Google Cloud Platform (GCP) HTTP(S) load balancing provides global load balancing for HTTP(S) requests destined for your instances. You can configure URL rules that route some URLs to one set of instances and route other URLs to other instances. Requests are always routed to the instance group that is closest to the user, provided that group has enough capacity and is appropriate for the request. If the closest group does not have enough capacity, the request is sent to the closest group that does have capacity.
HTTP(S) load balancing supports both IPv4 and IPv6 addresses for client traffic. Client IPv6 requests are terminated at the global load balancing layer, then proxied over IPv4 to your backends.
HTTP requests can be load balanced based on port 80 or port 8080. HTTPS requests can be load balanced on port 443.
The load balancer acts as an HTTP/2 to HTTP/1.1 translation layer, which means that the web servers always see and respond to HTTP/1.1 requests, but that requests from the browser can be HTTP/1.0, HTTP/1.1, or HTTP/2. HTTP/2 server push is not supported.
Before you begin
HTTP(S) load balancing uses instance groups to organize instances. Make sure you are familiar with instance groups before you use load balancing.
If you want to jump right in and build a working load balancer for testing, the following guides demonstrate two different scenarios using the HTTP(S) load balancing service. These scenarios provide a practical context for HTTP(S) load balancing and demonstrate how you might set up load balancing for your specific needs.
The rest of this page digs into more detail about how load balancers are constructed and how they work.
Creating a cross-region load balancer
You can use a global IP address that can intelligently route users based on proximity. For example, if you set up instances in North America, Europe, and Asia, users around the world will be automatically sent to the backends closest to them, assuming those instances have enough capacity. If the closest instances do not have enough capacity, cross-region load balancing automatically forwards users to the next closest region.
Creating a content-based load balancer
Content-based or content-aware load balancing uses HTTP(S) load balancing to
distribute traffic to different instances based on the incoming HTTP(S) URL. For
example, you can set up some instances to handle your video
content and another set to handle everything else. You can configure your load
balancer to direct traffic for
example.com/video to the video servers
example.com/ to the default servers.
Content-based and cross-region load-balancing can work together by using multiple backend services and multiple regions. You can build on top of the scenarios above to configure your own load balancing configuration that meets your needs.
An HTTP(S) load balancer is composed of several components. The following diagram illustrates the architecture of a complete HTTP(S) load balancer:
The following sections describe how each component works together to make up each type of load balancer. For a detailed description of each component, see Components below.
HTTP load balancing
A complete HTTP load balancer is structured as follows:
- A global forwarding rule directs incoming requests to a target HTTP proxy.
- The target HTTP proxy checks each request against a URL map to determine the appropriate backend service for the request.
- The backend service directs each request to an appropriate backend based on serving capacity, zone, and instance health of its attached backends. The health of each backend instance is verified using either an HTTP health check or an HTTPS health check. If the backend service is configured to use the latter, the request will be encrypted on its way to the backend instance.
HTTPS load balancing
An HTTPS load balancer shares the same basic structure as an HTTP load balancer (described above), but differs in the following ways:
- Uses a target HTTPS proxy instead of a target HTTP proxy
- Requires at least one signed SSL certificate installed on the target HTTPS proxy for the load balancer.
- The client SSL session terminates at the load balancer. Sessions between the load balancer and the instance can either be HTTPS (recommended). or HTTP. If you use HTTPS, each instance in the backend services must have an SSL certificate.
Global forwarding rules and addresses
Global forwarding rules route traffic by IP address, port, and protocol to a load balancing configuration consisting of a target proxy, URL map, and one or more backend services.
Each global forwarding rule provides a single global IP address that can be used in DNS records for your application. No DNS-based load balancing is required. You can either specify the IP address to be used or let Google Compute Engine assign one for you.
Target proxies terminate HTTP(S) connections from clients, and are referenced by one or more global forwarding rules and route the incoming requests to a URL map.
