Files
Agent-System-Backend/internal/runmgr/multi_instance_test.go
weichen 0a7413903d
ci / vet-and-test (pull_request) Successful in 48s
Harden cross-instance HITL/cancel command paths and add disk-full rejection tests (U7)
Why: two-process HTTP acceptance (Dev tests/integration/multi-instance, added in
the matching Dev-side change) exposed three real cross-instance defects that the
U5 library-level tests bypassed via direct Manager calls, plus the U7 fault
matrix needs explicit disk-full rejection evidence.
Evidence: go test ./... green on SQLite and real PG (AGENT_PG_TEST_DSN);
new regressions TestCrossInstanceResolveDoesNotSettleCard (runmgr, PG-gated),
TestUploadRejectsCleanlyWhenObjectDirUnwritable and
TestArtifactOverwriteRejectsCleanlyWhenObjectDirUnwritable (server).
Impact: ResolveHITL no longer settles the card when the run executes on another
instance (command-bus fallback stays available); cancelRun enqueues the command
with the real conversation id and no longer reports a false local success;
object-store failures now provably leave zero DB/object residue.
Refs: collaboration storage proposal U7 (sections 3/5/7); task card
feature/u7-load-fault-release.
2026-10-12 02:28:48 +00:00

214 lines
7.5 KiB
Go
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
package runmgr
import (
"context"
"database/sql"
"errors"
"fmt"
"os"
"path/filepath"
"strings"
"testing"
"time"
"agent-system-backend/internal/store"
core "agent-system-core"
)
// 协作存储 U5:双实例语义(KTD4/KTD5)。两个独立 Manager(A/B)共享同一
// PG schema(等价两真实进程的存储视图;进程级隔离由 CI 双进程验收与部署
// 合同覆盖——本测试验证所有权/围栏/命令的库级正确性)。
// 场景:同会话争抢、跨实例取消、旧 owner 迟到结算被拒。
func pgSchemaStore(t *testing.T) *store.Store {
t.Helper()
dsn := os.Getenv("AGENT_PG_TEST_DSN")
if dsn == "" {
t.Skip("AGENT_PG_TEST_DSN 未设置:真库多实例语义由 CI 必跑 job 覆盖")
}
schema := "u5_" + strings.ToLower(store.NewID("t")) + "_" + fmt.Sprint(os.Getpid())
admin, err := sql.Open("pgx", dsn)
if err != nil {
t.Fatal(err)
}
if _, err := admin.Exec(`CREATE EXTENSION IF NOT EXISTS pg_trgm`); err != nil {
admin.Close()
t.Fatalf("pg_trgm: %v", err)
}
if _, err := admin.Exec("CREATE SCHEMA " + schema); err != nil {
admin.Close()
t.Fatalf("schema: %v", err)
}
t.Cleanup(func() {
admin.Exec("DROP SCHEMA IF EXISTS " + schema + " CASCADE")
admin.Close()
})
st, err := store.OpenPostgres(store.AppendPGRuntimeParam(dsn, "search_path", schema+",public"), filepath.Join(t.TempDir(), "art"))
if err != nil {
t.Fatal(err)
}
t.Cleanup(func() { st.Close() })
return st
}
func newOwnedManager(t *testing.T, st *store.Store, tag string) *Manager {
t.Helper()
mgr := New(st, nil, nil, nil, nil, noopRunner{}, noopRunner{}, noopRunner{}, noopRunner{})
mgr.SetInstanceTag(tag, "127.0.0.1:0")
t.Cleanup(mgr.Shutdown)
return mgr
}
// 同会话争抢:A 占用后 B 的 ClaimConversation 被 409 拒(库级租约裁决)。
func TestMultiInstanceConversationContention(t *testing.T) {
st := pgSchemaStore(t)
a := newOwnedManager(t, st, "instA")
b := newOwnedManager(t, st, "instB")
if err := a.ClaimConversation("cnv_race"); err != nil {
t.Fatalf("A claim: %v", err)
}
defer a.ReleaseConversation("cnv_race")
if err := b.ClaimConversation("cnv_race"); err == nil || !errors.Is(err, ErrConversationBusy) {
t.Fatalf("B claim should be busy: %v", err)
}
// A 释放后 B 可占。
a.ReleaseConversation("cnv_race")
if err := b.ClaimConversation("cnv_race"); err != nil {
t.Fatalf("B claim after release: %v", err)
}
}
// 旧 owner 迟到结算被拒(fencing):A 持 epoch N;租约到期 B 接管(N+1);
// A 的 SettleRunFenced 被围栏拒绝、run 不变;B 的结算成功。
func TestMultiInstanceStaleOwnerSettleFenced(t *testing.T) {
st := pgSchemaStore(t)
conv := recoveryConversation(t, st)
// A 以 1 秒租约占用并开 run。
eA, err := st.TryClaimExecution(string(conv.ID), "instA", "", 1)
if err != nil {
t.Fatal(err)
}
run := recoveryRun(t, st, conv, core.RunRunning)
time.Sleep(1100 * time.Millisecond) // 租约到期
// B 接管。
eB, err := st.TryClaimExecution(string(conv.ID), "instB", "", 60)
if err != nil || eB != eA+1 {
