Proxy-only subnets for internal HTTP(S) load balancers

This page describes how to work with proxy-only subnets for regional load balancers.

For a general overview of internal HTTP(S) load balancers, see the Internal HTTP(S) Load Balancing overview.

A proxy-only subnet is reserved exclusively for regional load balancer proxies. It cannot be used for other purposes. Only one proxy-only subnet for each VPC network in each region can be active.

You must create a proxy-only subnet before you create forwarding rules for your load balancers. Each region of a VPC network in which you use regional load balancers must have a proxy-only subnet.

The regional load balancer provides a pool of proxies for your network. The proxies evaluate where each HTTP(S) request should go based on the URL map, the backend service's session affinity, the balancing mode of each backend instance group or NEG, and other factors.

  1. A client makes a connection to the IP address and port of the load balancer's forwarding rule.

  2. One of the proxies receives and terminates the client's network connection.

  3. The proxy establishes a connection to the appropriate backend VM or endpoint in a NEG, as determined by the load balancer's URL map and backend services.

Each of the load balancer's proxies is assigned an internal IP address. The proxies for all regional load balancers in a region use internal IP addresses from a single, proxy-only subnet in that region of your VPC network. A proxy-only subnet must provide 64 or more IP addresses. This corresponds to a prefix length of /26 or shorter.

Only the proxies created by Google Cloud for a region's regional load balancers use the proxy-only subnet. The IP address for the load balancer's forwarding rule doesn't come from the proxy-only subnet. Also, the IP addresses of the backend VMs and endpoints don't come from the proxy-only subnet.

Each proxy listens on the IP address and port specified by the corresponding load balancer's forwarding rule. Each packet sent from a proxy to a backend VM or endpoint has a source IP address from the proxy-only subnet.

Proxy allocation is at the VPC level, not the load balancer level. This means that multiple regional load balancers that are in the same region and same VPC network share proxies for load balancing.

How proxy-only subnets fit into the load balancer's architecture

The following diagram shows the Google Cloud resources required for a regional load balancer.

Internal HTTP(S) load balancer numbered components (click to enlarge)
Internal HTTP(S) load balancer numbered components (click to enlarge)

As shown in the diagram, a regional load balancer deployment requires at least two subnets:

  • The load balancer's backend VMs and backend endpoints use a single subnet whose primary IP address range is 10.1.2.0/24 (in this example). This subnet isn't the proxy-only subnet. You can use multiple subnets for your backend VMs and endpoints if the subnets are in the same region as the load balancer. For internal HTTP(S) load balancers, the load balancer's IP address associated with the forwarding rule also must be in this subnet.
  • The proxy-only subnet is 10.129.0.0/23 (in this example).
  • Creating a proxy-only subnet

    Reserving a subnet for a regional load balancer is essentially the same procedure as creating any subnet, except with the addition of some flags.

    You can't reuse an existing subnet as the proxy-only subnet; you must create a new subnet in each region that has an internal HTTP(S) load balancer. This is in part because the subnets update command doesn't allow modifying a subnet's --purpose field. For the proxy-only subnet, the --purpose must be set to INTERNAL_HTTPS_LOAD_BALANCER.

    You must create a proxy-only subnet for use by the load balancers' proxies, before creating forwarding rules for your regional load balancers. If you try to configure a regional load balancer without first creating a proxy-only subnet for the region, the load balancer creation process fails.

    You must create one proxy-only subnet in each region of a virtual network (VPC) in which you use regional load balancers. This subnet is shared by all of your regional load balancers in the region.

    You must create proxy-only subnets regardless of whether your network is auto-mode or custom. The recommended subnet size is /23 (512 proxy-only addresses). It must be at least /26 (64 proxy-only addresses).

    The gcloud compute networks subnets create command creates a proxy-only subnet.

    gcloud compute networks subnets create SUBNET_NAME \
        --purpose=INTERNAL_HTTPS_LOAD_BALANCER \
        --role=ACTIVE \
        --region=REGION \
        --network=VPC_NETWORK_NAME \
        --range=CIDR_RANGE
    

    The fields are defined as follows:

    • SUBNET_NAME is the name of the proxy-only subnet.
    • REGION is the region of the proxy-only subnet.
    • VPC_NETWORK_NAME is the name of the VPC network that contains the subnet.
    • CIDR_RANGE is the primary IP address range of the subnet. You must use a subnet mask no longer than 26 so that at least 64 IP addresses are available for proxies in the region. The recommended subnet mask length is /23.

