Files
ARTEX-chixuweihu/server/proxy_autoresume_test.go
T

501 lines
18 KiB
Go

package server
import (
"context"
"errors"
"sync"
"testing"
"time"
"github.com/Autumn-27/artex/agent"
)
func proxyTestManager(t *testing.T) *Manager {
t.Helper()
m, err := NewManager(t.TempDir(), "")
if err != nil {
t.Skipf("postgres unavailable (%v) - skipping", err)
}
t.Cleanup(func() { _ = m.Close() })
return m
}
func TestProxyFailureClassification(t *testing.T) {
for _, tc := range []struct {
name string
err error
want bool
}{
{"proxy connect", errors.New("proxyconnect tcp: connection refused"), true},
{"reset", errors.New("dial tcp: connection reset by peer"), false},
{"eof", errors.New("unexpected EOF"), false},
{"quota", errors.New("status 429: insufficient_quota"), false},
{"auth", errors.New("status 401: invalid api key"), false},
} {
t.Run(tc.name, func(t *testing.T) {
if got := isProxyFailure(tc.err); got != tc.want {
t.Fatalf("isProxyFailure(%v)=%v, want %v", tc.err, got, tc.want)
}
})
}
}
func TestGlobalProxyTrafficFailureClassification(t *testing.T) {
if !isGlobalProxyDialFailure(errors.New("dial tcp 127.0.0.1:8080: connect: connection refused")) {
t.Fatal("upstream dial refusal should be eligible for probe confirmation")
}
if isGlobalProxyDialFailure(errors.New("status 404: target not found")) {
t.Fatal("ordinary target failure must not be treated as proxy failure")
}
}
func TestProxyPauseCauseIsDistinct(t *testing.T) {
ctx, cancel := context.WithCancelCause(context.Background())
cancel(agent.AbortPausedByProxy)
code, short, _, ok := agent.AbortReason(ctx)
if !ok || code != "paused_by_proxy" || short == "" {
t.Fatalf("proxy cause = %q/%q/%v", code, short, ok)
}
}
func TestProxyReconnectReasonGate(t *testing.T) {
for _, tc := range []struct {
reason string
want bool
}{
{"paused_by_proxy", true},
{"paused_by_user", false},
{"paused_by_schedule", false},
{"paused_by_orchestrator", false},
} {
got := tc.reason == "paused_by_proxy"
if got != tc.want {
t.Errorf("reason %q gate=%v, want %v", tc.reason, got, tc.want)
}
}
}
func TestProxyAdmissionRechecksReasonAfterSnapshot(t *testing.T) {
m := proxyTestManager(t)
task, err := m.CreateTask("proxy gate", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
if err := m.ApplyProxyPause(task.ID); err != nil {
t.Fatal(err)
}
s := newAdmissionTestServer(m, nil)
s.engine.Pause(task.ID, agent.AbortPausedByProxy)
// Simulate a user re-pause winning after proxyReconnectSuccess took its
// snapshot but before the admission lock was acquired.
