Author SHA1 Message Date
sneak 73793e68e0 Write each handler batch from its flush goroutine, outside the mutex (closes #65)
check / check (push) Waiting to run
The AS, peer and prefix handlers wrote their batch to the database while
holding the mutex HandleMessage needs, and HandleMessage wrote a full batch
itself. For as long as a write took, the handler took nothing off its queue,
and the streamer dropped updates once the queue was half full.

Now only the flush goroutine writes: it takes the batch under the mutex,
releases it, then writes. A full batch wakes the flush goroutine instead.
Stop still writes what the handler holds.

While a write runs, the next batch has no size limit; the README memory
section says so.

Model: opus-5-5
2026-10-14 17:50:38 +00:00
13 changed files with 501 additions and 329 deletions
+4
View File
@@ -233,6 +233,10 @@ 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 has no size
limit. A database that stays slower than the feed grows memory instead of
dropping updates.
With `DEBUG=routewatch` the daemon logs a `System stats` line every 60 seconds
with the goroutine count and Go memory figures.
+5 -3
View File
@@ -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", PeerIP: "192.0.2.1", IPVersion: ipVersionV4},
{Prefix: "203.0.113.0/24", OriginASN: 64501, PeerIP: "192.0.2.1", IPVersion: ipVersionV4},
}); err != nil {
t.Fatalf("DeleteLiveRouteBatch: %v", err)
}
@@ -327,11 +327,13 @@ func announceRoutes(db *Database, batch bool, routes []*LiveRoute) error {
}
// withdrawRoutes removes routes in one DeleteLiveRouteBatch, or with one
// DeleteLiveRoute each.
// 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.
func withdrawRoutes(db *Database, batch bool, routes []*LiveRoute) error {
if !batch {
for _, route := range routes {
if err := db.DeleteLiveRoute(route.Prefix, route.PeerIP); err != nil {
if err := db.DeleteLiveRoute(route.Prefix, 0, route.PeerIP); err != nil {
return err
}
}
+79 -35
View File
@@ -228,22 +228,21 @@ func (d *Database) beginTx() (*loggingTx, error) {
}
// A live-route upsert is an UPDATE followed, only when no row matched, by an
// 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.
// 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.
const (
updateLiveRouteV4SQL = `UPDATE live_routes_v4 SET mask_length = ?, origin_asn = ?, as_path = ?,
next_hop = ?, last_updated = ? WHERE prefix = ? AND peer_ip = ?`
updateLiveRouteV4SQL = `UPDATE live_routes_v4 SET mask_length = ?, as_path = ?, next_hop = ?,
last_updated = ? WHERE prefix = ? AND origin_asn = ? 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 = ?, origin_asn = ?, as_path = ?,
next_hop = ?, last_updated = ? WHERE prefix = ? AND peer_ip = ?`
updateLiveRouteV6SQL = `UPDATE live_routes_v6 SET mask_length = ?, as_path = ?, next_hop = ?,
last_updated = ? WHERE prefix = ? AND origin_asn = ? AND peer_ip = ?`
insertLiveRouteV6SQL = `INSERT INTO live_routes_v6 (id, prefix, mask_length, origin_asn, peer_ip,
as_path, next_hop, last_updated) VALUES (?, ?, ?, ?, ?, ?, ?, ?)`
@@ -262,8 +261,8 @@ 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, route.OriginASN, pathJSON, route.NextHop, route.LastUpdated,
route.Prefix, route.PeerIP)
res, err := upd.Exec(route.MaskLength, pathJSON, route.NextHop, route.LastUpdated,
route.Prefix, route.OriginASN, route.PeerIP)
if err != nil {
return false, false, err
}
@@ -294,8 +293,8 @@ 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, route.OriginASN, pathJSON, route.NextHop, route.LastUpdated,
route.Prefix, route.PeerIP)
res, err := upd.Exec(route.MaskLength, pathJSON, route.NextHop, route.LastUpdated,
route.Prefix, route.OriginASN, route.PeerIP)
if err != nil {
return false, false, err
}
@@ -442,18 +441,32 @@ func (d *Database) DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error {
}
}()
// Prepare statements for both IPv4 and IPv6 tables
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 statement: %w", err)
}
