GCLOUD: Re-try on 502 errors

providers/gcloud/gcloudProvider.go

@@ -5,10 +5,8 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
-	"log"
 	"regexp"
 	"strings"
-	"time"
 
 	"github.com/StackExchange/dnscontrol/v4/models"
 	"github.com/StackExchange/dnscontrol/v4/pkg/diff2"
@@ -488,70 +486,3 @@ func (g *gcloudProvider) EnsureZoneExists(domain string) error {
 	g.zones[domain+"."], err = g.client.ManagedZones.Create(g.project, mz).Do()
 	return err
 }
-
-const (
-	initialBackoff = time.Second * 10 // First delay duration
-	maxBackoff     = time.Minute * 3  // Maximum backoff delay
-)
-
-// backoff is the amount of time to sleep if a 429 or 504 is received.
-// It is doubled after each use.
-var (
-	backoff    = initialBackoff
-	backoff404 = false // Set if the last call requested a retry of a 404
-)
-
-func retryNeeded(resp *googleapi.ServerResponse, err error) bool {
-	if err != nil {
-		return false // Not an error.
-	}
-	serr, ok := err.(*googleapi.Error)
-	if !ok {
-		return false // Not a google error.
-	}
-	if serr.Code == 200 {
-		backoff = initialBackoff // Reset
-		return false             // Success! No need to retry.
-	}
-
-	if serr.Code == 404 {
-		// serr.Code == 404 happens occasionally when GCLOUD hasn't
-		// finished updating the database yet.  We pause and retry
-		// exactly once. There should be a better way to do this, such as
-		// a callback that would tell us a transaction is complete.
-		if backoff404 {
-			backoff404 = false
-			return false // Give up. We've done this already.
-		}
-		log.Printf("Special 404 pause-and-retry for GCLOUD: Pausing %s\n", backoff)
-		time.Sleep(backoff)
-		backoff404 = true
-		return true // Request a retry.
-	}
-	backoff404 = false
-
-	if serr.Code != 429 && serr.Code != 502 && serr.Code != 503 {
-		return false // Not an error that permits retrying.
-	}
-
-	// TODO(tlim): In theory, resp.Header has a header that says how
-	// long to wait but I haven't been able to capture that header in
-	// the wild. If you get these "RUNCHANGE HEAD" messages, please
-	// file a bug with the contents!
-
-	if resp != nil {
-		log.Printf("NOTE: If you see this message, please file a bug with the output below:\n")
-		log.Printf("RUNCHANGE CODE = %+v\n", resp.HTTPStatusCode)
-		log.Printf("RUNCHANGE HEAD = %+v\n", resp.Header)
-	}
-
-	// a simple exponential back-off
-	log.Printf("Pausing due to ratelimit: %v seconds\n", backoff)
-	time.Sleep(backoff)
-	backoff = backoff + (backoff / 2)
-	if backoff > maxBackoff {
-		backoff = maxBackoff
-	}
-
-	return true // Request the API call be re-tried.
-}

providers/gcloud/retry.go

@@ -0,0 +1,94 @@
+package gcloud
+
+import (
+	"log"
+	"time"
+
+	"google.golang.org/api/googleapi"
+)
+
+const (
+	initialBackoff = time.Second * 10 // First delay duration
+	maxBackoff     = time.Minute * 3  // Maximum backoff delay
+)
+
+// backoff is the amount of time to sleep if a 429 or 504 is received.
+// It is doubled after each use.
+var (
+	backoff    = initialBackoff
+	backoff404 = false // Set if the last call requested a retry of a 404
+	backoff502 = false // Set if the last call requested a retry of a 502
+)
+
+func retryNeeded(resp *googleapi.ServerResponse, err error) bool {
+	if err == nil {
+		return false // Not an error.
+	}
+
+	serr, ok := err.(*googleapi.Error)
+	if !ok {
+		return false // Not a google error.
+	}
+
+	if serr.Code == 200 {
+		backoff = initialBackoff // Reset
+		return false             // Success! No need to retry.
+	}
+
+	if serr.Code == 404 {
+		// serr.Code == 404 happens occasionally when GCLOUD hasn't
+		// finished updating the database yet.  We pause and retry
+		// exactly once. There should be a better way to do this, such as
+		// a callback that would tell us a transaction is complete.
+		if backoff404 {
+			backoff404 = false
+			return false // Give up. We've done this already.
+		}
+		log.Printf("Special 404 pause-and-retry for GCLOUD: Pausing %s\n", backoff)
+		time.Sleep(backoff)
+		backoff404 = true
+		return true // Request a retry.
+	}
+	backoff404 = false
+
+	if serr.Code == 502 {
+		// serr.Code == 502 happens occasionally when "The server
+		// encountered a temporary error and could not complete your
+		// request. Please try again in 30 seconds.  That’s all we know."
+		// We pause and retry exactly once.
+		if backoff502 {
+			backoff502 = false
+			return false // Give up. We've done this already.
+		}
+		log.Printf("Special 502 pause-and-retry for GCLOUD: Pausing %s\n", backoff)
+		time.Sleep(31 * time.Second)
+		backoff502 = true
+		return true // Request a retry.
+	}
+	backoff502 = false
+
+	if serr.Code != 429 && serr.Code != 503 {
+		return false // Not an error that permits retrying.
+	}
+
+	// TODO(tlim): In theory, resp.Header has a header that says how
+	// long to wait but I haven't been able to capture that header in
+	// the wild. If you get these "RUNCHANGE HEAD" messages, please
+	// file a bug with the contents!
+
+	if resp != nil {
+		log.Printf("NOTE: If you see this message, please file a bug with the output below:\n")
+		log.Printf("RUNCHANGE CODE = %+v\n", resp.HTTPStatusCode)
+		log.Printf("RUNCHANGE HEAD = %+v\n", resp.Header)
+	}
+
+	// a simple exponential back-off
+	log.Printf("Pausing due to ratelimit: %v seconds\n", backoff)
+	time.Sleep(backoff)
+	backoff = backoff + (backoff / 2)
+	if backoff > maxBackoff {
+		backoff = maxBackoff
+	}
+
+	return true // Request the API call be re-tried.
+}