if _, err := m.pg.Exec(`UPDATE tasks SET pause_reason='paused_by_user' WHERE id=$1`, task.ID); err != nil {
t.Fatal(err)
}
task.updateLifecycle(func(v *taskLifecycleState) { v.PauseReason = "paused_by_user" })
if _, err := s.admitTaskWhenScheduledWithReason(task, "resume", true, false, "paused_by_proxy", true); err == nil {
t.Fatal("proxy admission passed after user changed pause reason")
}
if !s.engine.IsPaused(task.ID) {
t.Fatal("user re-pause barrier was cleared")
}
}
func TestProxyQueuedAdmissionDoesNotResumeEngine(t *testing.T) {
m := proxyTestManager(t)
if err := m.SetConcurrency(true, 1); err != nil {
t.Fatal(err)
}
holder, err := m.CreateTask("holder", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
proxyTask, err := m.CreateTask("queued proxy", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() {
_, _ = m.DeleteTask(holder.ID, DeleteTaskOptions{})
_, _ = m.DeleteTask(proxyTask.ID, DeleteTaskOptions{})
}()
if err := m.ApplyProxyPause(proxyTask.ID); err != nil {
t.Fatal(err)
}
s := newAdmissionTestServer(m, nil)
s.engine.started.Store(holder.ID, true)
s.engine.Pause(proxyTask.ID, agent.AbortPausedByProxy)
queued, err := s.admitTaskWhenScheduledWithReason(proxyTask, "resume", true, false, "paused_by_proxy", true)
if err != nil {
t.Fatal(err)
}
if !queued {
t.Fatal("proxy task was expected to queue behind the running holder")
}
if !s.engine.IsPaused(proxyTask.ID) {
t.Fatal("queued proxy task was resumed")
}
}
func TestTerminalStatusClearsProxyReconnectState(t *testing.T) {
m := proxyTestManager(t)
task, err := m.CreateTask("terminal proxy", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
if err := m.ApplyProxyPause(task.ID); err != nil {
t.Fatal(err)
}
task.updateLifecycle(func(v *taskLifecycleState) {
v.ReconnectAttempt, v.ReconnectMax, v.ReconnectNextAt = 2, 5, 123
v.ReconnectExhausted, v.ReconnectLastError = true, "proxy down"
})
if err := m.SetTaskStatus(task.ID, "timeout"); err != nil {
t.Fatal(err)
}
stored, err := m.pg.GetTask(mustTaskID(t, task.ID))
if err != nil {
t.Fatal(err)
}
if stored.PauseReason != "" {
t.Fatalf("terminal row retained pause reason %q", stored.PauseReason)
}
st := task.lifecycleSnapshot()
if st.PauseReason != "" || st.ReconnectAttempt != 0 || st.ReconnectNextAt != 0 || st.ReconnectExhausted {
t.Fatalf("terminal lifecycle retained reconnect state: %+v", st)
}
}
func TestProxyAdmissionPreservesCalendarOverride(t *testing.T) {
m := proxyTestManager(t)
task, err := m.CreateTask("calendar proxy", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
if _, err := m.pg.Exec(`UPDATE tasks SET schedule_paused=true, schedule_manual_until='2099-01-01 00:00:00+00' WHERE id=$1`, task.ID); err != nil {
t.Fatal(err)
}
if err := m.ApplyProxyPause(task.ID); err != nil {
t.Fatal(err)
}
s := newAdmissionTestServer(m, nil)
s.engine.Pause(task.ID, agent.AbortPausedByProxy)
if queued, err := s.admitTaskWhenScheduledWithReason(task, "resume", true, false, "paused_by_proxy", true); err != nil || queued {
t.Fatalf("proxy admission queued=%v err=%v", queued, err)
}
var schedulePaused bool
var manualUntil time.Time
if err := m.pg.QueryRow(`SELECT schedule_paused, schedule_manual_until FROM tasks WHERE id=$1`, task.ID).Scan(&schedulePaused, &manualUntil); err != nil {
t.Fatal(err)
}
if !schedulePaused || !manualUntil.Equal(time.Date(2099, 1, 1, 0, 0, 0, 0, time.UTC)) {
t.Fatalf("proxy admission changed calendar state: paused=%v until=%v", schedulePaused, manualUntil)
}
}
func TestProxyReconnectSuccessKeepsCompetingUserPauseBarrier(t *testing.T) {
m := proxyTestManager(t)
task, err := m.CreateTask("proxy success race", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
s := newAdmissionTestServer(m, nil)
m.SetTaskControlMutex(&s.concMu)
m.globalProxy = "http://proxy.invalid"
if _, err := m.ApplyProxyPauseWithCallback(task.ID, func() { s.engine.Pause(task.ID, agent.AbortPausedByProxy) }); err != nil {
