As discussed on the issue, remove the hard-coded delay on startup in

favor of attempting to detect if the initial container ever enters
running state, and erroring out if not. It will re-check the container
once every 500ms for 15 seconds total; future work could make that
configurable.

builtin/providers/docker/resource_docker_container_funcs.go

@@ -11,6 +11,10 @@ import (
 	"github.com/hashicorp/terraform/helper/schema"
 )
 
+var (
+	creationTime time.Time
+)
+
 func resourceDockerContainerCreate(d *schema.ResourceData, meta interface{}) error {
 	var err error
 	client := meta.(*dc.Client)
@@ -20,15 +24,12 @@ func resourceDockerContainerCreate(d *schema.ResourceData, meta interface{}) err
 		return err
 	}
 
-	delayStart := false
-
 	image := d.Get("image").(string)
 	if _, ok := data.DockerImages[image]; !ok {
 		if _, ok := data.DockerImages[image+":latest"]; !ok {
 			return fmt.Errorf("Unable to find image %s", image)
-		} else {
-			image = image + ":latest"
 		}
+		image = image + ":latest"
 	}
 
 	// The awesome, wonderful, splendiferous, sensical
@@ -109,15 +110,9 @@ func resourceDockerContainerCreate(d *schema.ResourceData, meta interface{}) err
 
 	if v, ok := d.GetOk("links"); ok {
 		hostConfig.Links = stringSetToStringSlice(v.(*schema.Set))
-		delayStart = true
-	}
-
-	// For instance, Docker will fail to start conatiners with links
-	// to other containers if the containers haven't started yet
-	if delayStart {
-		time.Sleep(3 * time.Second)
 	}
 
+	creationTime = time.Now()
 	if err := client.StartContainer(retContainer.ID, hostConfig); err != nil {
 		return fmt.Errorf("Unable to start container: %s", err)
 	}
@@ -132,21 +127,49 @@ func resourceDockerContainerRead(d *schema.ResourceData, meta interface{}) error
 	if err != nil {
 		return err
 	}
-
 	if apiContainer == nil {
 		// This container doesn't exist anymore
 		d.SetId("")
-
 		return nil
 	}
 
-	container, err := client.InspectContainer(apiContainer.ID)
-	if err != nil {
-		return fmt.Errorf("Error inspecting container %s: %s", apiContainer.ID, err)
+	var container *dc.Container
+
+	loops := 1 // if it hasn't just been created, don't delay
+	if !creationTime.IsZero() {
+		loops = 30 // with 500ms spacing, 15 seconds; ought to be plenty
+	}
+	sleepTime := 500 * time.Millisecond
+
+	for i := loops; i > 0; i-- {
+		container, err = client.InspectContainer(apiContainer.ID)
+		if err != nil {
+			return fmt.Errorf("Error inspecting container %s: %s", apiContainer.ID, err)
+		}
+
+		if container.State.Running ||
+			(!container.State.Running && !d.Get("must_run").(bool)) {
+			break
+		}
+
+		if creationTime.IsZero() { // We didn't just create it, so don't wait around
+			return resourceDockerContainerDelete(d, meta)
+		}
+
+		if container.State.FinishedAt.After(creationTime) {
+			// It exited immediately, so error out so dependent containers
+			// aren't started
+			resourceDockerContainerDelete(d, meta)
+			return fmt.Errorf("Container %s exited after creation, error was: %s", apiContainer.ID, container.State.Error)
+		}
+
+		time.Sleep(sleepTime)
 	}
 
-	if d.Get("must_run").(bool) && !container.State.Running {
-		return resourceDockerContainerDelete(d, meta)
+	// Handle the case of the for loop above running its course
+	if !container.State.Running && d.Get("must_run").(bool) {
+		resourceDockerContainerDelete(d, meta)
+		return fmt.Errorf("Container %s failed to be in running state", apiContainer.ID)
 	}
 
 	// Read Network Settings