t.Fatalf("B takeover: epoch=%d (want %d) err=%v", eB, eA+1, err)
}
// A 迟到结算:fenced,run 不变(仍 running)。
runA := run
runA.Status = core.RunCompleted
updated, fenced, err := st.SettleRunFenced(&runA, nil, "instA", eA)
if err != nil || updated || !fenced {
t.Fatalf("A settle: updated=%v fenced=%v err=%v(应 fenced 拒写)", updated, fenced, err)
}
after, _ := st.GetRunByID(run.ID)
if after.Status != core.RunRunning {
t.Fatalf("fenced settle 不得改写 run:status=%s", after.Status)
}
// B 结算成功。
runB := run
runB.Status = core.RunCompleted
updated, fenced, err = st.SettleRunFenced(&runB, nil, "instB", eB)
if err != nil || !updated || fenced {
t.Fatalf("B settle: updated=%v fenced=%v err=%v", updated, fenced, err)
}
}
// 跨实例取消(KTD5):run 由 A 执行;B 侧投递取消命令;A 的消费循环领取并
// 取消(本地句柄路径);命令恰被消费一次。
func TestMultiInstanceCrossCancelViaCommand(t *testing.T) {
st := pgSchemaStore(t)
a := newOwnedManager(t, st, "instA")
conv := recoveryConversation(t, st)
if err := a.ClaimConversation(string(conv.ID)); err != nil {
t.Fatal(err)
}
defer a.ReleaseConversation(string(conv.ID))
run := recoveryRun(t, st, conv, core.RunRunning)
a.execMu.Lock()
a.execEpoch[string(conv.ID)], _ = st.TryClaimExecution(string(conv.ID), "instA", run.ID, 60)
a.execMu.Unlock()
// B 侧投递取消命令(等价 server 的跨实例路径)。
cmdID := store.NewID("cmd")
if err := st.EnqueueCommand(cmdID, "cancel", run.ID, string(conv.ID), 0,
CancelCommandPayload{Owner: conv.Owner, RunID: run.ID}); err != nil {
t.Fatal(err)
}
// A 的单轮消费:认领并执行(run 非本实例活跃 → 库级取消收口)。
a.consumeCommandsOnce()
state, _, err := st.CommandState(cmdID)
if err != nil || state != "consumed" {
t.Fatalf("command state=%s err=%v(应 consumed)", state, err)
}
after, _ := st.GetRunByID(run.ID)
if after.Status != core.RunCancelled {
t.Fatalf("run status=%s(应 cancelled)", after.Status)
}
// 重复消费不重复执行(命令已 consumed,单轮无动作)。
a.consumeCommandsOnce()
after2, _ := st.GetRunByID(run.ID)
if after2.Status != core.RunCancelled {
t.Fatalf("重复消费后 status=%s", after2.Status)
}
}
// 非执行实例不吃命令:命令的会话执行权在 B,A 的消费循环应跳过。
func TestCommandOnlyConsumedByOwner(t *testing.T) {
st := pgSchemaStore(t)
a := newOwnedManager(t, st, "instA")
conv := recoveryConversation(t, st)
run := recoveryRun(t, st, conv, core.RunRunning)
if _, err := st.TryClaimExecution(string(conv.ID), "instB", run.ID, 60); err != nil {
t.Fatal(err)
}
cmdID := store.NewID("cmd")
if err := st.EnqueueCommand(cmdID, "cancel", run.ID, string(conv.ID), 0, nil); err != nil {
t.Fatal(err)
}
a.consumeCommandsOnce()
state, _, _ := st.CommandState(cmdID)
if state != "pending" {
t.Fatalf("非执行实例不得消费命令:state=%s", state)
}
_ = context.Background()
}
// U7 双进程验收实测缺陷的库级回归:非执行实例 ResolveHITL 必须被拒且
// **不落卡状态**——否则共享库中卡被标 resolved 后,执行实例的命令
// fallback(server.resolveHITL 的 !card.Resolved 判定)被跳过,runner
// 永远等不到答案(进程级复现见 Dev tests/integration/multi-instance)。
func TestCrossInstanceResolveDoesNotSettleCard(t *testing.T) {
st := pgSchemaStore(t)
a := newOwnedManager(t, st, "instA")
b := newOwnedManager(t, st, "instB")
conv := recoveryConversation(t, st)
run := recoveryRun(t, st, conv, core.RunWaitingHITL)
card := core.HITLCard{
RunID: run.ID, ConversationID: conv.ID, Question: "跨实例审批",
Kind: core.HITLKindChoice,
Choices: []core.HITLChoice{
{ID: "md", Label: "Markdown"},
},
CreatedAt: time.Now().UTC().Format(time.RFC3339),
}
card.ID = store.NewID("hitl")
if err := st.SaveHITLCard(store.OwnerDefault, &card); err != nil {
t.Fatal(err)
}
// B(非执行实例):明确拒绝,卡保持未决,留给命令总线转交 A。
if _, _, err := b.ResolveHITL(store.OwnerDefault, string(card.ID), core.HITLResponse{ChoiceID: "md"}); err == nil || !strings.Contains(err.Error(), "not executing locally") {
t.Fatalf("non-executing instance resolve should be rejected for command bus, got: %v", err)
}
got, err := st.GetHITLCard(store.OwnerDefault, string(card.ID))
if err != nil {
t.Fatal(err)
}
if got.Resolved {
t.Fatal("non-executing instance resolve settled the card (would strand the executing runner)")
}
_ = a
}