fix(hatchery/swarm): adjust http client keepAlive (#3438)

Signed-off-by: Yvonnick Esnault <yvonnick.esnault@corp.ovh.com>
ovh · Oct 11, 2018 · cc8b7fb · cc8b7fb
1 parent 64909bd
commit cc8b7fb
Show file tree

Hide file tree

Showing 4 changed files with 34 additions and 16 deletions.
diff --git a/engine/hatchery/swarm/swarm.go b/engine/hatchery/swarm/swarm.go
@@ -115,21 +115,30 @@ func (h *HatcherySwarm) Init() error {
 
 			if tlsc != nil {
 				httpClient.Transport = &http.Transport{
-					Dial: (&net.Dialer{
-						Timeout: 30 * time.Second,
-					}).Dial,
-					MaxIdleConns:        500,
-					MaxIdleConnsPerHost: 500,
-					TLSHandshakeTimeout: 30 * time.Second,
-					TLSClientConfig:     tlsc,
+					DialContext: (&net.Dialer{
+						Timeout:   30 * time.Second,
+						KeepAlive: 0 * time.Second,
+						DualStack: true,
+					}).DialContext,
+					MaxIdleConns:          100,
+					IdleConnTimeout:       20 * time.Second,
+					TLSHandshakeTimeout:   10 * time.Second,
+					ExpectContinueTimeout: 1 * time.Second,
+					ResponseHeaderTimeout: 30 * time.Second,
+					TLSClientConfig:       tlsc,
 				}
 			} else {
 				httpClient.Transport = &http.Transport{
-					Dial: (&net.Dialer{
-						Timeout: 30 * time.Second,
-					}).Dial,
-					MaxIdleConns:        500,
-					MaxIdleConnsPerHost: 500,
+					DialContext: (&net.Dialer{
+						Timeout:   30 * time.Second,
+						KeepAlive: 0 * time.Second,
+						DualStack: true,
+					}).DialContext,
+					MaxIdleConns:          100,
+					IdleConnTimeout:       20 * time.Second,
+					TLSHandshakeTimeout:   10 * time.Second,
+					ExpectContinueTimeout: 1 * time.Second,
+					ResponseHeaderTimeout: 30 * time.Second,
 				}
 			}
 
@@ -247,7 +256,7 @@ func (h *HatcherySwarm) SpawnWorker(ctx context.Context, spawnArgs hatchery.Spaw
 					network = name + "-net"
 					networkAlias = "worker"
 					if err := h.createNetwork(ctx, dockerClient, network); err != nil {
-						log.Warning("hatchery> swarm> SpawnWorker> Unable to create network %s for jobID %d : %v", network, spawnArgs.JobID, err)
+						log.Warning("hatchery> swarm> SpawnWorker> Unable to create network %s on %s for jobID %d : %v", network, dockerClient.name, spawnArgs.JobID, err)
 						next()
 						return "", err
 					}
@@ -273,7 +282,7 @@ func (h *HatcherySwarm) SpawnWorker(ctx context.Context, spawnArgs hatchery.Spaw
 						m := strings.Replace(e, "CDS_SERVICE_MEMORY=", "", -1)
 						i, err := strconv.Atoi(m)
 						if err != nil {
-							log.Warning("hatchery> swarm> SpawnWorker> Unable to parse service option %s : %s", e, err)
+							log.Warning("hatchery> swarm> SpawnWorker> Unable to parse service option %s : %v", e, err)
 							continue
 						}
 						serviceMemory = int64(i)
@@ -308,7 +317,7 @@ func (h *HatcherySwarm) SpawnWorker(ctx context.Context, spawnArgs hatchery.Spaw
 				}
 
 				if err := h.createAndStartContainer(ctx, dockerClient, args, spawnArgs); err != nil {
-					log.Warning("hatchery> swarm> SpawnWorker> Unable to start required container: %s", err)
+					log.Warning("hatchery> swarm> SpawnWorker> Unable to start required container on %s: %s", dockerClient.name, err)
 					return "", err
 				}
 				services = append(services, serviceName)

diff --git a/sdk/cdsclient/client.go b/sdk/cdsclient/client.go
@@ -37,6 +37,7 @@ func NewHTTPClient(timeout time.Duration, insecureSkipVerifyTLS bool) *http.Clie
 			IdleConnTimeout:       30 * time.Second,
 			TLSHandshakeTimeout:   10 * time.Second,
 			ExpectContinueTimeout: 1 * time.Second,
+			ResponseHeaderTimeout: 30 * time.Second,
 			TLSClientConfig:       &tls.Config{InsecureSkipVerify: insecureSkipVerifyTLS},
 		},
 	}

diff --git a/sdk/cdsclient/http.go b/sdk/cdsclient/http.go
@@ -292,7 +292,9 @@ func (c *client) Stream(ctx context.Context, method string, path string, body io
 			}
 			continue
 		}
-
+		if resp != nil && resp.Body != nil {
+			resp.Body.Close()
+		}
 		if errDo != nil {
 			return nil, nil, 0, errDo
 		}

diff --git a/sdk/hatchery/register.go b/sdk/hatchery/register.go
@@ -33,6 +33,12 @@ func workerRegister(h Interface, startWorkerChan chan<- workerStarterRequest) er
 		if models[k].Type != h.ModelType() {
 			continue
 		}
+		if h.NeedRegistration(&models[k]) || models[k].CheckRegistration {
+			log.Debug("hatchery> workerRegister> need register")
+		} else {
+			continue
+		}
+
 		maxRegistration := int64(math.Floor(float64(h.Configuration().Provision.MaxWorker) / 4))
 		if atomic.LoadInt64(&nbRegisteringWorkerModels) > maxRegistration {
 			log.Debug("hatchery> workerRegister> max registering worker reached")