fix(container): preserve IP allocation across container operations (#13738)

2026-09-22 00:00:50 +00:00 · 2026-09-08 14:20:03 +08:00
parent ce74d96617
commit 8eac9a1808
3 changed files with 52 additions and 101 deletions
@@ -582,6 +582,10 @@ func stepRecreateContainer(recoverCtx *containerRecoverContext, taskItem *task.T
 	if config.Image == "" {
 		return fmt.Errorf("container image not found in backup file")
 	}
+	networkConf, extraNetworks := buildContainerRecoverNetworkConfig(recoverCtx.inspectInfo.NetworkSettings, hostConfig)
+	if err := normalizeContainerEndpointSettings(ctx, recoverCtx.client, networkConf, extraNetworks); err != nil {
+		return err
+	}
 	if !checkImageExist(recoverCtx.client, config.Image) {
 		if err := pullImages(taskItem, recoverCtx.client, config.Image); err != nil {
 			return err
@@ -596,7 +600,7 @@ func stepRecreateContainer(recoverCtx *containerRecoverContext, taskItem *task.T
 		return err
 	}

-	createRes, err := createContainerWithOldNetworks(ctx, recoverCtx.client, config, hostConfig, recoverCtx.inspectInfo.NetworkSettings, recoverCtx.targetName)
+	createRes, err := createContainerWithNetworks(ctx, recoverCtx.client, config, hostConfig, networkConf, extraNetworks, recoverCtx.targetName)
 	if err != nil {
 		return err
 	}
@@ -604,7 +608,7 @@ func stepRecreateContainer(recoverCtx *containerRecoverContext, taskItem *task.T
 	return nil
 }

-func normalizeContainerEndpointSettings(ctx context.Context, cli *client.Client, primary *network.NetworkingConfig, extras map[string]*network.EndpointSettings) {
+func normalizeContainerEndpointSettings(ctx context.Context, cli *client.Client, primary *network.NetworkingConfig, extras map[string]*network.EndpointSettings) error {
 	if cli.NewVersionError(ctx, "1.44", "specify mac-address per network") != nil {
 		removeEndpointMacAddresses(primary, extras)
 	}
@@ -619,11 +623,14 @@ func normalizeContainerEndpointSettings(ctx context.Context, cli *client.Client,
 			}
 			info, err := cli.NetworkInspect(ctx, netName, network.InspectOptions{})
 			if err != nil {
-				continue
+				return fmt.Errorf("inspect network %s failed: %w", netName, err)
+			}
+			if err := validateContainerEndpointStaticIP(netName, info, endpoint); err != nil {
+				return err
 			}
-			removeUnsupportedEndpointStaticIP(netName, info, endpoint)
 		}
 	}
+	return nil
 }

 func removeEndpointMacAddresses(primary *network.NetworkingConfig, extras map[string]*network.EndpointSettings) {
@@ -641,24 +648,28 @@ func removeEndpointMacAddresses(primary *network.NetworkingConfig, extras map[st
 	}
 }

-func removeUnsupportedEndpointStaticIP(netName string, info network.Inspect, endpoint *network.EndpointSettings) {
+func validateContainerEndpointStaticIP(netName string, info network.Inspect, endpoint *network.EndpointSettings) error {
 	if endpoint == nil || endpoint.IPAMConfig == nil {
-		return
+		return nil
 	}
-	if isDefaultBridgeNetwork(netName, info) {
-		endpoint.IPAMConfig = nil
-		return
+	ipam := endpoint.IPAMConfig
+	if err := ipam.Validate(); err != nil {
+		return fmt.Errorf("invalid IP configuration for network %s: %w", netName, err)
+	}
+	if ipam.IPv4Address == "" && ipam.IPv6Address == "" {
+		return nil
+	}
+	if netName == "host" || netName == "none" || isDefaultBridgeNetwork(netName, info) {
+		return fmt.Errorf("network %s does not support static IP configuration", netName)
 	}

