feat: add durable business inbox example and consumer crash recovery

This commit is contained in:
2026-09-17 20:49:51 +08:00
parent b17d9b1977
commit a8a8dcfe1e
27 changed files with 1601 additions and 17 deletions

View File

@@ -0,0 +1,184 @@
package inbox
import (
"context"
"encoding/binary"
"errors"
"io"
"net"
"strings"
"sync"
"sync/atomic"
"testing"
"time"
"github.com/go-sql-driver/mysql"
)
// commitLossGateway 只用于无TLS的隔离MySQL回归;捕获COMMIT的真实成功包后丢弃并断连。
// 不伪造SQL结果,也不修改数据库中的提交状态。
type commitLossGateway struct {
listener net.Listener
dropped chan struct{}
armed atomic.Bool
mu sync.Mutex
closing bool
connections map[net.Conn]struct{}
wg sync.WaitGroup
accepted chan struct{}
}
func newCommitLossGateway(t *testing.T, target string) *commitLossGateway {
t.Helper()
listener, err := net.Listen("tcp", "127.0.0.1:0")
if err != nil {
t.Fatal(err)
}
g := &commitLossGateway{listener: listener, dropped: make(chan struct{}, 1), connections: map[net.Conn]struct{}{}, accepted: make(chan struct{})}
go func() {
defer close(g.accepted)
for {
down, err := listener.Accept()
if err != nil {
return
}
g.wg.Add(1)
go func() {
defer g.wg.Done()
defer down.Close()
up, err := net.DialTimeout("tcp", target, 3*time.Second)
if err != nil {
return
}
defer up.Close()
g.mu.Lock()
if g.closing {
g.mu.Unlock()
return
}
g.connections[down] = struct{}{}
g.connections[up] = struct{}{}
g.mu.Unlock()
defer func() { g.mu.Lock(); delete(g.connections, down); delete(g.connections, up); g.mu.Unlock() }()
var committing atomic.Bool
done := make(chan struct{}, 2)
go func() {
defer func() { done <- struct{}{} }()
for {
packet, err := readMySQLPacket(down)
if err != nil {
return
}
if packet[3] == 0 && len(packet) > 4 && packet[4] == 3 && strings.EqualFold(strings.TrimSpace(string(packet[5:])), "COMMIT") {
committing.Store(true)
}
if writeMySQLPacket(up, packet) != nil {
return
}
}
}()
go func() {
defer func() { done <- struct{}{} }()
for {
packet, err := readMySQLPacket(up)
if err != nil {
return
}
if committing.Load() && packet[3] == 1 && len(packet) > 4 && packet[4] == 0 && g.armed.CompareAndSwap(true, false) {
g.dropped <- struct{}{}
return
}
if writeMySQLPacket(down, packet) != nil {
return
}
}
}()
<-done
down.Close()
up.Close()
<-done
}()
}
}()
t.Cleanup(func() {
listener.Close()
<-g.accepted
g.mu.Lock()
g.closing = true
for c := range g.connections {
c.Close()
}
g.mu.Unlock()
g.wg.Wait()
})
return g
}
func readMySQLPacket(r io.Reader) ([]byte, error) {
var header [4]byte
if _, err := io.ReadFull(r, header[:]); err != nil {
return nil, err
}
size := int(binary.LittleEndian.Uint32(header[:]) & 0x00ffffff)
packet := make([]byte, 4+size)
copy(packet, header[:])
_, err := io.ReadFull(r, packet[4:])
return packet, err
}
func writeMySQLPacket(w io.Writer, b []byte) error {
for len(b) > 0 {
n, err := w.Write(b)
if err != nil {
return err
}
if n == 0 {
return io.ErrShortWrite
}
b = b[n:]
}
return nil
}
func TestIntegrationReceiverCommitResponseLoss(t *testing.T) {
direct, dsn := integrationDB(t)
cfg, err := mysql.ParseDSN(dsn)
if err != nil || cfg.TLSConfig != "" {
t.Fatal("commit fault requires isolated plaintext MySQL")
}
gateway := newCommitLossGateway(t, cfg.Addr)
cfg.Addr = gateway.listener.Addr().String()
ctx, stop := context.WithTimeout(context.Background(), 15*time.Second)
defer stop()
proxied, err := Open(ctx, cfg.FormatDSN())
if err != nil {
t.Fatal(err)
}
defer proxied.Close()
s := testStore(t, proxied)
event := testEvent(t, s.queue)
gateway.armed.Store(true)
if _, err = s.Apply(ctx, event); !errors.Is(err, ErrCommitUnconfirmed) {
t.Fatal("lost COMMIT response acknowledged", err)
}
select {
case <-gateway.dropped:
case <-ctx.Done():
t.Fatal("real COMMIT success was not intercepted")
}
restarted, err := New(direct, s.consumer, s.queue)
if err != nil {
t.Fatal(err)
}
view, err := restarted.View(ctx, event.AggregateID)
if err != nil || view.AppliedEvents != 1 || inboxCount(t, ctx, restarted) != 1 {
t.Fatal("expected actual commit before response loss", err)
}
result, err := restarted.Apply(ctx, event)
if err != nil || !result.Duplicate || !result.Applied {
t.Fatal("unknown commit not reconciled from durable identity", err)
}
after, err := restarted.View(ctx, event.AggregateID)
if err != nil || after.AppliedEvents != 1 {
t.Fatal("unknown commit repeated business effect", err)
}
t.Log("proxy observed actual MySQL COMMIT OK and discarded it; caller returned unconfirmed, fresh store deduplicated the already committed event")
}