t.Fatal(err)
}
var pauseErr error
s.proxyReconnectSuccessWithHook(func() {
_, pauseErr = s.applyTaskControl(task, "pause")
})
if pauseErr != nil {
t.Fatalf("competing user pause failed: %v", pauseErr)
}
state := task.lifecycleSnapshot()
if state.PauseReason != "paused_by_user" || !state.Paused {
t.Fatalf("user pause was overwritten: %+v", state)
}
if !s.engine.IsPaused(task.ID) {
t.Fatal("user pause barrier was cleared by a post-admission Resume")
}
}
func TestProxyPauseBarrierInstalledBeforeControlUnlock(t *testing.T) {
m := proxyTestManager(t)
task, err := m.CreateTask("proxy barrier", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
s := newAdmissionTestServer(m, nil)
m.SetTaskControlMutex(&s.concMu)
m.globalProxy = "http://proxy.invalid"
userDone := make(chan error, 1)
transitioned, err := m.ApplyProxyPauseWithCallback(task.ID, func() {
s.engine.Pause(task.ID, agent.AbortPausedByProxy)
if !s.engine.IsPaused(task.ID) {
t.Fatal("proxy barrier was not installed in the locked callback")
}
go func() {
_, pauseErr := s.applyTaskControl(task, "pause")
userDone <- pauseErr
}()
select {
case <-userDone:
t.Fatal("user pause acquired concMu before proxy callback returned")
case <-time.After(20 * time.Millisecond):
}
})
if err != nil || !transitioned {
t.Fatalf("proxy transition=%v err=%v", transitioned, err)
}
if err := <-userDone; err != nil {
t.Fatalf("concurrent user pause failed: %v", err)
}
if !s.engine.IsPaused(task.ID) {
t.Fatal("concurrent user pause left the engine running")
}
if got := task.lifecycleSnapshot().PauseReason; got != "paused_by_user" {
t.Fatalf("concurrent user pause reason=%q", got)
}
}
func TestProxyPauseDoesNotReviveExhaustedOrNonProxyTasks(t *testing.T) {
m := proxyTestManager(t)
exhausted, err := m.CreateTask("exhausted proxy", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
userPaused, err := m.CreateTask("user paused", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
other, err := m.CreateTask("new proxy failure", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() {
_, _ = m.DeleteTask(exhausted.ID, DeleteTaskOptions{})
_, _ = m.DeleteTask(userPaused.ID, DeleteTaskOptions{})
_, _ = m.DeleteTask(other.ID, DeleteTaskOptions{})
}()
s := newAdmissionTestServer(m, nil)
m.SetTaskControlMutex(&s.concMu)
m.globalProxy = "http://proxy.invalid"
if err := m.ApplyProxyPause(exhausted.ID); err != nil {
t.Fatal(err)
}
if err := m.ApplyTaskPauseOriginReason(userPaused.ID, "paused_by_user", false); err != nil {
t.Fatal(err)
}
exhausted.updateLifecycle(func(v *taskLifecycleState) {
v.ReconnectAttempt, v.ReconnectMax, v.ReconnectNextAt = 5, 5, 12345
v.ReconnectExhausted, v.ReconnectLastError = true, "proxy down"
})
userPaused.updateLifecycle(func(v *taskLifecycleState) {
v.ReconnectAttempt, v.ReconnectMax, v.ReconnectNextAt = 3, 7, 23456
v.ReconnectExhausted, v.ReconnectLastError = true, "manual pause state"
})
exhaustedBefore := exhausted.lifecycleSnapshot()
userBefore := userPaused.lifecycleSnapshot()
// A no-op proxy pause on an already paused task must not reset either state.
s.pauseProxyTask(exhausted.ID, errors.New("another proxy failure"))
s.pauseProxyTask(userPaused.ID, errors.New("proxy failure while user paused"))
if got := exhausted.lifecycleSnapshot(); got.ReconnectNextAt != exhaustedBefore.ReconnectNextAt || !got.ReconnectExhausted || got.ReconnectLastError != exhaustedBefore.ReconnectLastError {
t.Fatalf("exhausted task was revived: before=%+v after=%+v", exhaustedBefore, got)
}
if got := userPaused.lifecycleSnapshot(); got.PauseReason != "paused_by_user" || got.ReconnectNextAt != userBefore.ReconnectNextAt || !got.ReconnectExhausted {
t.Fatalf("user-paused task was mutated: before=%+v after=%+v", userBefore, got)
}
// A genuinely new proxy pause of another running task initializes only that
// task's reconnect round.