    Make sure to set the --purpose flag to the correct load balancer type, as follows:

    • Regional external HTTP(S) load balancer: Use --purpose=REGIONAL_MANAGED_PROXY.
    • Internal HTTP(S) load balancer: Use --purpose=INTERNAL_HTTPS_LOAD_BALANCER.

    For a complete configuration example, refer to Configuring the proxy-only subnet.

    You must configure a firewall rule for your backends to accept connections from the proxy-only subnet. See the fw-allow-proxies configuration in Configuring firewall rules.

    Proxy availability

    Sometimes Google Cloud regions don't have enough proxy capacity for a new regional load balancer. If this happens, the Cloud Console provides a proxy availability warning message when you are creating your load balancer. To resolve this issue, you can do one of the following:

    • Select a different region for your load balancer. This can be a practical option if you have backends in another region.
    • Select a VPC network that already has an allocated proxy-only subnet.
    • Wait for the capacity issue to be resolved.

    Changing the size or address range of a proxy-only subnet

    When you change the number of backends in your deployment, you might need to change the size or address range of your proxy-only subnet.

    You can't expand a proxy-only subnet in the same way that you would for a primary address range (with the expand-ip-range command). Instead, you must create a backup proxy-only subnet that meets your needs and then promote it to the active role. This is because only one proxy-only subnet can be active per region and per VPC network and because you can only expand a subnet's primary IP address range.

    You can create only one active and one backup proxy-only subnet per region, per VPC network.

    1. Create a backup proxy-only subnet in the same region, specifying a primary IP address range that meets your needs, using the gcloud compute networks subnets create command with the --role=BACKUP flag.

      gcloud compute networks subnets create BACKUP_PROXY_SUBNET \
         --purpose=INTERNAL_HTTPS_LOAD_BALANCER \
         --role=BACKUP \
         --region=REGION \
         --network=VPC_NETWORK_NAME \
         --range=CIDR_RANGE
      

      Make sure to set the --purpose flag to the correct load balancer type, as follows:

      • Regional external HTTP(S) load balancer: Use --purpose=REGIONAL_MANAGED_PROXY.
      • Internal HTTP(S) load balancer: Use --purpose=INTERNAL_HTTPS_LOAD_BALANCER.
    2. Create or modify ingress allow firewall rules that apply to your backend VMs or endpoints so that they include the primary IP address range of the backup proxy-only subnet.

    3. The following gcloud command promotes a backup proxy-only subnet to the active role and demotes the previously active proxy-only subnet to the backup role:

      gcloud compute networks subnets update BACKUP_PROXY_SUBNET \
         --region=REGION \
         --role=ACTIVE \
         --drain-timeout=CONNECTION_DRAINING_TIMEOUT
      

      Replace the following:

      • BACKUP_PROXY_SUBNET: the name of the newly-created backup proxy-only subnet
      • REGION: the region of the newly-created backup proxy-only subnet
      • CONNECTION_DRAINING_TIMEOUT: the amount of time, in seconds, that Google Cloud uses to migrate existing connections away from proxies in the previously-active proxy-only subnet

      Switching a proxy-only subnet from backup to active does not interrupt new connections:

      • The newly activated proxy-only subnet is used for new connections.
      • The previously active (now backup) proxy-only subnet is no longer used for new connections.
      • Google Cloud begins draining existing connections from proxies in the previously active (now backup) proxy-only subnet.
    4. After the connection draining timeout, or after you're confident that connections to your backend VMs or endpoints aren't coming from proxies in the previously active (now backup) proxy-only subnet, you can do the following:

      • Modify ingress allow firewall rules that apply to your backend VMs or endpoints so they don't include the primary IP address range of the previously active (now backup) proxy-only subnet.
      • Delete the previously active (now backup) proxy-only subnet to release the IP addresses that the subnet used for its primary IP address range.