The proxies set HTTP request/response headers as follows:
Via: 1.1 google(requests and responses)
X-Forwarded-Proto: [http | https](requests only)
X-Forwarded-For: <unverified IP(s)>, <immediate client IP>, <global forwarding rule external IP>, <proxies running in GCP>(requests only)
A comma-separated list of IP addresses appended by the intermediaries the request traveled through. If you are running proxies inside GCP that append data to the
X-Forwarded-Forheader, then your software must take into account the existence and number of those proxies. Only the <immediate client IP> and <global forwarding rule external IP> entries are provided by the load balancer. All other entries in the list are passed along without verification. The <immediate client IP> entry is the client that connected directly to the load balancer. The <global forwarding rule external IP> entry is the external IP address of the load balancer's forwarding rule. If there are more entries than that, then the first entry in the list is the address of the original client. Other entries before the <immediate client IP> entry represent other proxies that forwarded the request along to the load balancer.
X-Cloud-Trace-Context: <trace-id>/<span-id>;<trace-options>(requests only)
Parameters for Stackdriver Trace.
URL maps define matching patterns for URL-based routing of requests to the appropriate backend services. A default service is defined to handle any requests that do not match a specified host rule or path matching rule. In some situations, such as the cross-region load balancing example, you might not define any URL rules and rely only on the default service. For content-based routing of traffic, the URL map allows you to divide your traffic by examining the URL components to send requests to different sets of backends.
If you are using HTTPS load balancing, you must install one or more SSL certificates on the target HTTPS proxy. You can have up to ten (10) SSL certificates installed. They are used by target HTTPS proxies to authenticate communications between the load balancer and the client.
For more information on installing SSL certificates, see SSL Certificates.
If you are using HTTPS from the load balancer to the backends, you must install SSL certificates on each VM instance. To install SSL certificates on a VM instance, use the instructions in your application documentation.
SSL policies give you the ability to control the features of SSL that your HTTPS load balancer negotiates with HTTPS clients.
By default, HTTPS load balancing uses a set of SSL features that provides good security and wide compatibility. Some applications require more control over which SSL versions and ciphers are used for their HTTPS or SSL connections. You can define SSL policies that control the features of SSL that your load balancer negotiates and associate an SSL policy with your target HTTPS proxy.
Backend services direct incoming traffic to one or more attached backends. Each backend is composed of an instance group and additional serving capacity metadata. Backend serving capacity can be based on CPU or requests per second (RPS).
Each backend service also specifies which health checks will be performed against the available instances.
HTTP(S) load balancing supports Compute Engine Autoscaler, which allows users to perform autoscaling on the instance groups in a backend service. For more information, see Scaling Based on HTTP load balancing serving capacity.
You can enable connection draining on backend services to ensure minimal interruption to your users when an instance that is serving traffic is terminated, removed manually, or removed by an autoscaler. To learn more about connection draining, read the Enabling Connection Draining documentation.
Backend buckets direct incoming traffic to Google Cloud Storage buckets. See Adding a Cloud Storage bucket to content-based load balancing for an example of adding buckets to an existing load balancer setup.
You must create a
firewall rule that allows traffic from
188.8.131.52/16 to reach your instances. These are IP
address ranges that the load balancer uses to connect to backend instances.
This rule allows traffic from both the load balancer and the health checker.
The rule must allow traffic on the port your global forwarding rule has been
configured to use, and your health checker should be configured to use the
same port. If your health checker uses a different port, then you must create
another firewall rule for that port.
Note that firewall rules block and allow traffic at the instance level, not at the edges of the network. They cannot prevent traffic from reaching the load balancer itself.
If you need to determine external IP addresses at a particular time, use the instructions in the Google Compute Engine FAQ.
Load distribution algorithm
HTTP(S) load balancing provides two methods of determining instance load.
Within the backend service object, the
balancingMode property selects between
requests per second (RPS)
and CPU utilization modes. Both modes allow a
maximum value to be specified; the HTTP load balancer will try to ensure that
load remains under the limit, but short bursts above the limit can occur during
failover or load spike events.
Incoming requests are sent to the region closest to the user, provided that region has available capacity. If more than one zone is configured with backends in a region, the traffic is distributed across the instance groups in each zone according to each group's capacity. Within the zone, the requests are spread evenly over the instances using a round-robin algorithm. Round-robin distribution can be overridden by configuring session affinity.
Session affinity sends all request from the same client to the same virtual machine instance as long as the instance stays healthy and has capacity.