-	if endpoint.IPAMConfig.IPv4Address != "" && !networkSupportsStaticIP(info, endpoint.IPAMConfig.IPv4Address, false) {
-		endpoint.IPAMConfig.IPv4Address = ""
+	if ipam.IPv4Address != "" && !networkSupportsStaticIP(info, ipam.IPv4Address, false) {
+		return fmt.Errorf("static IPv4 address %s is not in a configured subnet of network %s", ipam.IPv4Address, netName)
 	}
-	if endpoint.IPAMConfig.IPv6Address != "" && !networkSupportsStaticIP(info, endpoint.IPAMConfig.IPv6Address, true) {
-		endpoint.IPAMConfig.IPv6Address = ""
-	}
-	if endpoint.IPAMConfig.IPv4Address == "" && endpoint.IPAMConfig.IPv6Address == "" && len(endpoint.IPAMConfig.LinkLocalIPs) == 0 {
-		endpoint.IPAMConfig = nil
+	if ipam.IPv6Address != "" && !networkSupportsStaticIP(info, ipam.IPv6Address, true) {
+		return fmt.Errorf("static IPv6 address %s is not in a configured subnet of network %s", ipam.IPv6Address, netName)
 	}
+	return nil
 }

 func isDefaultBridgeNetwork(netName string, info network.Inspect) bool {
@@ -673,6 +684,7 @@ func networkSupportsStaticIP(info network.Inspect, ip string, isIPv6 bool) bool
 	if err != nil {
 		return false
 	}
+	addr = addr.Unmap()
 	if addr.Is6() != isIPv6 {
 		return false
 	}
@@ -826,39 +838,6 @@ func buildContainerRecoverNetworkConfig(networkSettings *container.NetworkSettin
 	return config, extraNetworks
 }

-const unsupportedUserSpecifiedIPAddress = "user specified IP address is supported only when connecting to networks with user configured subnets"
-
-func clearUnsupportedDynamicEndpointIPAM(err error, endpoints map[string]*network.EndpointSettings, networkSettings *container.NetworkSettings) bool {
-	if err == nil || !strings.Contains(err.Error(), unsupportedUserSpecifiedIPAddress) {
-		return false
-	}
-	for name, endpoint := range endpoints {
-		if !isDynamicContainerNetwork(networkSettings, name) || endpoint == nil || endpoint.IPAMConfig == nil {
-			continue
-		}
-		if strings.Contains(err.Error(), "network "+name+":") {
-			endpoint.IPAMConfig = nil
-			return true
-		}
-	}
-	cleared := false
-	for name, endpoint := range endpoints {
-		if isDynamicContainerNetwork(networkSettings, name) && endpoint != nil && endpoint.IPAMConfig != nil {
-			endpoint.IPAMConfig = nil
-			cleared = true
-		}
-	}
-	return cleared
-}
-
-func isDynamicContainerNetwork(networkSettings *container.NetworkSettings, name string) bool {
-	if networkSettings == nil || name == "bridge" {
-		return false
-	}
-	endpoint := networkSettings.Networks[name]
-	return endpoint != nil && endpoint.IPAMConfig == nil && (endpoint.IPAddress != "" || endpoint.GlobalIPv6Address != "")
-}
-
 func cloneContainerConfig(config *container.Config) *container.Config {
 	if config == nil {
 		return &container.Config{}
@@ -536,7 +536,9 @@ func (u *ContainerService) ContainerCreate(req dto.ContainerOperate, inThread bo
 		if err != nil {
 			return err
 		}
-		normalizeContainerEndpointSettings(ctx, client, networkConf, nil)
+		if err := normalizeContainerEndpointSettings(ctx, client, networkConf, nil); err != nil {
+			return err
+		}
 		con, err := client.ContainerCreate(ctx, config, hostConf, networkConf, &v1.Platform{}, req.Name)
 		if err != nil {
 			taskItem.Log(i18n.GetMsgByKey("ContainerCreateFailed"))
@@ -646,14 +648,9 @@ func loadContainerNetworkInfo(name string, endpoint *network.EndpointSettings) d
 	if endpoint.IPAMConfig != nil {
 		item.LinkLocalIPs = append([]string(nil), endpoint.IPAMConfig.LinkLocalIPs...)
 	}
-	if name != "bridge" {
-		if endpoint.IPAMConfig != nil {
-			item.Ipv4 = endpoint.IPAMConfig.IPv4Address
-			item.Ipv6 = endpoint.IPAMConfig.IPv6Address
-		} else {
-			item.Ipv4 = endpoint.IPAddress
-			item.Ipv6 = endpoint.GlobalIPv6Address
-		}
+	if name != "bridge" && endpoint.IPAMConfig != nil {
+		item.Ipv4 = endpoint.IPAMConfig.IPv4Address
+		item.Ipv6 = endpoint.IPAMConfig.IPv6Address
 	}
 	return item
 }
@@ -67,12 +67,12 @@ func (u *ContainerService) ContainerUpdate(req dto.ContainerOperate) error {
 			if err != nil {
 				return err
 			}
-			normalizeContainerEndpointSettings(ctx, client, networkConf, nil)
+			if err := normalizeContainerEndpointSettings(ctx, client, networkConf, nil); err != nil {
+				return err
+			}

