fix(worker): harden control-plane lifecycle
Все проверки выполнены успешно
CI / test (push) Successful in 2m32s
Docker / Build and publish worker image (push) Successful in 18m17s
Все проверки выполнены успешно
CI / test (push) Successful in 2m32s
Docker / Build and publish worker image (push) Successful in 18m17s
- reconnect safely after token rotation and retry leased results - reject malformed tasks and remove production cluster debug mutation - validate environment files and require immutable container images BREAKING CHANGE: Docker install, deploy, and Compose now require an immutable repository@sha256 image reference.
Этот коммит содержится в:
@@ -1,6 +1,9 @@
|
||||
package distworker
|
||||
|
||||
import (
|
||||
"encoding/json"
|
||||
"net/http"
|
||||
"net/http/httptest"
|
||||
"strconv"
|
||||
"sync"
|
||||
"sync/atomic"
|
||||
@@ -8,6 +11,7 @@ import (
|
||||
"time"
|
||||
|
||||
"github.com/Jeffail/tunny"
|
||||
"github.com/gorilla/websocket"
|
||||
"github.com/stretchr/testify/assert"
|
||||
"github.com/stretchr/testify/require"
|
||||
|
||||
@@ -174,6 +178,835 @@ func TestEnqueueRespectsBackpressure(t *testing.T) {
|
||||
}
|
||||
}
|
||||
|
||||
func TestStopRejectsTerminalResultsWithoutClosingResultChannel(t *testing.T) {
|
||||
r := NewRunner(&Config{MaxConcurrency: 1})
|
||||
r.results = make(chan resultEnvelope, 1)
|
||||
r.Stop()
|
||||
|
||||
assert.False(t, r.enqueueFailedCheck(wire.CheckJob{JobID: "job-1", LeaseToken: "lease-1"}, "unsupported_kind: rkn"))
|
||||
assert.Empty(t, r.results)
|
||||
|
||||
select {
|
||||
case r.results <- resultEnvelope{}:
|
||||
default:
|
||||
t.Fatal("results channel should remain open after Stop")
|
||||
}
|
||||
}
|
||||
|
||||
func TestRotateTokenReconnectsWithoutStoppingRunner(t *testing.T) {
|
||||
const (
|
||||
oldToken = "old-token"
|
||||
newToken = "new-token"
|
||||
)
|
||||
|
||||
connections := make(chan string, 2)
|
||||
results := make(chan wire.WorkerMessage, 1)
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
switch req.URL.Path {
|
||||
case "/api/internal/workers/rotate-token":
|
||||
if req.Header.Get("Authorization") != "Bearer "+oldToken {
|
||||
http.Error(w, "unexpected rotation token", http.StatusUnauthorized)
|
||||
return
|
||||
}
|
||||
_ = json.NewEncoder(w).Encode(struct {
|
||||
AuthToken string `json:"auth_token"`
|
||||
}{AuthToken: newToken})
|
||||
case "/worker":
|
||||
conn, err := upgrader.Upgrade(w, req, nil)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
defer conn.Close()
|
||||
token := req.URL.Query().Get("token")
|
||||
connections <- token
|
||||
if token == oldToken {
|
||||
_, _, _ = conn.ReadMessage() // Rotation must close this connection.