s.pauseProxyTask(other.ID, errors.New("new proxy failure"))
if got := exhausted.lifecycleSnapshot(); got.ReconnectNextAt != exhaustedBefore.ReconnectNextAt || !got.ReconnectExhausted {
t.Fatalf("subsequent proxy pause revived exhausted task: %+v", got)
}
if got := other.lifecycleSnapshot(); got.PauseReason != "paused_by_proxy" || got.ReconnectMax == 0 || got.ReconnectExhausted {
t.Fatalf("new proxy task did not initialize reconnect state: %+v", got)
}
}
func TestProxyReconnectRoundResetsAfterManualResume(t *testing.T) {
m := proxyTestManager(t)
task, err := m.CreateTask("proxy retry round", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
s := newAdmissionTestServer(m, nil)
m.SetTaskControlMutex(&s.concMu)
m.globalProxy = "http://proxy.invalid"
if err := m.ApplyProxyPause(task.ID); err != nil {
t.Fatal(err)
}
s.engine.Pause(task.ID, agent.AbortPausedByProxy)
task.updateLifecycle(func(v *taskLifecycleState) {
v.ReconnectAttempt, v.ReconnectMax, v.ReconnectNextAt = 5, 5, 12345
v.ReconnectExhausted, v.ReconnectLastError = true, "proxy exhausted"
})
if _, err := s.applyTaskControl(task, "resume"); err != nil {
t.Fatal(err)
}
resumed := task.lifecycleSnapshot()
if resumed.PauseReason != "" || resumed.ReconnectAttempt != 0 || resumed.ReconnectMax != 0 || resumed.ReconnectNextAt != 0 || resumed.ReconnectExhausted || resumed.ReconnectLastError != "" {
t.Fatalf("manual resume retained reconnect state: %+v", resumed)
}
s.pauseProxyTask(task.ID, errors.New("proxy failed again"))
paused := task.lifecycleSnapshot()
if paused.PauseReason != "paused_by_proxy" || paused.ReconnectAttempt != 0 || paused.ReconnectExhausted {
t.Fatalf("new proxy round was not reset: %+v", paused)
}
if paused.ReconnectMax == 0 || paused.ReconnectNextAt == 0 || paused.ReconnectLastError == "" {
t.Fatalf("new proxy round did not initialize reconnect state: %+v", paused)
}
}
func TestProxyReconnectDropsStaleProbeFailureAfterNewRound(t *testing.T) {
m := proxyTestManager(t)
task, err := m.CreateTask("proxy stale probe", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
s := newAdmissionTestServer(m, nil)
m.SetTaskControlMutex(&s.concMu)
m.globalProxy = "http://proxy.invalid"
// Keep the canceled test server from starting its supervisor. The test owns
// the reconnect tick and injects a blocking probe below.
s.proxyReconnect.mu.Lock()
s.proxyReconnect.wake = make(chan struct{}, 1)
s.proxyReconnect.mu.Unlock()
oldMax, hadMax, err := m.pg.GetSetting(settingProxyReconnectMax)
if err != nil {
t.Fatal(err)
}
defer func() {
if hadMax {
_ = m.pg.SetSetting(settingProxyReconnectMax, oldMax)
} else {
_, _ = m.pg.Exec(`DELETE FROM settings WHERE key=$1`, settingProxyReconnectMax)
}
}()
if err := m.pg.SetSetting(settingProxyReconnectMax, "1"); err != nil {
t.Fatal(err)
}
s.proxyProbe = func(context.Context, string, string, []string, time.Duration) egressProbeResult {
return egressProbeResult{Error: "initial probe failure"}
}
s.pauseProxyTask(task.ID, errors.New("initial proxy failure"))
task.updateLifecycle(func(v *taskLifecycleState) { v.ReconnectNextAt = 0 })
oldRound := task.lifecycleSnapshot().ReconnectGeneration
probeStarted := make(chan struct{})
releaseProbe := make(chan struct{})
var probeStartedOnce sync.Once
s.proxyProbe = func(context.Context, string, string, []string, time.Duration) egressProbeResult {
probeStartedOnce.Do(func() { close(probeStarted) })
<-releaseProbe
return egressProbeResult{Error: "stale probe failure"}
}
tickDone := make(chan struct{})
go func() {
s.proxyReconnectTick()
close(tickDone)
}()
select {