 			cleanupErr, err := switchContainer(ctx, client, req.Name, oldContainer, func() (container.CreateResponse, error) {
-				return createContainerWithDynamicIPFallback(func() (container.CreateResponse, error) {
-					return client.ContainerCreate(ctx, config, hostConf, networkConf, &v1.Platform{}, req.Name)
-				}, networkConf.EndpointsConfig, oldContainer.NetworkSettings)
+				return client.ContainerCreate(ctx, config, hostConf, networkConf, &v1.Platform{}, req.Name)
 			}, config.Tty, t)
 			if err != nil {
 				return fmt.Errorf("update container failed, err: %v", err)
@@ -138,8 +138,14 @@ func (u *ContainerService) ContainerUpgrade(req dto.ContainerUpgrade) error {
 				config.Image = req.Image
 				hostConf := cloneContainerHostConfig(oldContainer.HostConfig)
 				preserveContainerVolumeMounts(hostConf, oldContainer.Mounts)
+				networkConf, extraNetworks := buildContainerRecoverNetworkConfig(oldContainer.NetworkSettings, hostConf)
+				if err := normalizeContainerEndpointSettings(ctx, client, networkConf, extraNetworks); err != nil {
+					upgradeErr := fmt.Errorf("prepare networks for container %s failed: %w", item, err)
+					upgradeErrors = append(upgradeErrors, upgradeErr)
+					return upgradeErr
+				}
 				cleanupErr, err := switchContainer(ctx, client, item, oldContainer, func() (container.CreateResponse, error) {
-					return createContainerWithOldNetworks(ctx, client, config, hostConf, oldContainer.NetworkSettings, item)
+					return createContainerWithNetworks(ctx, client, config, hostConf, networkConf, extraNetworks, item)
 				}, config.Tty, t)
 				if err != nil {
 					upgradeErr := fmt.Errorf("upgrade container %s failed: %w", item, err)
@@ -242,9 +248,8 @@ func (l *containerOperationMutex) lock(names ...string) func() {
 }

 type containerNetworkAttachment struct {
-	name      string
-	endpoint  *network.EndpointSettings
-	isDynamic bool
+	name     string
+	endpoint *network.EndpointSettings
 }

