Change default detach retry limit to 100 & enable unlimit option

cmd/disk/main.go

@@ -18,14 +18,15 @@ package main
 
 import (
 	"flag"
+	"math/rand"
+	"os"
+	"time"
+
 	"github.com/yunify/qingcloud-csi/pkg/cloud"
 	"github.com/yunify/qingcloud-csi/pkg/common"
 	"github.com/yunify/qingcloud-csi/pkg/disk/driver"
 	"github.com/yunify/qingcloud-csi/pkg/disk/rpcserver"
 	"k8s.io/klog"
-	"math/rand"
-	"os"
-	"time"
 )
 
 const (
@@ -35,13 +36,13 @@ const (
 )
 
 var (
-	configPath       = flag.String("config", defaultConfigPath, "server config file path")
-	driverName       = flag.String("drivername", defaultProvisionName, "name of the driver")
-	endpoint         = flag.String("endpoint", "unix://tmp/csi.sock", "CSI endpoint")
-	maxVolume        = flag.Int64("maxvolume", 10, "Maximum number of volumes that controller can publish to the node.")
-	nodeId           = flag.String("nodeid", "", "If driver cannot get instance ID from /etc/qingcloud/instance-id, we would use this flag.")
-	retryIntervalMax = flag.Duration("retry-interval-max", 2*time.Minute, "Maximum retry interval of failed deletion.")
-	retryTimesMax    = flag.Int("retry-times-max", 10, "Maximum retry times of failed detach volume.")
+	configPath          = flag.String("config", defaultConfigPath, "server config file path")
+	driverName          = flag.String("drivername", defaultProvisionName, "name of the driver")
+	endpoint            = flag.String("endpoint", "unix://tmp/csi.sock", "CSI endpoint")
+	maxVolume           = flag.Int64("maxvolume", 10, "Maximum number of volumes that controller can publish to the node.")
+	nodeId              = flag.String("nodeid", "", "If driver cannot get instance ID from /etc/qingcloud/instance-id, we would use this flag.")
+	retryIntervalMax    = flag.Duration("retry-interval-max", 2*time.Minute, "Maximum retry interval of failed deletion.")
+	retryDetachTimesMax = flag.Int("retry-detach-times-max", 100, "Maximum retry times of failed detach volume. Set to 0 to disable the limit.")
 )
 
 func main() {
@@ -88,5 +89,5 @@ func handle() {
 	mounter := common.NewSafeMounter()
 	driver := driver.GetDiskDriver()
 	driver.InitDiskDriver(diskDriverInput)
-	rpcserver.Run(driver, cloud, mounter, *endpoint, rt, *retryTimesMax)
+	rpcserver.Run(driver, cloud, mounter, *endpoint, rt, *retryDetachTimesMax)
 }

pkg/common/common.go

@@ -19,10 +19,11 @@ package common
 import (
 	"fmt"
 	"hash/fnv"
-	"k8s.io/apimachinery/pkg/util/wait"
-	"k8s.io/client-go/util/retry"
 	"sync"
 	"time"
+
+	"k8s.io/apimachinery/pkg/util/wait"
+	"k8s.io/client-go/util/retry"
 )
 
 // EntryFunction print timestamps
@@ -84,7 +85,7 @@ func (r *retryLimiter) Add(id string) {
 func (r *retryLimiter) Try(id string) bool {
 	r.mux.RLock()
 	defer r.mux.RUnlock()
-	return r.record[id] <= r.maxRetry
+	return r.maxRetry == 0 || r.record[id] <= r.maxRetry
 }
 
 func (r *retryLimiter) GetMaxRetryTimes() int {

pkg/common/common_test.go

@@ -108,3 +108,18 @@ func TestRetryLimiter(t *testing.T) {
 		t.Errorf("expect false but actually true")
 	}
 }
+
+func TestUnlimitedRetryLimiter(t *testing.T) {
+	r := NewRetryLimiter(0)
+	key := "key"
+	for i := 1; i <= 5; i++ {
+		r.Add(key)
+		current := r.GetCurrentRetryTimes(key)
+		if current != i {
+			t.Errorf("expect %d but actually %d", i, current)
+		}
+		if r.Try(key) != true {
+			t.Errorf("expect true but actually false")
+		}
+	}
+}