defer func() { _ = deleteV4.Close() }()
// No longer need to separate deletions since we handle them in the loop below
deleteV6, err := tx.Prepare(`DELETE FROM live_routes_v6 WHERE prefix = ? AND peer_ip = ?`)
// 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 = ?`)
if err != nil {
return fmt.Errorf("failed to prepare IPv6 delete statement: %w", err)
return fmt.Errorf("failed to prepare IPv4 delete with origin statement: %w", err)
}
defer func() { _ = deleteV6.Close() }()
defer func() { _ = stmtV4WithOrigin.Close() }()
stmtV4WithoutOrigin, err := tx.Prepare(`DELETE FROM live_routes_v4 WHERE prefix = ? AND peer_ip = ?`)
if err != nil {
return fmt.Errorf("failed to prepare IPv4 delete without origin 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() }()
hasV4, err := tx.Prepare(prefixHasLiveRouteV4SQL)
if err != nil {
@@ -472,18 +485,38 @@ 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 {
stmt := deleteV4
if del.IPVersion != ipVersionV4 {
stmt = deleteV6
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)
}
res, err := stmt.Exec(del.Prefix, del.PeerIP)
if err != nil {
return fmt.Errorf("failed to delete route %s: %w", del.Prefix, err)
}
// The feed often withdraws a route that is not live, which deletes
// nothing, so the in-memory route counters use the rows-affected count.
// 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.
affected, err := res.RowsAffected()
if err != nil {
return fmt.Errorf("failed to count deleted route %s: %w", del.Prefix, err)
@@ -1249,8 +1282,9 @@ func (d *Database) UpsertLiveRoute(route *LiveRoute) error {
return nil
}
// DeleteLiveRoute deletes the live route the peer has for the prefix
func (d *Database) DeleteLiveRoute(prefix, peerIP string) error {
// 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 {
d.lock("DeleteLiveRoute")
defer d.unlock()
@@ -1275,12 +1309,22 @@ func (d *Database) DeleteLiveRoute(prefix, peerIP string) error {
}()
// Literal per-table queries (rather than one formatted with the table name)
// so the delete carries no dynamically built SQL.
// so the delete carries no dynamically built SQL. A delete with no origin
// ASN can remove several rows.
var res sql.Result
if isV4 {
switch {
case isV4 && originASN == 0:
res, err = tx.Exec(`DELETE FROM live_routes_v4 WHERE prefix = ? AND peer_ip = ?`, prefix, peerIP)
} else {
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:
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
+3 -37
View File
@@ -13,7 +13,6 @@ 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
@@ -123,8 +122,8 @@ func TestGetIPInfoFindsMostSpecificLiveRoute(t *testing.T) {
lookup("2001:db9::1", "")
err = db.DeleteLiveRouteBatch([]LiveRouteDeletion{
{Prefix: "10.1.2.0/24", PeerIP: testPeerIP, IPVersion: ipVersionV4},
{Prefix: "2001:db8:1:2::/64", PeerIP: testPeerIP, IPVersion: ipVersionV6},
{Prefix: "10.1.2.0/24", OriginASN: 64502, PeerIP: testPeerIP, IPVersion: ipVersionV4},
{Prefix: "2001:db8:1:2::/64", OriginASN: 64502, PeerIP: testPeerIP, IPVersion: ipVersionV6},
})
if err != nil {
t.Fatalf("DeleteLiveRouteBatch: %v", err)
@@ -206,40 +205,6 @@ 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
@@ -341,6 +306,7 @@ func TestVacuumReturnsFreePages(t *testing.T) {
routes = append(routes, route)
deletions = append(deletions, LiveRouteDeletion{
Prefix: route.Prefix,
OriginASN: asn,
PeerIP: route.PeerIP,
IPVersion: ipVersionV4,
})
+1 -1
View File
@@ -55,7 +55,7 @@ type Store interface {
// Live route operations
UpsertLiveRoute(route *LiveRoute) error
UpsertLiveRouteBatch(routes []*LiveRoute) error
DeleteLiveRoute(prefix, peerIP string) error
DeleteLiveRoute(prefix string, originASN int, peerIP string) error
DeleteLiveRouteBatch(deletions []LiveRouteDeletion) error
GetPrefixDistribution() (ipv4 []PrefixDistribution, ipv6 []PrefixDistribution, err error)
GetPrefixDistributionContext(ctx context.Context) (ipv4 []PrefixDistribution, ipv6 []PrefixDistribution, err error)
+1
View File
@@ -114,6 +114,7 @@ type IPInfo struct {