|
||||
return
|
||||
}
|
||||
if token != newToken {
|
||||
return
|
||||
}
|
||||
if conn.WriteJSON(wire.WorkerMessage{Kind: "task", Task: &wire.CheckJob{
|
||||
JobID: "after-rotation", LeaseToken: "lease", CheckID: 1, Kind: "http",
|
||||
}}) != nil {
|
||||
return
|
||||
}
|
||||
for {
|
||||
var message wire.WorkerMessage
|
||||
if err := conn.ReadJSON(&message); err != nil {
|
||||
return
|
||||
}
|
||||
if message.Kind == "result" && message.Result != nil && message.Result.JobID == "after-rotation" {
|
||||
results <- message
|
||||
return
|
||||
}
|
||||
}
|
||||
default:
|
||||
http.NotFound(w, req)
|
||||
}
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: oldToken, MaxConcurrency: 1})
|
||||
r.executor = func(payload interface{}) interface{} {
|
||||
job := payload.(wire.CheckJob)
|
||||
return []wire.CheckResultReport{{JobID: job.JobID, CheckID: job.CheckID, State: "OK"}}
|
||||
}
|
||||
startDone := make(chan error, 1)
|
||||
go func() { startDone <- r.Start() }()
|
||||
t.Cleanup(func() {
|
||||
r.Stop()
|
||||
select {
|
||||
case err := <-startDone:
|
||||
require.NoError(t, err)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("runner did not stop")
|
||||
}
|
||||
})
|
||||
|
||||
select {
|
||||
case token := <-connections:
|
||||
require.Equal(t, oldToken, token)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("worker did not establish its initial control connection")
|
||||
}
|
||||
|
||||
gotToken, err := r.RotateToken(t.Context())
|
||||
require.NoError(t, err)
|
||||
require.Equal(t, newToken, gotToken)
|
||||
require.Equal(t, newToken, r.Token())
|
||||
|
||||
select {
|
||||
case token := <-connections:
|
||||
require.Equal(t, newToken, token)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("worker did not reconnect with the replacement token")
|
||||
}
|
||||
select {
|
||||
case result := <-results:
|
||||
require.NotNil(t, result.Result)
|
||||
assert.Equal(t, "OK", result.Result.State)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("runner did not execute work after token rotation")
|
||||
}
|
||||
assert.False(t, r.stopped(), "rotation must not stop runner-owned subsystems")
|
||||
}
|
||||
|
||||
func TestStopClosesAndJoinsIdleControlConnection(t *testing.T) {
|
||||
closed := make(chan struct{}, 1)
|
||||
connected := make(chan struct{}, 1)
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
if req.URL.Path != "/worker" {
|
||||
http.NotFound(w, req)
|
||||
return
|
||||
}
|
||||
conn, err := upgrader.Upgrade(w, req, nil)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
defer conn.Close()
|
||||
connected <- struct{}{}
|
||||
_, _, _ = conn.ReadMessage()
|
||||
closed <- struct{}{}
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: "token", MaxConcurrency: 1})
|
||||
startDone := make(chan error, 1)
|
||||
go func() { startDone <- r.Start() }()
|
||||
t.Cleanup(func() { r.Stop() })
|
||||
|
||||
select {
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("worker did not establish idle control connection")
|
||||
case <-connected:
|
||||
}
|
||||
r.Stop()
|
||||
select {
|
||||
case <-closed:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Stop did not close the idle control connection")
|
||||
}
|
||||
select {
|
||||
case err := <-startDone:
|
||||
require.NoError(t, err)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Stop did not join the control loop")
|
||||
}
|
||||
}
|
||||
|
||||
func TestStopCancelsDialInProgress(t *testing.T) {
|
||||
dialStarted := make(chan struct{})
|
||||
allowUpgrade := make(chan struct{})
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
if req.URL.Path != "/worker" {
|
||||
http.NotFound(w, req)
|
||||
return
|
||||
}
|
||||
close(dialStarted)
|
||||
<-allowUpgrade
|
||||
_, _ = upgrader.Upgrade(w, req, nil)
|
||||
}))
|
||||
defer func() {
|
||||
close(allowUpgrade)
|
||||
server.Close()
|
||||
}()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: "token", MaxConcurrency: 1})
|
||||
startDone := make(chan error, 1)
|
||||
go func() { startDone <- r.Start() }()
|
||||
select {
|
||||
case <-dialStarted:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("worker did not begin websocket dial")
|
||||
}
|
||||
r.Stop()
|
||||
select {
|
||||
case err := <-startDone:
|
||||
require.NoError(t, err)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Stop did not join a canceled websocket dial")
|
||||