case <-probeStarted:
case <-time.After(time.Second):
t.Fatal("reconnect tick did not start its probe")
}
if _, err := s.applyTaskControl(task, "resume"); err != nil {
t.Fatal(err)
}
s.pauseProxyTask(task.ID, errors.New("new proxy failure"))
newRound := task.lifecycleSnapshot()
if newRound.ReconnectGeneration == oldRound || newRound.ReconnectAttempt != 0 || newRound.ReconnectExhausted {
t.Fatalf("new proxy round was not established: old=%d new=%+v", oldRound, newRound)
}
close(releaseProbe)
select {
case <-tickDone:
case <-time.After(time.Second):
t.Fatal("reconnect tick did not finish")
}
after := task.lifecycleSnapshot()
if after.PauseReason != "paused_by_proxy" || after.ReconnectGeneration != newRound.ReconnectGeneration || after.ReconnectAttempt != 0 || after.ReconnectExhausted || after.ReconnectNextAt != newRound.ReconnectNextAt || after.ReconnectLastError != newRound.ReconnectLastError {
t.Fatalf("stale probe mutated the new round: before=%+v after=%+v", newRound, after)
}
}
func TestProxyReconnectManualSuccessDoesNotAdmitNewRound(t *testing.T) {
m := proxyTestManager(t)
task, err := m.CreateTask("proxy stale manual success", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
s := newAdmissionTestServer(m, nil)
m.SetTaskControlMutex(&s.concMu)
m.globalProxy = "http://proxy.invalid"
if err := m.ApplyProxyPause(task.ID); err != nil {
t.Fatal(err)
}
s.engine.Pause(task.ID, agent.AbortPausedByProxy)
guards := s.snapshotProxyReconnectGuards(false)
task.updateLifecycle(func(v *taskLifecycleState) {
v.ReconnectGeneration++
v.ReconnectNextAt = 987654
v.ReconnectLastError = "new round still failing"
})
before := task.lifecycleSnapshot()
s.proxyReconnectSuccessWithGuards(guards, nil)
after := task.lifecycleSnapshot()
if after.PauseReason != "paused_by_proxy" || !s.engine.IsPaused(task.ID) {
t.Fatalf("stale manual success admitted new round: %+v", after)
}
if after.ReconnectGeneration != before.ReconnectGeneration || after.ReconnectNextAt != before.ReconnectNextAt || after.ReconnectLastError != before.ReconnectLastError {
t.Fatalf("stale manual success wiped new round state: before=%+v after=%+v", before, after)
}
}
func TestProxyReconnectSettingsSuccessGuardsGenerationZero(t *testing.T) {
m := proxyTestManager(t)
task, err := m.CreateTask("proxy generation zero", "goal", nil, 0, 0)
if err != nil {
t.Fatal(err)
}
defer func() { _, _ = m.DeleteTask(task.ID, DeleteTaskOptions{}) }()
s := newAdmissionTestServer(m, nil)
m.SetTaskControlMutex(&s.concMu)
m.globalProxy = "http://proxy.invalid"
if err := m.ApplyProxyPause(task.ID); err != nil {
t.Fatal(err)
}
s.engine.Pause(task.ID, agent.AbortPausedByProxy)
initial := task.lifecycleSnapshot()
if initial.ReconnectGeneration != 0 {
t.Fatalf("fresh persisted proxy pause generation=%d, want zero", initial.ReconnectGeneration)
}
guards := s.snapshotProxyReconnectGuards(false)
task.updateLifecycle(func(v *taskLifecycleState) {
v.ReconnectGeneration = 1
v.ReconnectNextAt = 13579
v.ReconnectLastError = "generation one failure"
})
before := task.lifecycleSnapshot()
s.proxyReconnectSuccessWithGuards(guards, nil)
after := task.lifecycleSnapshot()
if after.PauseReason != "paused_by_proxy" || !s.engine.IsPaused(task.ID) {
t.Fatalf("generation-zero stale settings success admitted new round: %+v", after)
}
if after.ReconnectGeneration != before.ReconnectGeneration || after.ReconnectNextAt != before.ReconnectNextAt || after.ReconnectLastError != before.ReconnectLastError {
t.Fatalf("generation-zero stale settings success wiped state: before=%+v after=%+v", before, after)
}
}