// LiveRouteDeletion represents parameters for deleting a live route
type LiveRouteDeletion struct {
Prefix string
OriginASN int
PeerIP string
IPVersion int
}
+2 -2
View File
@@ -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, peer_ip) -- a peer has at most one route for a prefix
UNIQUE(prefix, origin_asn, peer_ip)
);
-- IPv6 routing table maintained by PrefixHandler
@@ -120,7 +120,7 @@ CREATE TABLE IF NOT EXISTS live_routes_v6 (
as_path TEXT NOT NULL, -- JSON array
next_hop TEXT NOT NULL,
last_updated DATETIME NOT NULL,
UNIQUE(prefix, peer_ip) -- a peer has at most one route for a prefix
UNIQUE(prefix, origin_asn, peer_ip)
);
-- Indexes for live_routes_v4 table
+1 -1
View File
@@ -205,7 +205,7 @@ func (m *mockStore) UpsertLiveRoute(route *database.LiveRoute) error {
}
// DeleteLiveRoute mock implementation
func (m *mockStore) DeleteLiveRoute(prefix, peerIP string) error {
func (m *mockStore) DeleteLiveRoute(prefix string, originASN int, peerIP string) error {
// Simple mock - just return nil
return nil
}
+49 -27
View File
@@ -32,12 +32,15 @@ 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 never
// waits for a database write.
mu sync.Mutex
batch []asnOp
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 {
@@ -52,11 +55,12 @@ 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{}),
}
// Start the flush timer goroutine
@@ -85,8 +89,8 @@ 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
@@ -118,13 +122,18 @@ 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.
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 +142,46 @@ 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)
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 +192,6 @@ 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()
}
+192
View File
@@ -0,0 +1,192 @@
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"
)
// handleMessageTimeout is how long a test waits for HandleMessage to return,
// or for a write to start, before it fails.
const handleMessageTimeout = 2 * time.Second
// blockingStore is a mockStore whose batch writes for the AS, peer and prefix
// handlers wait until the test closes release. writeStarted is closed when the
// first of them starts. It records what each write was given.
type blockingStore struct {
*mockStore
release chan struct{}
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
}
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
}
}
// handleWhileAWriteIsBlocked hands h the messages in fill, which fill its
// batch so that the handler writes it to store, waits until the write starts,
// 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 close(store.release)
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")
}
if !returnsInTime(func() { h.HandleMessage(extra) }) {
t.Fatal("HandleMessage did not return while a database write was blocked")
}
}
func TestPrefixHandlerHandlesMessagesWhileAWriteIsBlocked(t *testing.T) {
const (
fillPrefix = "192.0.2.0/24"
extraPrefix = "198.51.100.0/24"
)
announce := func(prefixes []string) *ristypes.RISMessage {
return &ristypes.RISMessage{
Peer: testPeerIP,
Path: ristypes.ASPath{testASNA},
Announcements: []ristypes.RISAnnouncement{{NextHop: testPeerIP, Prefixes: prefixes}},
}
}
store := newBlockingStore()
h := NewPrefixHandler(store, logger.New())
handleWhileAWriteIsBlocked(t, store, h,
[]*ristypes.RISMessage{announce(slices.Repeat([]string{fillPrefix}, prefixBatchSize))},
announce([]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)
}
}
+49 -27
View File
@@ -35,12 +35,15 @@ 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 never
// waits for a database write.
mu sync.Mutex
peerBatch []peerUpdate
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 {
@@ -56,11 +59,12 @@ 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{}),
}
// Start the flush timer goroutine
@@ -87,8 +91,8 @@ 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
@@ -109,13 +113,18 @@ 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.