}
|
||||
}
|
||||
|
||||
func TestStartAndStopRegisterControlLoopSafely(t *testing.T) {
|
||||
server := httptest.NewServer(http.NotFoundHandler())
|
||||
defer server.Close()
|
||||
for i := 0; i < 25; i++ {
|
||||
r := NewRunner(&Config{URL: server.URL, Token: "token", MaxConcurrency: 1})
|
||||
startDone := make(chan error, 1)
|
||||
stopDone := make(chan struct{})
|
||||
go func() { startDone <- r.Start() }()
|
||||
go func() {
|
||||
r.Stop()
|
||||
close(stopDone)
|
||||
}()
|
||||
select {
|
||||
case <-stopDone:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Stop did not complete")
|
||||
}
|
||||
select {
|
||||
case <-startDone:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Start did not return after concurrent Stop")
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
func TestRotateTokenPreservesDequeuedResult(t *testing.T) {
|
||||
const (
|
||||
oldToken = "old-token"
|
||||
newToken = "new-token"
|
||||
)
|
||||
connected := make(chan string, 2)
|
||||
delivered := make(chan wire.WorkerMessage, 1)
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
switch req.URL.Path {
|
||||
case "/api/internal/workers/rotate-token":
|
||||
_ = json.NewEncoder(w).Encode(struct {
|
||||
AuthToken string `json:"auth_token"`
|
||||
}{AuthToken: newToken})
|
||||
case "/worker":
|
||||
conn, err := upgrader.Upgrade(w, req, nil)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
defer conn.Close()
|
||||
token := req.URL.Query().Get("token")
|
||||
connected <- token
|
||||
if token == oldToken {
|
||||
var message wire.WorkerMessage
|
||||
if conn.ReadJSON(&message) == nil {
|
||||
delivered <- message
|
||||
}
|
||||
return
|
||||
}
|
||||
if token != newToken {
|
||||
return
|
||||
}
|
||||
var message wire.WorkerMessage
|
||||
if conn.ReadJSON(&message) == nil {
|
||||
delivered <- message
|
||||
}
|
||||
}
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: oldToken, MaxConcurrency: 1})
|
||||
enteredWrite := make(chan struct{})
|
||||
releaseWrite := make(chan struct{})
|
||||
var once sync.Once
|
||||
r.beforeControlWrite = func() {
|
||||
once.Do(func() {
|
||||
close(enteredWrite)
|
||||
<-releaseWrite
|
||||
})
|
||||
}
|
||||
startDone := make(chan error, 1)
|
||||
go func() { startDone <- r.Start() }()
|
||||
t.Cleanup(func() {
|
||||
r.Stop()
|
||||
select {
|
||||
case <-startDone:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("runner did not stop")
|
||||
}
|
||||
})
|
||||
|
||||
select {
|
||||
case token := <-connected:
|
||||
require.Equal(t, oldToken, token)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("worker did not establish its initial control connection")
|
||||
}
|
||||
r.results <- resultEnvelope{job: wire.CheckJob{JobID: "result", LeaseToken: "lease"}, reports: []wire.CheckResultReport{{JobID: "result", State: "OK"}}}
|
||||
select {
|
||||
case <-enteredWrite:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("writer did not dequeue result")
|
||||
}
|
||||
rotated := make(chan error, 1)
|
||||
go func() {
|
||||
_, err := r.RotateToken(t.Context())
|
||||
rotated <- err
|
||||
}()
|
||||
close(releaseWrite)
|
||||
require.NoError(t, <-rotated)
|
||||
|
||||
select {
|
||||
case message := <-delivered:
|
||||
require.NotNil(t, message.Result)
|
||||
assert.Equal(t, "result", message.Result.JobID)
|
||||
assert.Equal(t, "lease", message.Result.LeaseToken)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("dequeued result was lost during rotation")
|
||||
}
|
||||
}
|
||||
|
||||
func TestRotateTokenSerializesWithStop(t *testing.T) {
|
||||
rotationStarted := make(chan struct{})
|
||||
releaseHandler := make(chan struct{})
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
switch req.URL.Path {
|
||||
case "/api/internal/workers/rotate-token":
|
||||
close(rotationStarted)
|
||||
<-releaseHandler
|
||||
case "/worker":
|
||||
conn, err := upgrader.Upgrade(w, req, nil)
|
||||
if err == nil {
|
||||
defer conn.Close()
|
||||
_, _, _ = conn.ReadMessage()
|
||||
}
|
||||
}
|
||||
}))
|
||||
defer func() {
|
||||
close(releaseHandler)
|
||||
server.Close()
|
||||
}()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: "old-token", MaxConcurrency: 1})
|
||||
startDone := make(chan error, 1)
|
||||
go func() { startDone <- r.Start() }()
|
||||
t.Cleanup(func() { r.Stop() })
|
||||
|
||||
// Wait for Start to install its client before beginning rotation.
