fix(go): singleflight tdengine child table ensures

2026-07-03 19:49:04 +08:00
parent 3a0d9bd840
commit 68b729fa7b
3 changed files with 151 additions and 42 deletions
--- a/docs/ops/go-vehicle-ingest-memory.md
+++ b/docs/ops/go-vehicle-ingest-memory.md
@@ -22,6 +22,8 @@ Go 版本车辆数据接入链路已经作为生产主链路运行在 ECS `115.2

 2026-07-03 之后，`nats-fast-writer` 的快速路径是 TDengine 批写 + Redis 批量 pipeline：一个 NATS fetch batch 先批量写 TDengine，再用 `FastUpdateBatch` 把实时 KV、在线 TTL、last_seen 和协议索引合并到一次 Redis pipeline，最后逐条 ack NATS。

+TDengine writer 的 raw/location 子表创建有进程内单飞保护：同一子表 key 并发首次写入时，只有一个 goroutine 执行 `CREATE TABLE IF NOT EXISTS` 和 tag 更新，其他 goroutine 等待结果后继续 INSERT，避免 10W 车辆启动或回放时对 TDengine 形成重复 DDL 风暴。
+
 ## 服务和端口

 ECS：`115.29.187.205`
--- a/go/vehicle-gateway/internal/history/writer.go
+++ b/go/vehicle-gateway/internal/history/writer.go
@@ -40,8 +40,45 @@ type payloadChunk struct {
 }

 type tableCache struct {
-	mu   sync.Mutex
-	seen map[string]struct{}
+	mu       sync.Mutex
+	seen     map[string]struct{}
+	inflight map[string]*tableEnsure
+}
+
+type tableEnsure struct {
+	done chan struct{}
+	err  error
+}
+
+func (c *tableCache) doOnce(key string, fn func() error) error {
+	c.mu.Lock()
+	if _, ok := c.seen[key]; ok {
+		c.mu.Unlock()
+		return nil
+	}
+	if c.inflight == nil {
+		c.inflight = map[string]*tableEnsure{}
+	}
+	if entry, ok := c.inflight[key]; ok {
+		c.mu.Unlock()
+		<-entry.done
+		return entry.err
+	}
+	entry := &tableEnsure{done: make(chan struct{})}
+	c.inflight[key] = entry
+	c.mu.Unlock()
+
+	err := fn()
+
+	c.mu.Lock()
+	if err == nil {
+		c.seen[key] = struct{}{}
+	}
+	entry.err = err
+	delete(c.inflight, key)
+	close(entry.done)
+	c.mu.Unlock()
+	return err
 }

 func NewWriter(exec Execer) *Writer {
@@ -230,53 +267,39 @@ VALUES %s`, w.qualify(table), strings.Join(rows, ","))); err != nil {

 func (w *Writer) ensureRawChild(ctx context.Context, table string, stable string, env envelope.FrameEnvelope) error {
 	key := stable + "." + table
-	w.cache.mu.Lock()
-	_, ok := w.cache.seen[key]
-	w.cache.mu.Unlock()
-	if ok {
-		return nil
-	}
-	statement := fmt.Sprintf("CREATE TABLE IF NOT EXISTS %s USING %s TAGS ('%s', '%s', '%s', '%s', '%s')",
-		w.qualify(table), w.qualify(stable),
-		quote(string(env.Protocol)),
-		quote(env.VehicleKey()),
-		quote(env.VIN),
-		quote(normalizePhoneTag(env.Phone)),
-		quote(env.DeviceID))
-	if _, err := w.exec.ExecContext(ctx, statement); err != nil {
-		return err
-	}
-	if phone := normalizePhoneTag(env.Phone); phone != "" {
-		if _, err := w.exec.ExecContext(ctx, fmt.Sprintf("ALTER TABLE %s SET TAG phone = '%s'", w.qualify(table), quote(phone))); err != nil {
+	return w.cache.doOnce(key, func() error {
+		statement := fmt.Sprintf("CREATE TABLE IF NOT EXISTS %s USING %s TAGS ('%s', '%s', '%s', '%s', '%s')",
+			w.qualify(table), w.qualify(stable),
+			quote(string(env.Protocol)),
+			quote(env.VehicleKey()),
+			quote(env.VIN),
+			quote(normalizePhoneTag(env.Phone)),
+			quote(env.DeviceID))
+		if _, err := w.exec.ExecContext(ctx, statement); err != nil {
 			return err
 		}
-	}
-	w.cache.mu.Lock()
-	w.cache.seen[key] = struct{}{}
-	w.cache.mu.Unlock()
-	return nil
+		if phone := normalizePhoneTag(env.Phone); phone != "" {
+			if _, err := w.exec.ExecContext(ctx, fmt.Sprintf("ALTER TABLE %s SET TAG phone = '%s'", w.qualify(table), quote(phone))); err != nil {
+				return err
+			}
+		}
+		return nil
+	})
 }