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 +133,46 @@ 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)
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 +194,6 @@ 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()
}
+90 -40
View File
@@ -40,12 +40,15 @@ 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 never
// waits for a database write.
mu sync.Mutex
batch []prefixUpdate
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 {
@@ -60,11 +63,12 @@ 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{}),
}
// Start the flush timer goroutine
@@ -108,14 +112,14 @@ func (h *PrefixHandler) HandleMessage(msg *ristypes.RISMessage) {
// Withdrawals are queued before announcements, because the batch keeps
// each peer's last update for a prefix: a message that lists a prefix as
// both withdrawn and announced keeps the route (RFC 4271 section 4.3).
// A withdrawal names only the prefix and peer: the message's AS path, and
// so its origin AS, belongs to the announced prefixes.
for _, prefix := range msg.Withdrawals {
h.batch = append(h.batch, prefixUpdate{
prefix: canonicalPrefix(prefix),
originASN: originASN, // Use the originASN from path if available
peer: msg.Peer,
messageType: "withdrawal",
timestamp: timestamp,
path: msg.Path,
})
// Record withdrawal in metrics
if h.metrics != nil {
@@ -140,13 +144,18 @@ 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.
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 +164,45 @@ 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)
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 +210,7 @@ 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
}
@@ -221,6 +245,7 @@ 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,
})
@@ -260,8 +285,6 @@ func (h *PrefixHandler) flushBatchLocked() {
"duration_ms", elapsed.Milliseconds(),
)
// Clear batch
h.batch = h.batch[:0]
h.lastFlush = time.Now()
}
@@ -424,24 +447,51 @@ 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) {
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,
)
// 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,
)
}
}
}
// 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) {
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,
)
// 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,
)
}
}
}
+25 -156
View File
@@ -3,7 +3,6 @@ package routewatch
import (
"context"
"errors"
"fmt"
"slices"
"testing"
"time"
@@ -31,11 +30,6 @@ 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{
@@ -47,7 +41,7 @@ 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",
@@ -67,7 +61,7 @@ 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) {
@@ -165,9 +159,7 @@ 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()
@@ -183,155 +175,32 @@ 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}}
}
assertCounts(t, db, ipv4Version, len(tt.wantPeers), wantDistribution)
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)
}
})
}
}
// TestPrefixHandlerKeepsOneRoutePerPrefixAndPeer runs updates from one peer
// through the prefix handler into a real database, flushing each message as its
// own batch, once with IPv4 prefixes and once with IPv6 prefixes. A peer has at
// most one route for a prefix: a later announcement replaces it whatever its
// origin AS, and a withdrawal removes it whatever AS the message's AS path ends
// in. The in-memory route count and prefix distribution must match the route
// table.
func TestPrefixHandlerKeepsOneRoutePerPrefixAndPeer(t *testing.T) {
families := []struct {
ipVersion int
prefix, otherPrefix string
// The mask length of both prefixes.
maskLength int
}{
{ipVersion: ipv4Version, prefix: "198.51.100.0/24", otherPrefix: "203.0.113.0/24", maskLength: 24},
{ipVersion: ipv6Version, prefix: "2001:db8:1::/48", otherPrefix: "2001:db8:2::/48", maskLength: 48},
}
ts := time.Date(2026, 1, 2, 3, 4, 5, 0, time.UTC)
announce := func(path ristypes.ASPath, announced string) *ristypes.RISMessage {
return &ristypes.RISMessage{
Peer: testPeerIP,
Path: path,
ParsedTimestamp: ts,
Announcements: []ristypes.RISAnnouncement{{NextHop: testPeerIP, Prefixes: []string{announced}}},
}
}
for _, family := range families {
prefix, otherPrefix := family.prefix, family.otherPrefix
// One message that announces otherPrefix with origin AS testASNB and
// withdraws prefix.
announceOtherAndWithdraw := announce(ristypes.ASPath{testASNA, testASNB}, otherPrefix)
announceOtherAndWithdraw.Withdrawals = []string{prefix}
tests := []struct {
name string
messages []*ristypes.RISMessage
// The origin AS of the peer's live route for each prefix. A prefix
// left out must have no live route.