|
||||
deadline := time.After(time.Second)
|
||||
for {
|
||||
r.clientMu.Lock()
|
||||
started := r.client != nil
|
||||
r.clientMu.Unlock()
|
||||
if started {
|
||||
break
|
||||
}
|
||||
select {
|
||||
case <-deadline:
|
||||
t.Fatal("runner did not start")
|
||||
default:
|
||||
time.Sleep(time.Millisecond)
|
||||
}
|
||||
}
|
||||
|
||||
rotated := make(chan error, 1)
|
||||
go func() {
|
||||
_, err := r.RotateToken(t.Context())
|
||||
rotated <- err
|
||||
}()
|
||||
select {
|
||||
case <-rotationStarted:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("rotation request did not start")
|
||||
}
|
||||
stopped := make(chan struct{})
|
||||
go func() {
|
||||
r.Stop()
|
||||
close(stopped)
|
||||
}()
|
||||
require.Error(t, <-rotated, "Stop must cancel an in-flight rotation request")
|
||||
select {
|
||||
case <-stopped:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Stop did not complete after canceling rotation")
|
||||
}
|
||||
select {
|
||||
case err := <-startDone:
|
||||
require.NoError(t, err)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("runner did not stop")
|
||||
}
|
||||
_, err := r.RotateToken(t.Context())
|
||||
require.Error(t, err, "rotation cannot succeed after shutdown")
|
||||
}
|
||||
|
||||
func TestStopUnblocksRotationWaitingForWriter(t *testing.T) {
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
connected := make(chan struct{}, 1)
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
switch req.URL.Path {
|
||||
case "/api/internal/workers/rotate-token":
|
||||
_ = json.NewEncoder(w).Encode(struct {
|
||||
AuthToken string `json:"auth_token"`
|
||||
}{AuthToken: "new-token"})
|
||||
case "/worker":
|
||||
conn, err := upgrader.Upgrade(w, req, nil)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
defer conn.Close()
|
||||
connected <- struct{}{}
|
||||
_, _, _ = conn.ReadMessage()
|
||||
}
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: "old-token", MaxConcurrency: 1})
|
||||
writeBlocked := make(chan struct{})
|
||||
var once sync.Once
|
||||
r.beforeControlWrite = func() {
|
||||
once.Do(func() {
|
||||
close(writeBlocked)
|
||||
<-r.controlCtx.Done()
|
||||
})
|
||||
}
|
||||
rotationReady := make(chan struct{})
|
||||
r.beforeTokenCommit = func() { close(rotationReady) }
|
||||
startDone := make(chan error, 1)
|
||||
go func() { startDone <- r.Start() }()
|
||||
t.Cleanup(func() { r.Stop() })
|
||||
select {
|
||||
case <-connected:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("worker did not connect")
|
||||
}
|
||||
r.results <- resultEnvelope{job: wire.CheckJob{JobID: "blocked", LeaseToken: "lease"}, reports: []wire.CheckResultReport{{JobID: "blocked", State: "OK"}}}
|
||||
select {
|
||||
case <-writeBlocked:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("writer did not block")
|
||||
}
|
||||
rotated := make(chan error, 1)
|
||||
go func() {
|
||||
_, err := r.RotateToken(t.Context())
|
||||
rotated <- err
|
||||
}()
|
||||
select {
|
||||
case <-rotationReady:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("rotation did not reach writer serialization")
|
||||
}
|
||||
stopped := make(chan struct{})
|
||||
go func() {
|
||||
r.Stop()
|
||||
close(stopped)
|
||||
}()
|
||||
select {
|
||||
case <-stopped:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Stop deadlocked behind rotation waiting for writer")
|
||||
}
|
||||
select {
|
||||
case <-rotated:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("rotation did not unblock after Stop closed the connection")
|
||||
}
|
||||
select {
|
||||
case err := <-startDone:
|
||||
require.NoError(t, err)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("runner did not stop")
|
||||
}
|
||||
}
|
||||
|
||||
func TestStopWinsBeforePostHTTPRotationCommit(t *testing.T) {
|
||||
commitReady := make(chan struct{})
|
||||
releaseCommit := make(chan struct{})
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
if req.URL.Path != "/api/internal/workers/rotate-token" {
