03225e31a9
EINO_COMPOSE 默认改为开(留 =0 逃生舱回退权威 graph.go)。翻默认前补平 compose 路径三个会静默吃掉治理能力的缺口: - HITL 审批:execDSLNode 无 approval 分支 → 审批节点被当未识别跳过、下游照跑; 补 case 调 approvalNode。 - 忠实度评测:executeGraph 把 refs 硬写 nil → 评测静默失效;runComposeGraph 改签名回传 refsOf(b)。 - 终态传播:rejected/fatalErr 不传播也不阻断下游 → 任务误判 done-空;加节点 lambda 入口守卫 + branch cond 守卫 + 终态上抛 errRejected/fatalErr,并让 runComposeConversation 模型失败置 fatalErr(对齐 graph.go 的 break 语义)。 新增等价测试:审批拒绝停下游、refs 回流。go test ./... 全绿。 soak 无回归后即可物理退役 graph.go。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
138 lines
5.3 KiB
Go
138 lines
5.3 KiB
Go
package eino
|
||
|
||
import (
|
||
"context"
|
||
"encoding/json"
|
||
"errors"
|
||
"strings"
|
||
"testing"
|
||
"time"
|
||
|
||
"github.com/sundynix/sundynix-dispatcher/internal/harness"
|
||
"github.com/sundynix/sundynix-dispatcher/internal/llm"
|
||
"github.com/sundynix/sundynix-shared/contract"
|
||
)
|
||
|
||
// echoLLM 回显最后一条 user 消息内容(确定性),便于两条执行路径逐字对比。
|
||
func echoLLM() *fakeLLM {
|
||
return &fakeLLM{
|
||
ready: true,
|
||
stream: func(m []llm.ChatMessage) string {
|
||
for i := len(m) - 1; i >= 0; i-- {
|
||
if m[i].Role == "user" {
|
||
return "ANS:" + m[i].Content
|
||
}
|
||
}
|
||
return "ANS:"
|
||
},
|
||
}
|
||
}
|
||
|
||
func runBoth(t *testing.T, graph string) (interp, comp string) {
|
||
t.Helper()
|
||
task := &contract.Task{ID: "t_eq", Graph: []byte(graph)}
|
||
|
||
o1 := &Orchestrator{pool: echoLLM(), breaker: harness.NewCircuitBreaker(), sink: &fakeSink{}}
|
||
a1, _, err := o1.runGraph(context.Background(), task, &execTracer{})
|
||
if err != nil {
|
||
t.Fatalf("runGraph: %v", err)
|
||
}
|
||
o2 := &Orchestrator{pool: echoLLM(), breaker: harness.NewCircuitBreaker(), sink: &fakeSink{}}
|
||
a2, _, err := o2.runComposeGraph(context.Background(), task, &execTracer{})
|
||
if err != nil {
|
||
t.Fatalf("runComposeGraph: %v", err)
|
||
}
|
||
return a1, a2
|
||
}
|
||
|
||
// rejectWaiter 是恒拒绝的 HITL 审批替身(fail-safe 路径回归用)。
|
||
type rejectWaiter struct{ note string }
|
||
|
||
func (r *rejectWaiter) WaitApproval(context.Context, string, time.Duration) (*contract.ApprovalDecision, error) {
|
||
return &contract.ApprovalDecision{Approved: false, Note: r.note}, nil
|
||
}
|
||
|
||
// TestComposeApprovalRejectStopsDownstream 钉死 compose 路径的 HITL 审批:
|
||
// 审批拒绝须置 errRejected、产出拒绝语,且下游 agent 绝不执行(否则等于审批形同虚设)。
|
||
// 这是翻默认前最高危的回归点——execDSLNode 一度没有 approval 分支会把审批节点当未识别跳过。
|
||
func TestComposeApprovalRejectStopsDownstream(t *testing.T) {
|
||
graph := `{"version":"1","nodes":[
|
||
{"id":"in","kind":"input","config":{"text":"hi"}},
|
||
{"id":"ap","kind":"approval","config":{"title":"上线审批"}},
|
||
{"id":"a","kind":"agent","config":{"system":"机密操作"}}
|
||
],"edges":[
|
||
{"source":"in","target":"ap"},{"source":"ap","target":"a"}
|
||
]}`
|
||
o := &Orchestrator{pool: echoLLM(), breaker: harness.NewCircuitBreaker(),
|
||
sink: &fakeSink{}, approval: &rejectWaiter{note: "不批"}}
|
||