 type containerSwitchLogger interface {
@@ -604,9 +609,8 @@ func disconnectOriginalContainerNetworks(ctx context.Context, cli containerSwitc
 			return disconnected, fmt.Errorf("disconnect original container from network %s failed: %w", name, err)
 		}
 		disconnected = append(disconnected, containerNetworkAttachment{
-			name:      name,
-			endpoint:  endpoints[name],
-			isDynamic: isDynamicContainerNetwork(oldContainer.NetworkSettings, name),
+			name:     name,
+			endpoint: endpoints[name],
 		})
 	}
 	return disconnected, nil
@@ -616,10 +620,6 @@ func reconnectOriginalContainerNetworks(ctx context.Context, cli containerSwitch
 	var reconnectErr error
 	for _, attachment := range attachments {
 		err := cli.NetworkConnect(ctx, attachment.name, containerID, attachment.endpoint)
-		if err != nil && attachment.isDynamic && strings.Contains(err.Error(), unsupportedUserSpecifiedIPAddress) {
-			attachment.endpoint.IPAMConfig = nil
-			err = cli.NetworkConnect(ctx, attachment.name, containerID, attachment.endpoint)
-		}
 		if err != nil {
 			reconnectErr = errors.Join(reconnectErr, fmt.Errorf("reconnect original container to network %s failed: %w", attachment.name, err))
 		}
@@ -652,7 +652,7 @@ func restoreOriginalContainer(ctx context.Context, cli containerSwitchClient, ol
 	reconnectErr := reconnectOriginalContainerNetworks(ctx, cli, oldContainerID, disconnectedNetworks)
 	logContainerSwitchStep(logger, "ContainerRollbackReconnectOld", currentName, reconnectErr)
 	rollbackErr = errors.Join(rollbackErr, reconnectErr)
-	if wasRunning {
+	if wasRunning && reconnectErr == nil {
 		restartErr := restartOriginalContainer(ctx, cli, oldContainerID)
 		logContainerSwitchStep(logger, "ContainerRollbackRestartOld", currentName, restartErr)
 		rollbackErr = errors.Join(rollbackErr, restartErr)
@@ -660,17 +660,8 @@ func restoreOriginalContainer(ctx context.Context, cli containerSwitchClient, ol
 	return rollbackErr
 }

-func createContainerWithOldNetworks(ctx context.Context, client *client.Client, config *container.Config, hostConf *container.HostConfig, networkSettings *container.NetworkSettings, name string) (container.CreateResponse, error) {
-	networkConf, extraNetworks := buildContainerRecoverNetworkConfig(networkSettings, hostConf)
-	normalizeContainerEndpointSettings(ctx, client, networkConf, extraNetworks)
-	var primaryEndpoints map[string]*network.EndpointSettings
-	if networkConf != nil {
-		primaryEndpoints = networkConf.EndpointsConfig
-	}
-
-	created, err := createContainerWithDynamicIPFallback(func() (container.CreateResponse, error) {
-		return client.ContainerCreate(ctx, config, hostConf, networkConf, nil, name)
-	}, primaryEndpoints, networkSettings)
+func createContainerWithNetworks(ctx context.Context, client *client.Client, config *container.Config, hostConf *container.HostConfig, networkConf *network.NetworkingConfig, extraNetworks map[string]*network.EndpointSettings, name string) (container.CreateResponse, error) {
+	created, err := client.ContainerCreate(ctx, config, hostConf, networkConf, nil, name)
 	if err != nil {
 		return created, err
 	}
@@ -682,9 +673,6 @@ func createContainerWithOldNetworks(ctx context.Context, client *client.Client,
 	sort.Strings(extraNames)
 	for _, item := range extraNames {
 		err := client.NetworkConnect(ctx, item, created.ID, extraNetworks[item])
-		if clearUnsupportedDynamicEndpointIPAM(err, map[string]*network.EndpointSettings{item: extraNetworks[item]}, networkSettings) {
-			err = client.NetworkConnect(ctx, item, created.ID, extraNetworks[item])
-		}
 		if err != nil {
 			_ = client.ContainerRemove(ctx, created.ID, container.RemoveOptions{Force: true})
 			return created, err
@@ -692,16 +680,3 @@ func createContainerWithOldNetworks(ctx context.Context, client *client.Client,
 	}
 	return created, nil
 }
-
-func createContainerWithDynamicIPFallback(
-	create func() (container.CreateResponse, error),
-	endpoints map[string]*network.EndpointSettings,
-	networkSettings *container.NetworkSettings,
-) (container.CreateResponse, error) {
-	for {
-		created, err := create()
-		if err == nil || created.ID != "" || !clearUnsupportedDynamicEndpointIPAM(err, endpoints, networkSettings) {
-			return created, err
-		}
-	}
-}