From db4201534248ecda7b43f9bb763bbbf82048a807 Mon Sep 17 00:00:00 2001
From: James Bardin <j.bardin@gmail.com>
Date: Wed, 9 Aug 2017 15:05:51 -0400
Subject: [PATCH] backoff retries in remote-exec provisioner

Add a simple backoff to the remote-exec retryFunc.
Backoff between tries, up to a 10s max.
---
 .../remote-exec/resource_provisioner.go       | 22 +++++++++++--
 .../remote-exec/resource_provisioner_test.go  | 33 +++++++++++++++++++
 2 files changed, 52 insertions(+), 3 deletions(-)

diff --git a/builtin/provisioners/remote-exec/resource_provisioner.go b/builtin/provisioners/remote-exec/resource_provisioner.go
index 7dd86daf0..ba811dafe 100644
--- a/builtin/provisioners/remote-exec/resource_provisioner.go
+++ b/builtin/provisioners/remote-exec/resource_provisioner.go
@@ -19,6 +19,10 @@ import (
 	"github.com/mitchellh/go-linereader"
 )
 
+// maxBackoffDealy is the maximum delay between retry attempts
+var maxBackoffDelay = 10 * time.Second
+var initialBackoffDelay = time.Second
+
 func Provisioner() terraform.ResourceProvisioner {
 	return &schema.Provisioner{
 		Schema: map[string]*schema.Schema{
@@ -246,7 +250,6 @@ func copyOutput(
 }
 
 // retryFunc is used to retry a function for a given duration
-// TODO: this should probably backoff too
 func retryFunc(ctx context.Context, timeout time.Duration, f func() error) error {
 	// Build a new context with the timeout
 	ctx, done := context.WithTimeout(ctx, timeout)
@@ -263,12 +266,13 @@ func retryFunc(ctx context.Context, timeout time.Duration, f func() error) error
 	go func() {
 		defer close(doneCh)
 
+		delay := time.Duration(0)
 		for {
 			// If our context ended, we want to exit right away.
 			select {
 			case <-ctx.Done():
 				return
-			default:
+			case <-time.After(delay):
 			}
 
 			// Try the function call
@@ -279,7 +283,19 @@ func retryFunc(ctx context.Context, timeout time.Duration, f func() error) error
 				return
 			}
 
-			log.Printf("Retryable error: %v", err)
+			log.Printf("[WARN] retryable error: %v", err)
+
+			delay *= 2
+
+			if delay == 0 {
+				delay = initialBackoffDelay
+			}
+
+			if delay > maxBackoffDelay {
+				delay = maxBackoffDelay
+			}
+
+			log.Printf("[INFO] sleeping for %s", delay)
 		}
 	}()
 
diff --git a/builtin/provisioners/remote-exec/resource_provisioner_test.go b/builtin/provisioners/remote-exec/resource_provisioner_test.go
index 67faf1fe4..8c447788d 100644
--- a/builtin/provisioners/remote-exec/resource_provisioner_test.go
+++ b/builtin/provisioners/remote-exec/resource_provisioner_test.go
@@ -211,6 +211,16 @@ func TestResourceProvider_CollectScripts_scriptsEmpty(t *testing.T) {
 }
 
 func TestRetryFunc(t *testing.T) {
+	origMax := maxBackoffDelay
+	maxBackoffDelay = time.Second
+	origStart := initialBackoffDelay
+	initialBackoffDelay = 10 * time.Millisecond
+
+	defer func() {
+		maxBackoffDelay = origMax
+		initialBackoffDelay = origStart
+	}()
+
 	// succeed on the third try
 	errs := []error{io.EOF, &net.OpError{Err: errors.New("ERROR")}, nil}
 	count := 0
@@ -235,6 +245,29 @@ func TestRetryFunc(t *testing.T) {
 	}
 }
 
+func TestRetryFuncBackoff(t *testing.T) {
+	origMax := maxBackoffDelay
+	maxBackoffDelay = time.Second
+	origStart := initialBackoffDelay
+	initialBackoffDelay = 100 * time.Millisecond
+
+	defer func() {
+		maxBackoffDelay = origMax
+		initialBackoffDelay = origStart
+	}()
+
+	count := 0
+
+	retryFunc(context.Background(), time.Second, func() error {
+		count++
+		return io.EOF
+	})
+
+	if count > 4 {
+		t.Fatalf("retry func failed to backoff. called %d times", count)
+	}
+}
+
 func testConfig(t *testing.T, c map[string]interface{}) *terraform.ResourceConfig {
 	r, err := config.NewRawConfig(c)
 	if err != nil {