test: verify slow consumer peer progress

This commit is contained in:
2026-09-20 14:59:23 +08:00
parent 6604c7f292
commit 465349cd6d
3 changed files with 97 additions and 3 deletions

View File

@@ -101,3 +101,95 @@ func TestIntegrationEventProcessAckCommitResponseLoss(t *testing.T) {
}
t.Log("ACK commit verified before response loss and SIGKILL; replacement process returned original receipt without another handler or delivery attempt")
}
func TestIntegrationEventProcessSlowConsumerPeerProgress(t *testing.T) {
a := startOrbitProcess(t)
_, _, token := integrationWorkerClient(t)
apiA := processClient(t, a, token)
b := startOrbitProcess(t)
apiB := processClient(t, b, token)
if a.nodeID == b.nodeID {
t.Fatal("server processes reused node identity")
}
ctx, stop := context.WithTimeout(context.Background(), 45*time.Second)
defer stop()
sub := integrationEventSubscription(t, ctx, apiA, 3)
_ = integrationCancelledEvent(t, ctx, apiA, sub)
started := make(chan struct{})
release := make(chan struct{})
var callsA, callsB atomic.Int64
consumerA, err := NewEventConsumer(apiA, EventConsumerOptions{SubscriptionID: sub.ID, Capacity: 1, ShutdownTimeout: time.Second, Handle: func(work context.Context, _ ResultEvent) error {
if callsA.Add(1) == 1 {
close(started)
select {
case <-release:
case <-work.Done():
return work.Err()
}
}
return nil
}})
if err != nil {
t.Fatal(err)
}
run, cancel := context.WithCancel(ctx)
defer cancel()
doneA := make(chan error, 1)
go func() { doneA <- consumerA.Run(run) }()
select {
case <-started:
case <-ctx.Done():
t.Fatal("node A slow handler did not start")
case err := <-doneA:
t.Fatal("node A consumer stopped", err)
}
consumerB, err := NewEventConsumer(apiB, EventConsumerOptions{SubscriptionID: sub.ID, Capacity: 1, ShutdownTimeout: time.Second, Handle: func(context.Context, ResultEvent) error {
callsB.Add(1)
return nil
}})
if err != nil {
t.Fatal(err)
}
doneB := make(chan error, 1)
go func() { doneB <- consumerB.Run(run) }()
awaitConsumerEvent(t, consumerB, "connected")
_ = integrationCancelledEvent(t, ctx, apiA, sub)
deadline := time.Now().Add(5 * time.Second)
for consumerB.Stats().Acknowledged < 1 {
select {
case <-ctx.Done():
t.Fatal("node B did not progress while node A was slow")
case err := <-doneB:
t.Fatal("node B consumer stopped", err)
case <-time.After(20 * time.Millisecond):
}
if time.Now().After(deadline) {
t.Fatal("node B did not acknowledge peer delivery before deadline")
}
}
if callsA.Load() != 1 || callsB.Load() != 1 || consumerA.Stats().Acknowledged != 0 {
t.Fatalf("peer progress violated slow consumer isolation: callsA=%d callsB=%d statsA=%+v statsB=%+v", callsA.Load(), callsB.Load(), consumerA.Stats(), consumerB.Stats())
}
close(release)
for consumerA.Stats().Acknowledged < 1 {
select {
case <-ctx.Done():
t.Fatal("node A did not drain its held delivery")
case err := <-doneA:
t.Fatal("node A consumer stopped", err)
case <-time.After(20 * time.Millisecond):
}
}
cancel()
if err := awaitConsumerStop(t, doneA); err != nil {
t.Fatal(err)
}
if err := awaitConsumerStop(t, doneB); err != nil {
t.Fatal(err)
}
page, err := apiB.ListEventDeliveries(ctx, sub.ID, DeliveryPageRequest{State: DeliveryAcked})
if err != nil || len(page.Items) != 2 || page.Items[0].Attempts != 1 || page.Items[1].Attempts != 1 {
t.Fatalf("peer delivery accounting failed: err=%v deliveries=%+v", err, page.Items)
}
t.Log("node B acknowledged a second delivery while node A's capacity-one handler was blocked; both nodes preserved one attempt per delivery")
}