|
||||
http.NotFound(w, req)
|
||||
return
|
||||
}
|
||||
_ = json.NewEncoder(w).Encode(struct {
|
||||
AuthToken string `json:"auth_token"`
|
||||
}{AuthToken: "new-token"})
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: "old-token", MaxConcurrency: 1})
|
||||
r.client = NewClient(server.URL, "old-token")
|
||||
r.beforeTokenCommit = func() {
|
||||
close(commitReady)
|
||||
<-releaseCommit
|
||||
}
|
||||
rotated := make(chan error, 1)
|
||||
go func() {
|
||||
_, err := r.RotateToken(t.Context())
|
||||
rotated <- err
|
||||
}()
|
||||
select {
|
||||
case <-commitReady:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("rotation did not reach post-HTTP commit")
|
||||
}
|
||||
stopped := make(chan struct{})
|
||||
go func() {
|
||||
r.Stop()
|
||||
close(stopped)
|
||||
}()
|
||||
select {
|
||||
case <-r.stopCh:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Stop did not win lifecycle ownership")
|
||||
}
|
||||
close(releaseCommit)
|
||||
require.Error(t, <-rotated, "rotation cannot succeed after Stop wins")
|
||||
select {
|
||||
case <-stopped:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Stop did not finish")
|
||||
}
|
||||
assert.Equal(t, "old-token", r.Token())
|
||||
}
|
||||
|
||||
func TestRotateTokenUnchangedResponseReleasesLifecycle(t *testing.T) {
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
_ = json.NewEncoder(w).Encode(struct {
|
||||
AuthToken string `json:"auth_token"`
|
||||
}{AuthToken: "old-token"})
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: "old-token", MaxConcurrency: 1})
|
||||
r.client = NewClient(server.URL, "old-token")
|
||||
_, err := r.RotateToken(t.Context())
|
||||
require.Error(t, err)
|
||||
|
||||
stopped := make(chan struct{})
|
||||
go func() {
|
||||
r.Stop()
|
||||
close(stopped)
|
||||
}()
|
||||
select {
|
||||
case <-stopped:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("Stop deadlocked after unchanged rotation response")
|
||||
}
|
||||
_, err = r.RotateToken(t.Context())
|
||||
require.Error(t, err, "later rotation must observe shutdown")
|
||||
}
|
||||
|
||||
func TestRotateTokenClosesStalledWriterBeforeWaiting(t *testing.T) {
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
connected := make(chan struct{}, 1)
|
||||
connectionClosed := make(chan struct{})
|
||||
resent := make(chan wire.WorkerMessage, 1)
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
switch req.URL.Path {
|
||||
case "/api/internal/workers/rotate-token":
|
||||
_ = json.NewEncoder(w).Encode(struct {
|
||||
AuthToken string `json:"auth_token"`
|
||||
}{AuthToken: "new-token"})
|
||||
case "/worker":
|
||||
conn, err := upgrader.Upgrade(w, req, nil)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
defer conn.Close()
|
||||
if req.URL.Query().Get("token") == "old-token" {
|
||||
connected <- struct{}{}
|
||||
_, _, _ = conn.ReadMessage()
|
||||
connectionClosed <- struct{}{}
|
||||
return
|
||||
}
|
||||
var message wire.WorkerMessage
|
||||
if conn.ReadJSON(&message) == nil {
|
||||
resent <- message
|
||||
}
|
||||
}
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: "old-token", MaxConcurrency: 1})
|
||||
writerBlocked := make(chan struct{})
|
||||
var once sync.Once
|
||||
r.beforeControlWrite = func() {
|
||||
once.Do(func() {
|
||||
close(writerBlocked)
|
||||
<-connectionClosed
|
||||
})
|
||||
}
|
||||
startDone := make(chan error, 1)
|
||||
go func() { startDone <- r.Start() }()
|
||||
t.Cleanup(func() { r.Stop() })
|
||||
select {
|
||||
case <-connected:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("worker did not connect")
|
||||
}
|
||||
r.results <- resultEnvelope{job: wire.CheckJob{JobID: "stalled", LeaseToken: "lease"}, reports: []wire.CheckResultReport{{JobID: "stalled", State: "OK"}}}
|
||||
select {
|
||||
case <-writerBlocked:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("writer did not stall")
|
||||
}
|
||||
rotated := make(chan error, 1)
|
||||
go func() {