 func (w *Writer) ensureLocationChild(ctx context.Context, table string, env envelope.FrameEnvelope) error {
 	key := "vehicle_locations." + table
-	w.cache.mu.Lock()
-	_, ok := w.cache.seen[key]
-	w.cache.mu.Unlock()
-	if ok {
+	return w.cache.doOnce(key, func() error {
+		statement := fmt.Sprintf("CREATE TABLE IF NOT EXISTS %s USING %s TAGS ('%s', '%s')",
+			w.qualify(table),
+			w.qualify("vehicle_locations"),
+			quote(string(env.Protocol)),
+			quote(strings.TrimSpace(env.VIN)))
+		if _, err := w.exec.ExecContext(ctx, statement); err != nil {
+			return err
+		}
 		return nil
-	}
-	statement := fmt.Sprintf("CREATE TABLE IF NOT EXISTS %s USING %s TAGS ('%s', '%s')",
-		w.qualify(table),
-		w.qualify("vehicle_locations"),
-		quote(string(env.Protocol)),
-		quote(strings.TrimSpace(env.VIN)))
-	if _, err := w.exec.ExecContext(ctx, statement); err != nil {
-		return err
-	}
-	w.cache.mu.Lock()
-	w.cache.seen[key] = struct{}{}
-	w.cache.mu.Unlock()
-	return nil
+	})
 }

 func rawValues(env envelope.FrameEnvelope, rawHex string, rawText string, parsedFields string) []any {
--- a/go/vehicle-gateway/internal/history/writer_test.go
+++ b/go/vehicle-gateway/internal/history/writer_test.go
@@ -4,6 +4,7 @@ import (
 	"context"
 	"database/sql"
 	"strings"
+	"sync"
 	"testing"
 	"time"

@@ -318,6 +319,47 @@ func TestSchemaStatementsCreateRawFramePayloadChunks(t *testing.T) {
 	}
 }

+func TestWriterEnsuresSameChildTableOnceUnderConcurrency(t *testing.T) {
+	exec := newBlockingExec("CREATE TABLE IF NOT EXISTS raw_")
+	writer := NewWriter(exec)
+	first := sampleEnvelope()
+	second := sampleEnvelope()
+	second.Sequence = 2
+	second.EventID = "second"
+
+	var wg sync.WaitGroup
+	errs := make(chan error, 2)
+	wg.Add(2)
+	go func() {
+		defer wg.Done()
+		errs <- writer.AppendAll(context.Background(), first)
+	}()
+	<-exec.blocked
+	go func() {
+		defer wg.Done()
+		errs <- writer.AppendAll(context.Background(), second)
+	}()
+	time.Sleep(25 * time.Millisecond)
+	exec.release <- struct{}{}
+	wg.Wait()
+	close(errs)
+	for err := range errs {
+		if err != nil {
+			t.Fatalf("AppendAll() error = %v", err)
+		}
+	}
+
+	if got := exec.count("CREATE TABLE IF NOT EXISTS raw_"); got != 1 {
+		t.Fatalf("raw child CREATE count = %d, want 1; calls=%#v", got, exec.calls())
+	}
+	if got := exec.count("ALTER TABLE raw_"); got != 1 {
+		t.Fatalf("raw child ALTER count = %d, want 1; calls=%#v", got, exec.calls())
+	}
+	if got := exec.count("INSERT INTO raw_"); got != 2 {
+		t.Fatalf("raw insert count = %d, want 2; calls=%#v", got, exec.calls())
+	}
+}
+
 func sampleEnvelope() envelope.FrameEnvelope {
 	return envelope.FrameEnvelope{
 		Protocol:       envelope.ProtocolJT808,
@@ -392,3 +434,45 @@ func (e *recordingExec) ExecContext(_ context.Context, query string, args ...any
 	e.calls = append(e.calls, execCall{query: query, args: args})
 	return nil, nil
 }
+
+type blockingExec struct {
+	mu           sync.Mutex
+	recorded     []execCall
+	blockPattern string
+	blocked      chan struct{}
+	release      chan struct{}
+	blockOnce    sync.Once
+}
+
+func newBlockingExec(blockPattern string) *blockingExec {
+	return &blockingExec{
+		blockPattern: blockPattern,
+		blocked:      make(chan struct{}),
+		release:      make(chan struct{}),
+	}
+}
+
+func (e *blockingExec) ExecContext(_ context.Context, query string, args ...any) (sql.Result, error) {
+	e.mu.Lock()
+	e.recorded = append(e.recorded, execCall{query: query, args: args})
+	e.mu.Unlock()
+	if strings.Contains(query, e.blockPattern) {
+		e.blockOnce.Do(func() {
+			close(e.blocked)
+			<-e.release
+		})
+	}
+	return nil, nil
+}
+
+func (e *blockingExec) calls() []execCall {
+	e.mu.Lock()
+	defer e.mu.Unlock()
+	out := make([]execCall, len(e.recorded))
+	copy(out, e.recorded)
+	return out
+}
+
+func (e *blockingExec) count(pattern string) int {
+	return countSQL(e.calls(), pattern)
+}