GCP HTTP(S) Load Balancing offers two types of session affinity:
- client IP affinity— forwards all requests from the same client IP address to the same instance.
- generated cookie affinity— sets a client cookie, then sends all requests with that cookie to the same instance.
WebSocket proxy support
The HTTP(S) load balancer has native support for the WebSocket protocol. Backends that use WebSocket to communicate with clients can use the HTTP(S) load balancer as a front end, for scale and availability. The load balancer does not need any additional configuration to proxy WebSocket connections.
The WebSocket protocol, which is defined in RFC6455, provides a full-duplex communication channel between clients and servers. The channel is initiated from an HTTP(S) request.
When the HTTP(S) load balancer recognizes a WebSocket Upgrade request from an HTTP(S) client and the request is followed by a successful Upgrade response from the backend instance, the load balancer proxies bidirectional traffic for the duration of the current connection. If the backend does not return a successful Upgrade response, the load balancer closes the connection.
The timeout for a WebSocket connection depends on the configurable response timeout of the load balancer, which is 30 seconds by default. This timeout is applied to WebSocket connections regardless of whether they are in use or not. For more information about the response timeout and how to configure it, refer to Timeouts and retries.
If you have configured either client IP or generated cookie session affinity for your HTTP(S) load balancer, all WebSocket connections from a client are sent to the same backend instance, provided the instance continues to pass health checks and has capacity.
Your HTTP(S) load balancing service can be configured and updated through the following interfaces:
gcloudcommand-line tool: a command-line tool included in the Cloud SDK. The HTTP(S) load balancing documentation calls on this tool frequently to accomplish tasks. For a complete overview of the tool, see the gcloud Tool Guide. You can find commands related to load balancing in the
gcloud computecommand group.
You can also get detailed help for any
gcloudcommand by using the
gcloud compute http-health-checks create --help
The Google Cloud Platform Console: Load balancing tasks can be accomplished through the Google Cloud Platform Console.
The REST API: All load balancing tasks can be accomplished using the Google Compute Engine API. The API reference docs describe the resources and methods available to you.
By default, an HTTPS target proxy accepts only TLS 1.0, 1.1, and 1.2 when terminating client SSL requests. You can use SSL policies to change this default behavior and control how the load balancer negotiates SSL with clients.
When the load balancer uses HTTPS as a backend service protocol, it can negotiate TLS 1.0, 1.1, or 1.2 to the backend.
Timeouts and retries
HTTP(S) load balancing has two distinct types of timeouts:
A configurable response timeout, which represents the amount of time the load balancer will wait for your backend to return a complete response. It is not an idle (keepalive) timeout. This timeout is configurable by modifying the timeout setting for your backend service. The default value is 30 seconds. Consider increasing this timeout under these circumstances:
- If you expect a backend to take longer to return HTTP responses, or
- If the connection is upgraded to a WebSocket.
A TCP session timeout, whose value is fixed at 10 minutes (600 seconds). This session timeout is sometimes called a keepalive or idle timeout, and its value is not configurable by modifying your backend service. You must configure the web server software used by your backends so that its keepalive timeout is longer than 600 seconds to prevent connections from being closed prematurely by the backend. This timeout does not apply to WebSockets.
This table illustrates changes necessary to modify keepalive timeouts for common web server software:
|Web Server Software||Parameter||Default Setting||Recommended Setting|
|Apache||KeepAliveTimeout||KeepAliveTimeout 5||KeepAliveTimeout 620|
|nginx||keepalive_timeout||keepalive_timeout 75s;||keepalive_timeout 620s;|
HTTP(S) load balancing retries failed GET requests in certain circumstances, such as when the response timeout is exhausted. It does not retry failed POST requests. Retries are limited to two attempts. Retried requests only generate one log entry for the final response. Refer to Logging for more information.
Illegal request handling
The HTTP(S) load balancer blocks client requests from reaching the backend for a number of reasons: some strictly for HTTP/1.1 compliance and others to avoid unexpected data being passed to the backends.
The load balancer blocks the following for HTTP/1.1 compliance:
- It cannot parse the first line of the request.
- A header is missing the
- Headers or the first line contain invalid characters.
- The content length is not a valid number, or there are multiple content length headers.
- There are multiple transfer encoding keys, or there are unrecognized transfer encoding values.
