From e660d8b8b7cc76438a3e145bcb27ede4c935a0f4 Mon Sep 17 00:00:00 2001
From: James Ravn <james.ravn@sky.uk>
Date: Thu, 18 Jan 2018 18:05:27 +0000
Subject: [PATCH] Retry docker status on startup

For https://github.com/google/cadvisor/issues/1866.
---
 container/docker/docker.go | 14 +++++++++++---
 manager/manager.go         | 37 ++++++++++++++++++++++++++++++++-----
 2 files changed, 43 insertions(+), 8 deletions(-)

diff --git a/container/docker/docker.go b/container/docker/docker.go
index b0ed227d..f1cda2be 100644
--- a/container/docker/docker.go
+++ b/container/docker/docker.go
@@ -29,19 +29,27 @@ import (
 	"github.com/google/cadvisor/machine"
 )
 
-const defaultTimeout = time.Second * 5
+var dockerTimeout = 10 * time.Second
 
 func defaultContext() context.Context {
-	ctx, _ := context.WithTimeout(context.Background(), defaultTimeout)
+	ctx, _ := context.WithTimeout(context.Background(), dockerTimeout)
 	return ctx
 }
 
+func SetTimeout(timeout time.Duration) {
+	dockerTimeout = timeout
+}
+
 func Status() (v1.DockerStatus, error) {
+	return StatusWithContext(defaultContext())
+}
+
+func StatusWithContext(ctx context.Context) (v1.DockerStatus, error) {
 	client, err := Client()
 	if err != nil {
 		return v1.DockerStatus{}, fmt.Errorf("unable to communicate with docker daemon: %v", err)
 	}
-	dockerInfo, err := client.Info(defaultContext())
+	dockerInfo, err := client.Info(ctx)
 	if err != nil {
 		return v1.DockerStatus{}, err
 	}
diff --git a/manager/manager.go b/manager/manager.go
index 08955833..10a725f7 100644
--- a/manager/manager.go
+++ b/manager/manager.go
@@ -50,6 +50,7 @@ import (
 
 	"github.com/golang/glog"
 	"github.com/opencontainers/runc/libcontainer/cgroups"
+	"golang.org/x/net/context"
 	"k8s.io/utils/clock"
 )
 
@@ -59,6 +60,8 @@ var eventStorageAgeLimit = flag.String("event_storage_age_limit", "default=24h",
 var eventStorageEventLimit = flag.String("event_storage_event_limit", "default=100000", "Max number of events to store (per type). Value is a comma separated list of key values, where the keys are event types (e.g.: creation, oom) or \"default\" and the value is an integer. Default is applied to all non-specified event types")
 var applicationMetricsCountLimit = flag.Int("application_metrics_count_limit", 100, "Max number of application metrics to store (per container)")
 
+const dockerClientTimeout = 10 * time.Second
+
 // The Manager interface defines operations for starting a manager and getting
 // container and machine information.
 type Manager interface {
@@ -154,11 +157,10 @@ func New(memoryCache *memory.InMemoryCache, sysfs sysfs.SysFs, maxHousekeepingIn
 		dockerStatus info.DockerStatus
 		rktPath      string
 	)
-	if tempDockerStatus, err := docker.Status(); err != nil {
-		glog.V(5).Infof("Docker not connected: %v", err)
-	} else {
-		dockerStatus = tempDockerStatus
-	}
+	docker.SetTimeout(dockerClientTimeout)
+	// Try to connect to docker indefinitely on startup.
+	dockerStatus = retryDockerStatus()
+
 	if tmpRktPath, err := rkt.RktPath(); err != nil {
 		glog.V(5).Infof("Rkt not connected: %v", err)
 	} else {
@@ -234,6 +236,31 @@ func New(memoryCache *memory.InMemoryCache, sysfs sysfs.SysFs, maxHousekeepingIn
 	return newManager, nil
 }
 
+func retryDockerStatus() info.DockerStatus {
+	startupTimeout := dockerClientTimeout
+	maxTimeout := 4 * startupTimeout
+	for {
+		ctx, _ := context.WithTimeout(context.Background(), startupTimeout)
+		dockerStatus, err := docker.StatusWithContext(ctx)
+		if err != nil {
+			return dockerStatus
+		}
+
+		switch err {
+		case context.DeadlineExceeded:
+			glog.Warningf("Timeout trying to communicate with docker during initialization, will retry")
+		default:
+			glog.V(5).Infof("Docker not connected: %v", err)
+			return info.DockerStatus{}
+		}
+
+		startupTimeout = 2 * startupTimeout
+		if startupTimeout > maxTimeout {
+			startupTimeout = maxTimeout
+		}
+	}
+}
+
 // A namespaced container name.
 type namespacedContainerName struct {
 	// The namespace of the container. Can be empty for the root namespace.