    Example

    1. Create a backup proxy-only subnet dedicated to the region.

      Make sure to set the --purpose flag to the correct load balancer type, as follows:

      • Regional external HTTP(S) load balancer: Use --purpose=REGIONAL_MANAGED_PROXY
      • Internal HTTP(S) load balancer: Use --purpose=INTERNAL_HTTPS_LOAD_BALANCER

      gcloud compute networks subnets create new-l7ilb-backend-subnet-us-west1 \
         --purpose=INTERNAL_HTTPS_LOAD_BALANCER \
         --role=BACKUP \
         --region=us-west1 \
         --network=default \
         --range=10.130.0.0/23
      

    2. Update your backend firewall to accept connections from the new subnet.

      gcloud compute firewall-rules update l7-ilb-firewall \
         --source-ranges 10.129.0.0/23,10.130.0.0/23
      
    3. Update the new subnet, setting it to be the ACTIVE proxy-only subnet in the region and waiting for the old subnet to drain.

      gcloud compute networks subnets update new-l7ilb-ip-range-us-west1 \
         --drain-timeout 1h --role ACTIVE
      

      To drain an IP address range immediately, set the --drain-timeout to 0s. This promptly ends all connections to proxies that have assigned addresses in the subnet that is being drained.

    4. Monitor the status of the drain by using a list or describe command. The status of the subnet is DRAINING while it is being drained.

      gcloud compute networks subnets list
      

      Wait for draining to complete. When the old proxy-only subnet is drained, the status of the subnet is READY.

    5. Update your backend firewall rule to only allow connections from the new subnet.

      gcloud compute firewall-rules update l7-ilb-firewall \
         --source-ranges 10.130.0.0/23
      
    6. Delete the old subnet.

      gcloud compute networks subnets delete l7ilb-ip-range-us-west1 \
         --region us-west1
      

    Deleting a proxy-only subnet

    Deleting a proxy-only subnet releases its primary IP address range so you can use the range for another purpose. Google Cloud enforces the following rules when it receives a request to delete a proxy-only subnet:

    • An active proxy-only subnet cannot be deleted if there is at least one regional load balancer in the same region and VPC network.

    • An active proxy-only subnet cannot be deleted if there is a backup proxy-only subnet in the same region and VPC network.

      If you try to delete an active proxy-only subnet before deleting the backup, the following error message appears: "Invalid resource usage: Cannot delete ACTIVE subnetwork because a BACKUP subnetwork exists."

    Practically, these rules have the following effect:

    • If no regional load balancer is defined in a given region and VPC network, you can delete the proxy-only subnets in that region. If a backup proxy-only subnet exists, you must first delete it before you can delete the active proxy-only subnet.

    • If you have at least one regional load balancer defined in a given region and VPC network, you cannot delete the active proxy-only subnet; however, you can promote a backup proxy-only subnet to the active role, which automatically demotes the previously active proxy-only subnet to the backup role. After connections are drained, you can delete the backup (previously active) proxy-only subnet.

    Refer to deleting subnets in the VPC network documentation for more information.

    Limitations

    The following constraints apply to proxy-only subnets:

    • You can't have both an INTERNAL_HTTPS_LOAD_BALANCER and a REGIONAL_MANAGED_PROXY subnet in the same network and region, in the same way you can't have two REGIONAL_MANAGED_PROXY proxies or two INTERNAL_HTTPS_LOAD_BALANCER proxies.

    • You can create only one active and one backup proxy-only subnet in each region in each VPC network.

    • You cannot create a backup proxy-only subnet unless you have already created an active proxy-only subnet in that region and network.

    • You can change the role of a proxy-only subnet from backup to active by updating the subnet. When you do that, Google Cloud automatically changes the previously active proxy-only subnet to backup. You cannot explicitly set the role of a proxy-only subnet to backup by updating it.

    • During a proxy-only subnet's connection draining period (--drain-timeout), you cannot change the role of a proxy-only subnet from backup to active.

    • Google Cloud doesn't warn you if your proxy-only subnet runs out of IP addresses.

    • Proxy-only subnets do not support VPC Flow Logs.

    What's next