move retrier to separate package and reuse code

Author: QxBytes

dhcp/dhcp_windows.go

@@ -7,6 +7,7 @@ import (
 	"time"
 
 	"github.com/Azure/azure-container-networking/cni/log"
+	"github.com/Azure/azure-container-networking/retry"
 	"github.com/pkg/errors"
 	"go.uber.org/zap"
 	"golang.org/x/sys/windows"
@@ -18,8 +19,9 @@ const (
 	addIPAddressDelay    = 4 * time.Second
 	deleteIPAddressDelay = 2 * time.Second
 	returnDelay          = 8 * time.Second // time to wait before returning from DiscoverRequest
-
-	socketTimeoutMillis = 1000
+	retryCount           = 5
+	retryDelayMillis     = 500
+	socketTimeoutMillis  = 1000
 )
 
 var (
@@ -160,7 +162,11 @@ func (c *DHCP) DiscoverRequest(ctx context.Context, macAddress net.HardwareAddr,
 		return errors.Wrap(err, "failed to create socket")
 	}
 
-	_, err = sock.Write(bytesToSend)
+	// retry sending the packet until it succeeds
+	err = retry.Do(func() error {
+		_, sockErr := sock.Write(bytesToSend)
+		return sockErr
+	}, retryCount, retryDelayMillis)
 	if err != nil {
 		return errors.Wrap(err, "failed to write to dhcp socket")
 	}

network/transparent_vlan_endpointclient_linux.go

@@ -4,7 +4,6 @@ import (
 	"fmt"
 	"net"
 	"strings"
-	"time"
 
 	"github.com/Azure/azure-container-networking/iptables"
 	"github.com/Azure/azure-container-networking/netio"
@@ -13,6 +12,7 @@ import (
 	"github.com/Azure/azure-container-networking/network/networkutils"
 	"github.com/Azure/azure-container-networking/network/snat"
 	"github.com/Azure/azure-container-networking/platform"
+	"github.com/Azure/azure-container-networking/retry"
 	"github.com/pkg/errors"
 	vishnetlink "github.com/vishvananda/netlink"
 	"go.uber.org/zap"
@@ -192,7 +192,7 @@ func (client *TransparentVlanEndpointClient) setLinkNetNSAndConfirm(name string,
 	}
 
 	// confirm veth was moved successfully
-	err = RunWithRetries(func() error {
+	err = retry.Do(func() error {
 		// retry checking in the namespace if the interface is not detected
 		return ExecuteInNS(client.nsClient, client.vnetNSName, func() error {
 			_, ifDetectedErr := client.netioshim.GetNetworkInterfaceByName(client.vlanIfName)
@@ -220,7 +220,7 @@ func (client *TransparentVlanEndpointClient) PopulateVM(epInfo *EndpointInfo) er
 		// We assume the only possible error is that the namespace doesn't exist
 		logger.Info("No existing NS detected. Creating the vnet namespace and switching to it", zap.String("message", existingErr.Error()))
 
-		err = RunWithRetries(func() error {
+		err = retry.Do(func() error {
 			return client.createNetworkNamespace(vmNS)
 		}, numRetries, sleepInMs)
 		if err != nil {
@@ -279,7 +279,7 @@ func (client *TransparentVlanEndpointClient) PopulateVM(epInfo *EndpointInfo) er
 		}()
 
 		// sometimes there is slight delay in interface creation. check if it exists
-		err = RunWithRetries(func() error {
+		err = retry.Do(func() error {
 			_, err = client.netioshim.GetNetworkInterfaceByName(client.vlanIfName)
 			return errors.Wrap(err, "failed to get vlan veth")
 		}, numRetries, sleepInMs)
@@ -316,7 +316,7 @@ func (client *TransparentVlanEndpointClient) PopulateVM(epInfo *EndpointInfo) er
 	}
 
 	// Ensure vnet veth is created, as there may be a slight delay
-	err = RunWithRetries(func() error {
+	err = retry.Do(func() error {
 		_, getErr := client.netioshim.GetNetworkInterfaceByName(client.vnetVethName)
 		return errors.Wrap(getErr, "failed to get vnet veth")
 	}, numRetries, sleepInMs)
@@ -326,7 +326,7 @@ func (client *TransparentVlanEndpointClient) PopulateVM(epInfo *EndpointInfo) er
 
 	// Ensure container veth is created, as there may be a slight delay
 	var containerIf *net.Interface
-	err = RunWithRetries(func() error {
+	err = retry.Do(func() error {
 		var getErr error
 		containerIf, getErr = client.netioshim.GetNetworkInterfaceByName(client.containerVethName)
 		return errors.Wrap(getErr, "failed to get container veth")
@@ -712,16 +712,3 @@ func ExecuteInNS(nsc NamespaceClientInterface, nsName string, f func() error) er
 	}()
 	return f()
 }
-
-func RunWithRetries(f func() error, maxRuns, sleepMs int) error {
-	var err error
-	for i := 0; i < maxRuns; i++ {
-		err = f()
-		if err == nil {
-			break
-		}
-		logger.Info("Retrying after delay...", zap.String("error", err.Error()), zap.Int("retry", i), zap.Int("sleepMs", sleepMs))
-		time.Sleep(time.Duration(sleepMs) * time.Millisecond)
-	}
-	return err
-}

network/transparent_vlan_endpointclient_linux_test.go

@@ -908,58 +908,3 @@ func TestTransparentVlanConfigureContainerInterfacesAndRoutes(t *testing.T) {
 		})
 	}
 }
-
-func createFunctionWithFailurePattern(errorPattern []error) func() error {
-	s := 0
-	return func() error {
-		if s >= len(errorPattern) {
-			return nil
-		}
-		result := errorPattern[s]
-		s++
-		return result
-	}
-}
-
-func TestRunWithRetries(t *testing.T) {
-	errMock := errors.New("mock error")
-	runs := 4
-
-	tests := []struct {
-		name    string
-		wantErr bool
-		f       func() error
-	}{
-		{
-			name:    "Succeed on first try",
-			f:       createFunctionWithFailurePattern([]error{}),
-			wantErr: false,
-		},
-		{
-			name:    "Succeed on first try do not check again",
-			f:       createFunctionWithFailurePattern([]error{nil, errMock, errMock, errMock}),
-			wantErr: false,
-		},
-		{
-			name:    "Succeed on last try",
-			f:       createFunctionWithFailurePattern([]error{errMock, errMock, errMock, nil, errMock}),
-			wantErr: false,
-		},
-		{
-			name:    "Fail after too many attempts",
-			f:       createFunctionWithFailurePattern([]error{errMock, errMock, errMock, errMock, nil, nil}),
-			wantErr: true,
-		},
-	}
-	for _, tt := range tests {
-		tt := tt
-		t.Run(tt.name, func(t *testing.T) {
-			err := RunWithRetries(tt.f, runs, 100)
-			if tt.wantErr {
-				require.Error(t, err)
-			} else {
-				require.NoError(t, err)
-			}
-		})
-	}
-}