.Fatalf("GetRandomPrefixesByLength(%d, %d): %v", want.MaskLength, want.IPVersion, err)
+		}
+		if len(routes) != 1 || routes[0].Prefix != want.Prefix {
+			t.Errorf("GetRandomPrefixesByLength(%d, %d) = %+v, want %s listed once",
+				want.MaskLength, want.IPVersion, routes, want.Prefix)
+		}
+	}
+}
+
+// TestGetRandomPrefixesByLengthListsNewestRoute stores a prefix learned from
+// two peers at different times, the older route first, and checks that the
+// prefix-length sample lists the newer route.
+func TestGetRandomPrefixesByLengthListsNewestRoute(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	older := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	olderRoute := mkV4Route(t, "198.51.100.0/24", 64500, older)
+	newerRoute := mkV4Route(t, "198.51.100.0/24", 64500, older.Add(time.Minute))
+	newerRoute.PeerIP = secondV4PeerIP
+	if err := db.UpsertLiveRouteBatch([]*LiveRoute{olderRoute, newerRoute}); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+
+	routes, err := db.GetRandomPrefixesByLength(newerRoute.MaskLength, ipVersionV4, sampleLimit)
+	if err != nil {
+		t.Fatalf("GetRandomPrefixesByLength: %v", err)
+	}
+	if len(routes) != 1 || routes[0].PeerIP != newerRoute.PeerIP {
+		t.Errorf("GetRandomPrefixesByLength = %+v, want only the newer route, from %s",
+			routes, newerRoute.PeerIP)
+	}
+}
+
+// TestGetRandomPrefixesByLengthSamplesManyPrefixes stores more /24 prefixes
+// than the sample asks for, each followed by a /16 so that routes of another
+// mask length sit between the /24 rows, and each prefix learned from two peers
+// in the same second. It checks both the sample the pages get, picked from the
+// shuffled prefixes, and the random lookups used for a mask length with more
+// than maxPrefixesToShuffle prefixes.
+func TestGetRandomPrefixesByLengthSamplesManyPrefixes(t *testing.T) {
+	cfg := &config.Config{StateDir: t.TempDir()}
+
+	db, err := New(cfg, logger.New())
+	if err != nil {
+		t.Fatalf("failed to create database: %v", err)
+	}
+	defer func() { _ = db.Close() }()
+
+	ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
+	var routes []*LiveRoute
+	var stored []string
+	for i := range sampledPrefixCount {
+		prefix := fmt.Sprintf("10.0.%d.0/24", i)
+		for _, p := range []string{prefix, fmt.Sprintf("172.%d.0.0/16", i)} {
+			first := mkV4Route(t, p, 64500, ts)
+			second := mkV4Route(t, p, 64500, ts)
+			second.PeerIP = secondV4PeerIP
+			routes = append(routes, first, second)
+		}
+		stored = append(stored, prefix)
+	}
+	if err := db.UpsertLiveRouteBatch(routes); err != nil {
+		t.Fatalf("UpsertLiveRouteBatch: %v", err)
+	}
+	slices.Sort(stored)
+
+	maskLength := routes[0].MaskLength
+	sample, err := db.GetRandomPrefixesByLength(maskLength, ipVersionV4, sampleLimit)
+	if err != nil {
+		t.Fatalf("GetRandomPrefixesByLength: %v", err)
+	}
+	var listed []string
+	for _, route := range sample {
+		listed = append(listed, route.Prefix)
+	}
+	checkPrefixSample(t, "GetRandomPrefixesByLength", listed, stored)
+
+	looked, err := db.samplePrefixesByLength(
+		context.Background(), "live_routes_v4", maskLength, sampleLimit, nil)
+	if err != nil {
+		t.Fatalf("samplePrefixesByLength: %v", err)
+	}
+	checkPrefixSample(t, "samplePrefixesByLength", looked, stored)
+}
+
+// checkPrefixSample checks that sample lists sampleLimit prefixes, each once
+// and each one of stored, and that it is not just the first sampleLimit of
+// stored, which is in text order.
+func checkPrefixSample(t *testing.T, name string, sample, stored []string) {
+	t.Helper()
+
+	if len(sample) != sampleLimit {
+		t.Errorf("%s listed %d prefixes, want %d", name, len(sample), sampleLimit)
+	}
+	listed := make(map[string]bool)
+	for _, prefix := range sample {
+		if listed[prefix] || !slices.Contains(stored, prefix) {
+			t.Errorf("%s lists %s twice or it is not one of the /24s", name, prefix)
+		}
+		listed[prefix] = true
+	}
+	if slices.Equal(slices.Sorted(slices.Values(sample)), stored[:sampleLimit]) {
+		t.Errorf("%s lists just the first %d prefixes in text order", name, sampleLimit)
+	}
+}

internal/metrics/metrics.go

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

internal/metrics/metrics_test.go

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