Merge pull request #957 from convox/cert-random

Stability updates for CI

api/controllers/builds.go

@@ -250,31 +250,12 @@ func BuildLogs(ws *websocket.Conn) *httperr.Error {
 	// in production loop through docker hosts that the rack is running on
 	// to find the build
 	if os.Getenv("DEVELOPMENT") != "true" {
-		pss, err := models.ListProcesses(os.Getenv("RACK"))
-
+		h, err := findBuildHost(build)
 		if err != nil {
 			return httperr.Server(err)
 		}
 
-		for _, ps := range pss {
-			client, err := ps.Docker()
-
-			if err != nil {
-				return httperr.Server(err)
-			}
-
-			res, err := client.ListContainers(docker.ListContainersOptions{
-				All: true,
-				Filters: map[string][]string{
-					"name": []string{fmt.Sprintf("build-%s", build)},
-				},
-			})
-
-			if len(res) > 0 {
-				host = fmt.Sprintf("http://%s:2376", ps.Host)
-				break
-			}
-		}
+		host = h
 	}
 
 	// proxy to docker container logs
@@ -341,6 +322,39 @@ ForLoop:
 	return httperr.Server(err)
 }
 
+// try to find the docker host that's running a build
+// try a few times with a sleep
+func findBuildHost(build string) (string, error) {
+	for i := 1; i < 5; i++ {
+		pss, err := models.ListProcesses(os.Getenv("RACK"))
+		if err != nil {
+			return "", httperr.Server(err)
+		}
+
+		for _, ps := range pss {
+			client, err := ps.Docker()
+			if err != nil {
+				return "", httperr.Server(err)
+			}
+
+			res, err := client.ListContainers(docker.ListContainersOptions{
+				All: true,
+				Filters: map[string][]string{
+					"name": []string{fmt.Sprintf("build-%s", build)},
+				},
+			})
+
+			if len(res) > 0 {
+				return fmt.Sprintf("http://%s:2376", ps.Host), nil
+			}
+		}
+
+		time.Sleep(2 * time.Second)
+	}
+
+	return "", fmt.Errorf("could not find build host")
+}
+
 func keepAlive(ws *websocket.Conn, quit chan bool) {
 	c := time.Tick(5 * time.Second)
 	b := []byte{}

api/models/release.go

@@ -4,6 +4,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"html/template"
+	"math/rand"
 	"os"
 	"os/exec"
 	"regexp"
@@ -271,7 +272,7 @@ func (r *Release) Promote() error {
 			switch app.Parameters[protoParam] {
 			case "https", "tls":
 				if app.Parameters[certParam] == "" {
-					name := fmt.Sprintf("cert-%s-%d", os.Getenv("RACK"), time.Now().Unix())
+					name := fmt.Sprintf("cert-%s-%d-%05d", os.Getenv("RACK"), time.Now().Unix(), rand.Intn(100000))
 
 					body, key, err := GenerateSelfSignedCertificate("*.*.elb.amazonaws.com")
 					if err != nil {