|
||||
_, err := r.RotateToken(t.Context())
|
||||
rotated <- err
|
||||
}()
|
||||
select {
|
||||
case err := <-rotated:
|
||||
require.NoError(t, err)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("rotation waited for stalled writer before closing its connection")
|
||||
}
|
||||
select {
|
||||
case message := <-resent:
|
||||
require.NotNil(t, message.Result)
|
||||
assert.Equal(t, "stalled", message.Result.JobID)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("failed check result was not requeued after rotation")
|
||||
}
|
||||
r.Stop()
|
||||
select {
|
||||
case err := <-startDone:
|
||||
require.NoError(t, err)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("runner did not stop")
|
||||
}
|
||||
}
|
||||
|
||||
func TestWriterDropsFailedServerMetricSnapshot(t *testing.T) {
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
upgraded := make(chan struct{})
|
||||
closeServer := make(chan struct{})
|
||||
serverClosed := make(chan struct{})
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
conn, err := upgrader.Upgrade(w, req, nil)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
defer conn.Close()
|
||||
close(upgraded)
|
||||
<-closeServer
|
||||
close(serverClosed)
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
conn, err := NewClient(server.URL, "token").WorkerSocket()
|
||||
require.NoError(t, err)
|
||||
defer conn.Close()
|
||||
select {
|
||||
case <-upgraded:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("websocket did not connect")
|
||||
}
|
||||
|
||||
r := NewRunner(&Config{})
|
||||
r.metricResults = make(chan metricEnvelope, 1)
|
||||
enteredWrite := make(chan struct{})
|
||||
releaseWrite := make(chan struct{})
|
||||
r.beforeControlWrite = func() {
|
||||
close(enteredWrite)
|
||||
<-releaseWrite
|
||||
}
|
||||
done := make(chan struct{})
|
||||
writerDone := make(chan struct{})
|
||||
go func() {
|
||||
var writeMu sync.Mutex
|
||||
r.writer(conn, &writeMu, done, 1)
|
||||
close(writerDone)
|
||||
}()
|
||||
r.metricResults <- metricEnvelope{generation: 1, report: wire.ServerMetricReport{ServerID: 1}}
|
||||
select {
|
||||
case <-enteredWrite:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("writer did not dequeue metric snapshot")
|
||||
}
|
||||
close(closeServer)
|
||||
select {
|
||||
case <-serverClosed:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("server did not close websocket")
|
||||
}
|
||||
_ = conn.Close()
|
||||
close(releaseWrite)
|
||||
select {
|
||||
case <-writerDone:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("writer did not return after failed metric write")
|
||||
}
|
||||
_, replayable := r.takeOutbox()
|
||||
assert.False(t, replayable, "failed metric snapshot must not enter result outbox")
|
||||
}
|
||||
|
||||
func TestWriterRequeuesFailedNotificationResult(t *testing.T) {
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
upgraded := make(chan struct{})
|
||||
closeServer := make(chan struct{})
|
||||
serverClosed := make(chan struct{})
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
conn, err := upgrader.Upgrade(w, req, nil)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
defer conn.Close()
|
||||
close(upgraded)
|
||||
<-closeServer
|
||||
close(serverClosed)
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
conn, err := NewClient(server.URL, "token").WorkerSocket()
|
||||
require.NoError(t, err)
|
||||
defer conn.Close()
|
||||
select {
|
||||
case <-upgraded:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("websocket did not connect")
|
||||
}
|
||||
|
||||
r := NewRunner(&Config{})
|
||||
r.notifyResults = make(chan notifyResultEnvelope, 1)
|
||||
enteredWrite := make(chan struct{})
|
||||
releaseWrite := make(chan struct{})
|
||||
r.beforeControlWrite = func() {
|
||||
close(enteredWrite)
|
||||
<-releaseWrite
|
||||
}
|
||||
done := make(chan struct{})
|
||||
writerDone := make(chan struct{})
|
||||
go func() {
|
||||
var writeMu sync.Mutex
|
