|
| 1 | +// Copyright 2025 Croupier Authors |
| 2 | +// Licensed under the Apache License, Version 2.0 |
| 3 | + |
| 4 | +package croupier |
| 5 | + |
| 6 | +import ( |
| 7 | + "context" |
| 8 | + "sync/atomic" |
| 9 | + "testing" |
| 10 | + "time" |
| 11 | + |
| 12 | + sdkv1 "github.com/cuihairu/croupier/sdks/go/pkg/pb/croupier/sdk/v1" |
| 13 | + "google.golang.org/protobuf/proto" |
| 14 | + |
| 15 | + "github.com/cuihairu/croupier/sdks/go/pkg/croupier/protocol" |
| 16 | +) |
| 17 | + |
| 18 | +// drain 请求：置位状态、立即回空确认、幂等；drain 期间新 Invoke 被拒。 |
| 19 | +func TestDrainHandler_AcksIdempotentAndRejectsInvoke(t *testing.T) { |
| 20 | + handler := newTestRPCHandler(t) |
| 21 | + handler.manager.handlers["test.fn"] = func(ctx context.Context, payload []byte) ([]byte, error) { |
| 22 | + return []byte(`{}`), nil |
| 23 | + } |
| 24 | + |
| 25 | + drainReq, _ := proto.Marshal(&sdkv1.ProviderDrainRequest{SessionId: "s-1", Reason: "rolling-restart", RetryAfterMs: 1000}) |
| 26 | + respBody, err := handler.handleDrain(context.Background(), protocol.MsgProviderDrainRequest, 1, drainReq) |
| 27 | + if err != nil { |
| 28 | + t.Fatalf("handleDrain: %v", err) |
| 29 | + } |
| 30 | + if err := proto.Unmarshal(respBody, &sdkv1.ProviderDrainResponse{}); err != nil { |
| 31 | + t.Fatalf("response must be ProviderDrainResponse: %v", err) |
| 32 | + } |
| 33 | + if !handler.manager.draining.Load() { |
| 34 | + t.Fatal("draining must be set after first drain request") |
| 35 | + } |
| 36 | + |
| 37 | + // 幂等：重复 drain 不重复触发恢复 |
| 38 | + if _, err := handler.handleDrain(context.Background(), protocol.MsgProviderDrainRequest, 2, drainReq); err != nil { |
| 39 | + t.Fatalf("idempotent drain: %v", err) |
| 40 | + } |
| 41 | + |
| 42 | + // drain 期间新 Invoke 被拒：返回 provider is draining 错误 payload，handler 不执行 |
| 43 | + called := false |
| 44 | + handler.manager.handlers["test.fn"] = func(ctx context.Context, payload []byte) ([]byte, error) { |
| 45 | + called = true |
| 46 | + return []byte(`{}`), nil |
| 47 | + } |
| 48 | + invokeReq, _ := proto.Marshal(&sdkv1.InvokeRequest{FunctionId: "test.fn"}) |
| 49 | + respBody, err = handler.invoke(context.Background(), protocol.MsgInvokeRequest, 3, invokeReq) |
| 50 | + if err != nil { |
| 51 | + t.Fatalf("invoke during drain must not be transport error: %v", err) |
| 52 | + } |
| 53 | + if called { |
| 54 | + t.Fatal("handler must not run while draining") |
| 55 | + } |
| 56 | + resp := &sdkv1.InvokeResponse{} |
| 57 | + if err := proto.Unmarshal(respBody, resp); err != nil { |
| 58 | + t.Fatalf("unmarshal: %v", err) |
| 59 | + } |
| 60 | + if string(resp.Payload) != `{"error":"provider is draining"}` { |
| 61 | + t.Fatalf("unexpected draining payload: %s", resp.Payload) |
| 62 | + } |
| 63 | + |
| 64 | + // 恢复等待在途为 0 后 handleDisconnect：无 onDisconnect、Reconnect nil → 断开 |
| 65 | + if handler.manager.config.Reconnect != nil && handler.manager.config.Reconnect.Enabled { |
| 66 | + handler.manager.config.Reconnect.Enabled = false |
| 67 | + } |
| 68 | + handler.manager.drainAndRecover() |
| 69 | + if handler.manager.draining.Load() { |
| 70 | + t.Fatal("draining must clear after recovery") |
| 71 | + } |
| 72 | +} |
| 73 | + |
| 74 | +// 在途计数：invoke 期间 inflight>0，drainAndRecover 等待其清零。 |
| 75 | +func TestDrain_WaitsForInflightCalls(t *testing.T) { |
| 76 | + handler := newTestRPCHandler(t) |
| 77 | + handler.manager.config.Reconnect = nil |
| 78 | + handler.manager.handlers["slow.fn"] = func(ctx context.Context, payload []byte) ([]byte, error) { |
| 79 | + time.Sleep(200 * time.Millisecond) |
| 80 | + return []byte(`{}`), nil |
| 81 | + } |
| 82 | + |
| 83 | + done := make(chan struct{}) |
| 84 | + go func() { |
| 85 | + defer close(done) |
| 86 | + _, _ = handler.invoke(context.Background(), protocol.MsgInvokeRequest, 1, mustMarshal(t, &sdkv1.InvokeRequest{FunctionId: "slow.fn"})) |
| 87 | + }() |
| 88 | + time.Sleep(50 * time.Millisecond) // 等 invoke 进入 handler |
| 89 | + if n := handler.manager.inflightCalls.Load(); n != 1 { |
| 90 | + t.Fatalf("inflight = %d, want 1", n) |
| 91 | + } |
| 92 | + |
| 93 | + recovered := make(chan struct{}) |
| 94 | + go func() { |
| 95 | + handler.manager.draining.Store(true) |
| 96 | + handler.manager.drainAndRecover() |
| 97 | + close(recovered) |
| 98 | + }() |
| 99 | + select { |
| 100 | + case <-recovered: |
| 101 | + t.Fatal("drainAndRecover must wait for in-flight call") |
| 102 | + case <-time.After(100 * time.Millisecond): |
| 103 | + } |
| 104 | + <-done |
| 105 | + select { |
| 106 | + case <-recovered: |
| 107 | + case <-time.After(time.Second): |
| 108 | + t.Fatal("drainAndRecover did not finish after in-flight completed") |
| 109 | + } |
| 110 | +} |
| 111 | + |
| 112 | +// drain 后经 handleDisconnect 触发 onDisconnect（对齐既有重连编排入口）。 |
| 113 | +func TestDrain_FiresOnDisconnectForReconnect(t *testing.T) { |
| 114 | + handler := newTestRPCHandler(t) |
| 115 | + handler.manager.config.Reconnect = &ReconnectConfig{Enabled: true} |
| 116 | + var fired atomic.Bool |
| 117 | + handler.manager.onDisconnect = func() { fired.Store(true) } |
| 118 | + handler.manager.connected = true |
| 119 | + handler.manager.draining.Store(true) |
| 120 | + |
| 121 | + handler.manager.drainAndRecover() |
| 122 | + |
| 123 | + if !fired.Load() { |
| 124 | + t.Fatal("onDisconnect must fire so client.go reconnect loop takes over") |
| 125 | + } |
| 126 | +} |
| 127 | + |
| 128 | +func mustMarshal(t *testing.T, m proto.Message) []byte { |
| 129 | + t.Helper() |
| 130 | + b, err := proto.Marshal(m) |
| 131 | + if err != nil { |
| 132 | + t.Fatalf("marshal: %v", err) |
| 133 | + } |
| 134 | + return b |
| 135 | +} |
0 commit comments