- There's a non-chunked body and no content length specified.
- Body chunks are unparseable. This is the only case where some data will make it to the backend. The load balancer will close the connections to client and backend when it receives an unparseable chunk.
The load balancer also blocks the request if any of the following are true:
- The combination of request URL and headers is longer than about 15KB.
- The request method does not allow a body, but the request has one.
- The request contains an upgrade header.
- The HTTP version is unknown.
Each HTTP(S) request is logged temporarily via Stackdriver Logging. If you have been accepted into the Alpha testing phase, logging is automatic and does not need to be enabled.
How to view logs
To view logs, go to the Logs Viewer.
- To see all logs, in the first pull-down menu select Cloud HTTP Load Balancer > All forwarding_rule_name.
- To see logs for just one forwarding rule, select a single forwarding rule name from the list.
- To see logs for just one URL map used by a forwarding rule, select GCE URL Map and choose the URL map of interest.
Log fields of type boolean typically only appear if they have a value of
If a boolean field has a value of
false, that field is omitted from the log.
UTF-8 encoding is enforced for log fields. Characters that are not UTF-8 characters are replaced with question marks.
What is logged
HTTP(S) load balancing log entries contain information useful for monitoring and debugging your HTTP(S) traffic. Log entries contain the following types of information:
- General information shown in most GCP logs, such as severity, project ID, project number, timestamp, and so on.
- HttpRequest log fields.
statusDetailsfield inside the
structPayload. This field holds a string that explains why the load balancer returned the HTTP status that it did. The tables below contain further explanations of these log strings.
statusDetail HTTP success messages
|statusDetails (successful)||Meaning||Common Accompanying Response Codes|
||The HTTP request was served from cache.||Any cachable response code is possible.|
||The return code was set from a cached entry that was validated by a backend.||Any cachable response code is possible.|
||The HTTP request was proxied successfully to the backend.||Returned from VM backend - any response code is possible.|
statusDetail HTTP failure messages
|statusDetails (failure)||Meaning||Common Accompanying Response Codes|
||A request with body was aborted due to backend sending an early response with an error code. The response was forwarded to the client. The request was terminated.||4XX or 5XX|
||The backend sent a 503 that the load balancer could not recover from with retries.||503|
||The backend connection closed unexpectedly after a partial response had been sent to the client.||Returned from VM backend - any response code is possible. A 0 indicates the backend did not send full response headers.|
||The backend unexpectedly closed its connection to the load balancer before the response was proxied to the client. This can happen if the load balancer is sending traffic to another entity, such as a third party load balancer running on a VM instance, that has a TCP timeout that is shorter than the HTTP(S) load balancer's 10 minute (600 second) timeout. Manually setting the TCP timeout (keepalive) on the target service to greater than 600 seconds may fix this problem.||502|
||The backend sent a non-error response (1XX or 2XX) to a request before receiving the whole request body.||502|
||The HTTP response body sent by the backend has invalid chunked transfer-encoding or is otherwise corrupted.||Any response code possible depending on the nature of the corruption. Often 502.|
||The backend timed out while generating a response.||502|
||The client sent a HTTP request with a body, but the HTTP method used does not allow a body.||400|
||The load balancer failed to serve a full response from cache due to an internal error.||2XX|
||The connection to the client was broken after the load balancer sent a partial response.||Returned from the VM backend - any response code is possible.|
||The connection to the client was broken before the load balancer sent any response.||0|
||The load balancer idled out the client connection due to lack of progress while proxying either the request or response.||0|
||There was an error uncompressing a gzipped HTTP response.||503|
||The load balancer failed to connect to the backend.||502|
||The load balancer failed to pick a healthy backend to handle the request.||502|
||The request headers were larger than the maximum allowed.||413|
||HTTP version not supported. Currently only HTTP 0.9, 1.0, 1.1, and 2.0 are supported.||400|
||The load balancer canceled the HTTP/2 server push because the backend returned an invalid response code.||Can only happen when using http2 to the backend. Client will receive a RST_STREAM containing INTERNAL_ERROR.|
||Internal error at the load balancer.||400|
||The HTTP/2 headers from client are invalid.||400|
||The request body was improperly chunk encoded.||411|
||The HTTP request requires a body but the request headers did not include a content length or transfer-encoding chunked header.||400 or 403|
||A request with a https:// URL was received over a plaintext HTTP/1.1 connection.||400|
||The client supplied an unsupported HTTP request method.||400|
||The client HTTP request contained the Upgrade header and was refused.||400|
||The HTTP request URI was longer than the maximum allowed length.||414|
||The websocket connection was closed.|
||The websocket handshake failed.||501|
HTTP(S) Load Balancing exports monitoring data to Stackdriver. Monitoring metrics can be used to evaluate an HTTP(S) load balancer's configuration, usage, and performance; troubleshoot problems; and improve resource utilization and user experience.