||||
r.writer(conn, &writeMu, done, 0)
|
||||
close(writerDone)
|
||||
}()
|
||||
r.notifyResults <- notifyResultEnvelope{report: wire.NotificationResultReport{JobID: "notification", LeaseToken: "lease"}}
|
||||
select {
|
||||
case <-enteredWrite:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("writer did not dequeue notification result")
|
||||
}
|
||||
close(closeServer)
|
||||
select {
|
||||
case <-serverClosed:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("server did not close websocket")
|
||||
}
|
||||
_ = conn.Close()
|
||||
close(releaseWrite)
|
||||
select {
|
||||
case <-writerDone:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("writer did not return after failed notification write")
|
||||
}
|
||||
message, replayable := r.takeOutbox()
|
||||
require.True(t, replayable, "failed notification result must enter result outbox")
|
||||
require.NotNil(t, message.NotificationResult)
|
||||
assert.Equal(t, "notification", message.NotificationResult.JobID)
|
||||
}
|
||||
|
||||
func TestMetricGenerationRejectsDisconnectedAndSendsFreshMetric(t *testing.T) {
|
||||
upgrader := websocket.Upgrader{CheckOrigin: func(*http.Request) bool { return true }}
|
||||
firstClosed := make(chan struct{})
|
||||
secondConnected := make(chan struct{})
|
||||
received := make(chan wire.WorkerMessage, 1)
|
||||
var connections atomic.Int32
|
||||
server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, req *http.Request) {
|
||||
conn, err := upgrader.Upgrade(w, req, nil)
|
||||
if err != nil {
|
||||
return
|
||||
}
|
||||
defer conn.Close()
|
||||
if connections.Add(1) == 1 {
|
||||
close(firstClosed)
|
||||
return
|
||||
}
|
||||
close(secondConnected)
|
||||
var message wire.WorkerMessage
|
||||
if conn.ReadJSON(&message) == nil {
|
||||
received <- message
|
||||
}
|
||||
}))
|
||||
defer server.Close()
|
||||
|
||||
r := NewRunner(&Config{URL: server.URL, Token: "token", MaxConcurrency: 1})
|
||||
startDone := make(chan error, 1)
|
||||
go func() { startDone <- r.Start() }()
|
||||
t.Cleanup(func() { r.Stop() })
|
||||
select {
|
||||
case <-firstClosed:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("worker did not establish initial websocket")
|
||||
}
|
||||
// Wait until the old connection has fully torn down, then try to enqueue a
|
||||
// metric in the old-drain/disconnected interleaving.
|
||||
deadline := time.After(time.Second)
|
||||
for r.metricGeneration.Load() != 0 {
|
||||
select {
|
||||
case <-deadline:
|
||||
t.Fatal("old control generation did not clear")
|
||||
default:
|
||||
time.Sleep(time.Millisecond)
|
||||
}
|
||||
}
|
||||
staleCount := 1
|
||||
r.enqueueMetric(wire.ServerMetricReport{ServerID: 1, ProcessCount: &staleCount})
|
||||
assert.Empty(t, r.metricResults, "disconnected metric must not enter bounded channel")
|
||||
select {
|
||||
case r.reconnectCh <- struct{}{}:
|
||||
default:
|
||||
}
|
||||
select {
|
||||
case <-secondConnected:
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("worker did not reconnect")
|
||||
}
|
||||
deadline = time.After(time.Second)
|
||||
for r.metricGeneration.Load() == 0 {
|
||||
select {
|
||||
case <-deadline:
|
||||
t.Fatal("new control generation did not install")
|
||||
default:
|
||||
time.Sleep(time.Millisecond)
|
||||
}
|
||||
}
|
||||
// This enqueue occurs immediately after the new connection installation.
|
||||
freshCount := 2
|
||||
r.enqueueMetric(wire.ServerMetricReport{ServerID: 1, ProcessCount: &freshCount})
|
||||
select {
|
||||
case message := <-received:
|
||||
require.NotNil(t, message.ServerMetric)
|
||||
assert.Equal(t, 2, *message.ServerMetric.ProcessCount)
|
||||
case <-time.After(time.Second):
|
||||
t.Fatal("fresh metric was not sent after reconnect")
|
||||
}
|
||||
}
|
||||
|
||||
func TestApplyInitResizesPool(t *testing.T) {
|
||||
executor := func(payload interface{}) interface{} {
|
||||
return []wire.CheckResultReport{}
|
||||
|
||||
Ссылка в новой задаче
Block a user