ans, refs, err := o.runComposeGraph(context.Background(),
|
||
&contract.Task{ID: "t_ap", Graph: []byte(graph)}, &execTracer{})
|
||
if !errors.Is(err, errRejected) {
|
||
t.Fatalf("审批拒绝应返回 errRejected,got %v", err)
|
||
}
|
||
if !strings.Contains(ans, "已被拒绝") {
|
||
t.Fatalf("应产出拒绝语,got %q", ans)
|
||
}
|
||
if strings.Contains(ans, "ANS:") {
|
||
t.Fatalf("审批拒绝后下游 agent 不应执行,但成稿含 agent 产出: %q", ans)
|
||
}
|
||
if refs != nil {
|
||
t.Fatalf("拒绝终态 refs 应为 nil,got %v", refs)
|
||
}
|
||
}
|
||
|
||
// TestComposeReturnsRefs 钉死 compose 路径回传检索来源——曾被硬写成 nil,导致忠实度评测静默失效。
|
||
// 与 runGraph 同图对照:两路径都应回流含检索片段的 refs(喂 grounded judge)。
|
||
func TestComposeReturnsRefs(t *testing.T) {
|
||
g := `{"nodes":[
|
||
{"id":"i","kind":"input","config":{"text":"介绍杭州西湖"}},
|
||
{"id":"r","kind":"retriever","config":{"kb":"travel"}},
|
||
{"id":"a","kind":"agent","config":{"system":"你是导游"}}
|
||
],"edges":[{"source":"i","target":"r"},{"source":"r","target":"a"}]}`
|
||
ll := &fakeLLM{ready: true, stream: func(m []llm.ChatMessage) string { return m[0].Content }}
|
||
o := newOrch(ll, kbTool(nil), &fakeSink{}, &fakeExec{})
|
||
tk := &contract.Task{ID: "t_refs", Graph: json.RawMessage(g), Meta: map[string]any{contract.MetaUserID: "u42"}}
|
||
|
||
_, refs, err := o.runComposeGraph(context.Background(), tk, &execTracer{})
|
||
if err != nil {
|
||
t.Fatalf("runComposeGraph: %v", err)
|
||
}
|
||
if len(refs) == 0 || !strings.Contains(strings.Join(refs, ""), ragSnippet) {
|
||
t.Fatalf("compose 路径应回流含检索片段的 refs,got %v", refs)
|
||
}
|
||
}
|
||
|
||
// TestComposeEquivalentLinear 多节点线性图:input→memory→agent→output,两路径成稿应逐字一致。
|
||
func TestComposeEquivalentLinear(t *testing.T) {
|
||
graph := `{"version":"1","nodes":[
|
||
{"id":"in","kind":"input","config":{"text":"什么是图编排"}},
|
||
{"id":"m","kind":"memory","config":{}},
|
||
{"id":"a","kind":"agent","config":{"system":"你是助手"}},
|
||
{"id":"out","kind":"output","config":{}}
|
||
],"edges":[
|
||
{"source":"in","target":"m"},{"source":"m","target":"a"},{"source":"a","target":"out"}
|
||
]}`
|
||
interp, comp := runBoth(t, graph)
|
||
if interp == "" || interp != comp {
|
||
t.Fatalf("线性图不等价: interp=%q compose=%q", interp, comp)
|
||
}
|
||
}
|
||
|
||
// TestComposeEquivalentBranch 分支图:input→branch→(真)A/(假)B,条件恒真应都走 A,两路径一致。
|
||
func TestComposeEquivalentBranch(t *testing.T) {
|
||
graph := `{"version":"1","nodes":[
|
||
{"id":"in","kind":"input","config":{"text":"hi"}},
|
||
{"id":"br","kind":"branch","config":{"condition":""}},
|
||
{"id":"a","kind":"agent","config":{"system":"A"}},
|
||
{"id":"b","kind":"agent","config":{"system":"B"}}
|
||
],"edges":[
|
||
{"source":"in","target":"br"},
|
||
{"source":"br","target":"a","sourceHandle":"true"},
|
||
{"source":"br","target":"b","sourceHandle":"false"}
|
||
]}`
|
||
interp, comp := runBoth(t, graph)
|
||
if interp == "" || interp != comp {
|
||
t.Fatalf("分支图不等价: interp=%q compose=%q", interp, comp)
|
||
}
|
||
}
|