In addition to the predefined dashboards in Stackdriver, you can create custom dashboards, set up alerts, and query the metrics through the Stackdriver monitoring API.
Viewing Stackdriver monitoring dashboards
- Go to Stackdriver in the Google Cloud Platform Console.
Go to Stackdriver
- Select Resources > Google Cloud Load Balancers.
- Click the name of your load balancer.
In the left pane, you can see various details for this HTTP(S) load balancer. In the right pane you can see timeseries graphs. Click the Breakdowns link to see specific breakdowns.
Defining Stackdriver alerts
You can define Stackdriver alerts over various HTTP(S) Load Balancing metrics:
- Go to Stackdriver in the Google Cloud Platform Console.
Go to Stackdriver
- Select Alerting > Create a Policy.
- Click on Add Condition and select condition type.
- Select metrics and filters. For metrics, the resource type is HTTP(S) Load Balancer.
- Click Save Condition.
- Enter policy name and click Save Policy.
Defining Stackdriver custom dashboards
You can create custom Stackdriver dashboards over HTTP(S) Load Balancing metrics:
- Go to Stackdriver in the Google Cloud Platform Console.
Go to Stackdriver
- Select Dashboards > Create Dashboard.
- Click on Add Chart.
- Give the chart a title.
- Select metrics and filters. For metrics, the resource type is HTTP(S) Load Balancer.
- Click Save.
Metric reporting frequency and retention
Metrics for the HTTP(S) load balancers are exported to Stackdriver in 1-minute granularity batches. Monitoring data is retained for six (6) weeks. The dashboard provides data analysis in default intervals of 1H, 6H, 1D, 1W, and 6W. You can manually request analysis in any interval from 6W to 1 minute.
Monitoring metrics for HTTP(S) load balancers
The following metrics for HTTP(S) load balancers are reported into Stackdriver:
|Request count||The number of requests served by the HTTP(S) load balancer|
|Request bytes count||The number of bytes sent as requests from users to the HTTP(S) load balancer|
|Response bytes count||The number of bytes sent as responses from the HTTP(S) load balancer to users|
|Total latencies||A distribution of the latency measured from the time the first
byte of the request was received by the load balancer proxy until the proxy
receives an ACK from the requesting client on the last response byte.
Total latencies are measured by request/response, so pauses between requests
on the same connection using `Connection: keep-alive` do not affect the
measurement. This measurement is typically reduced to 95th percentile in
Example: a load balancer has 1 request per second from the UK, all with 100ms latency, and 9 requests per second from the US, all with 50ms latency. Over a certain minute there were 60 requests from the UK and 540 requests from the US. Monitoring metrics preserves the distribution over all dimensions. You can request the following:
|Frontend RTT(*)||A distribution of the smoothed RTT measured for each connection between client and proxy (measured by the proxy's TCP stack). Typically reduced to 95th percentile in Stackdriver views.|
|Backend latencies(*)||A distribution of the latency measured from when the first request byte was sent by the proxy to the backend, until the proxy received from the backend the last byte of the response. Typically reduced to 95th percentile in Stackdriver views.|
|Response code class fraction||Percentage of total HTTP(S) load balancer responses that are in each response code class (2XX, 4XX, ...). In Stackdriver, this value is only available via default dashboards. It is not available for custom dashboards. You can set alerts for it via the API.|
(*) The sum of Frontend RTT and Backend latencies is not guaranteed to be less than or equal to Total latencies. This is because although we poll RTT over the socket from proxy to client at the time the HTTP response is ACKed, we rely on kernel reporting for some of these measurements, and we cannot guarantee that the kernel will have an RTT measurement for the given HTTP response. The end result is a smoothed RTT value that is also affected by previous HTTP responses, SYN/ACKs, and SSL handshakes that aren't affecting current HTTP request actual timings.