wantOrigins map[string]int
}{
{
name: "the peer announces the prefix with a new origin AS",
messages: []*ristypes.RISMessage{
announce(ristypes.ASPath{testASNA}, prefix),
announce(ristypes.ASPath{testASNA, testASNB}, prefix),
},
wantOrigins: map[string]int{prefix: testASNB},
},
{
name: "one message announces a prefix with another origin AS and withdraws the prefix",
messages: []*ristypes.RISMessage{
announce(ristypes.ASPath{testASNA}, prefix),
announceOtherAndWithdraw,
},
wantOrigins: map[string]int{otherPrefix: testASNB},
},
}
for _, tt := range tests {
t.Run(fmt.Sprintf("IPv%d %s", family.ipVersion, tt.name), func(t *testing.T) {
db, err := database.New(&config.Config{StateDir: t.TempDir()}, logger.New())
if err != nil {
t.Fatalf("failed to create database: %v", err)
}
defer func() { _ = db.Close() }()
// Built without NewPrefixHandler's flush timer, so the test flushes itself.
h := &PrefixHandler{db: db, logger: logger.New()}
for _, msg := range tt.messages {
h.HandleMessage(msg)
h.mu.Lock()
h.flushBatchLocked()
h.mu.Unlock()
}
for _, p := range []string{prefix, otherPrefix} {
routes, _, err := db.GetPrefixDetailsContext(context.Background(), p)
if err != nil && !errors.Is(err, database.ErrNoRoute) {
t.Fatalf("GetPrefixDetailsContext(%s): %v", p, err)
}
var gotOrigins []int
for _, route := range routes {
gotOrigins = append(gotOrigins, route.OriginASN)
}
var wantOrigins []int
if origin, ok := tt.wantOrigins[p]; ok {
wantOrigins = []int{origin}
}
if !slices.Equal(gotOrigins, wantOrigins) {
t.Errorf("origin ASes of the live routes for %s = %v, want %v", p, gotOrigins, wantOrigins)
}
}
wantDistribution := []database.PrefixDistribution{
{MaskLength: family.maskLength, Count: len(tt.wantOrigins)},
}
assertCounts(t, db, family.ipVersion, len(tt.wantOrigins), wantDistribution)
})
}
}
}
// assertCounts checks that the live route count and prefix distribution of one
// IP version are wantRoutes and wantDistribution, both as held in memory and as
// read from the route table.
func assertCounts(t *testing.T, db *database.Database, ipVersion, wantRoutes int,
wantDistribution []database.PrefixDistribution) {
t.Helper()
ctx := context.Background()
stats, err := db.GetStatsContext(ctx)
if err != nil {
t.Fatalf("GetStatsContext: %v", err)
}
tableRoutesV4, tableRoutesV6, err := db.GetLiveRouteCountsContext(ctx)
if err != nil {
t.Fatalf("GetLiveRouteCountsContext: %v", err)
}
queryDistributionV4, queryDistributionV6, err := db.GetPrefixDistributionContext(ctx)
if err != nil {
t.Fatalf("GetPrefixDistributionContext: %v", err)
}
memoryRoutes, tableRoutes := stats.IPv4Routes, tableRoutesV4
memoryDistribution, queryDistribution := stats.IPv4PrefixDistribution, queryDistributionV4
if ipVersion == ipv6Version {
memoryRoutes, tableRoutes = stats.IPv6Routes, tableRoutesV6
memoryDistribution, queryDistribution = stats.IPv6PrefixDistribution, queryDistributionV6
}
if memoryRoutes != wantRoutes || tableRoutes != wantRoutes {
t.Errorf("IPv%d live routes: %d in memory, %d in the table, want %d",
ipVersion, memoryRoutes, tableRoutes, wantRoutes)
}
if !slices.Equal(memoryDistribution, wantDistribution) ||
!slices.Equal(queryDistribution, wantDistribution) {
t.Errorf("IPv%d prefix distribution: %v in memory, %v from the table, want %v",
ipVersion, memoryDistribution, queryDistribution, wantDistribution)
}
}