Filtering dimensions for HTTP(S) load balancer metrics
Metrics are aggregated for each HTTP(S) load balancer. You can filter aggregated metrics by the following dimensions(*):
|BACKEND ZONE||The Google Cloud Platform zone of the instance group that served the user
|PROXY CONTINENT||Continent of the HTTP(S) proxy that terminated the user HTTP(S) connection.
|INSTANCE GROUP||The name of the instance group that served the user request.
If no instance group was available or if the request was served by another entity, you will see one of the following values instead of an instance group.
|BACKEND SERVICE||The name of the backend service that served the user request.|
|MATCHED URL RULE||The URL map path rule that matched the prefix of the user HTTP(S) request (up to 50 characters).|
(*) Currently the "Response code class fraction" metric is available per entire load balancer only, with no further breakdowns available.
Notes and Restrictions
- HTTP(S) load balancing supports the
HTTP/1.1 100 Continueresponse.
- If your load balanced instances are running a public operating system image supplied by Compute Engine, then firewall rules in the operating system will be configured automatically to allow load balanced traffic. If you are using a custom image, you have to configure the operating system firewall manually. This is separate from the GCP firewall rule that must be created as part of configuring an HTTP(S) load balancer.
- Load balancing does not keep instances in sync. You must set up your own mechanisms, such as using Deployment Manager, for ensuring that your instances have consistent configurations and data.
- The HTTP(S) load balancer does not support sending an HTTP DELETE with a body
to the load balancer. Such requests will receive an error message:
Error 400 (Bad Request)!! Your client has issued a malformed or illegal request.Only DELETE requests without bodies are supported.
Load balanced traffic does not have a source address of the original client
Traffic from the load balancer to your instances has an IP address in the
184.108.40.206/16. When viewing logs on your
load balanced instances, you will not see the source address of the
original client. Instead, you will see source addresses from this range.
Getting a permission error when trying to view an object in my Cloud Storage bucket
In order to serve objects through load balancing, the Cloud Storage objects must be publicly accessible. Make sure to update the permissions of the objects being served so they are publicly readable.
URL doesn’t serve expected Cloud Storage object
The Cloud Storage object to serve is determined based on your URL map and the URL that you request. If the request path maps to a backend bucket in your URL map, the Cloud Storage object is determined by appending the full request path onto the Cloud Storage bucket that the URL map specifies.
For example, if you map
gs://[EXAMPLE_BUCKET], the request to
https://<GCLB IP or Host>/static/path/to/content.jpg will try to serve
gs://[EXAMPLE_BUCKET]/static/path/to/content.jpg. If that object doesn’t exist,
you will get the following error message instead of the object:
The specified key does not exist.
Compression isn't working
If responses served by HTTP(S) Load Balancing are not compressed but should be, check that
the web server software running on your instances is configured to compress
responses. By default, some web server software will automatically disable
compression for requests that include a
Via header. The presence of a
Via header indicates the request was forwarded by a proxy. Because it is a
proxy, HTTP(S) Load Balancing adds a
Via header to each request as
required by the HTTP specification.
To enable compression, you may have to override your web server's default
configuration to tell it to compress responses even if the request had a
If you are using the nginx web server
software, modify the
nginx.conf configuration file to enable
compression. The location of this file depends on where nginx is installed. In
many Linux distributions, the file is stored at
allow nginx compression to work with HTTP(S) load balancing, add the following
two lines to the
http section of nginx.conf:
gzip_proxied any; gzip_vary on;
The first line enables compression even for requests forwarded
by a proxy like HTTP(S) Load Balancing. The second line adds a
Vary: Accept-Encoding header to responses.
Vary: Accept-Encoding notifies
caching proxies such as Cloud CDN that they should maintain separate cache
entries for compressed and non-compressed variants of compressible resources.
After modifying nginx.conf, you need to restart nginx before it will use the new
configuration. In many Linux distributions, nginx can be restarted by running
sudo service nginx restart or