Varnish

Varnish is a popular HTTP proxy server, often used for caching. You usually don’t need it with Platform.sh as the standard router includes HTTP cache and a CDN would cover more advanced uses. But you can include Varnish as a service.

Supported versions

You can select the major and minor version.

Patch versions are applied periodically for bug fixes and the like. When you deploy your app, you always get the latest available patches.

How it works

All incoming requests go through the standard router. The Varnish service sits between the router and all apps in the project.

Usage example

1. Configure the service

To define the service, use the varnish type:

# The name of the service container. Must be unique within a project.
<SERVICE_NAME>:
    type: varnish:<VERSION>
    relationships:
        <RELATIONSHIP_NAME>: '<APP_NAME>:http'
    configuration:
        vcl: !include
            type: string
            path: config.vcl

Note that changing the name of the service replaces it with a brand new service and all existing data is lost. Back up your data before changing the service.

The relationships block defines the connection between Varnish and your app. You can define RELATIONSHIP_NAME as you like. APP_NAME should match your app’s name in the app configuration.

The configuration block must reference a VCL file inside the .platform directory. The path defines the file relative to the .platform directory.

Example configuration

Service definition

# The name of the service container. Must be unique within a project.
varnish:
    type: varnish:7.3
    relationships:
        application: 'myapp:http'
    configuration:
        vcl: !include
            type: string
            path: config.vcl

Notice the relationship (application) defined for the service varnish granting access to the application container myapp.

2. Create a VCL template

To tell Varnish how to handle traffic, in the .platform directory add a Varnish Configuration Language (VCL) template.

This template is supplemented by automatic additions from Platform.sh. So you MUST NOT include certain features that you might elsewhere:

• A vcl_init() function: The function is automatically generated based on the relationships you defined in Step 1. Each defined relationship results in a backend with the same name.
• The VCL version at the start: This is automatically generated.
• Imports for std or directors: These are already imported. You can import other modules.

The file MUST include:

• A definition of which backend to use in a vcl_recv() subroutine.

The logic varies based on whether you have one or more apps.

Example VCL template with one app

To serve one app, your VCL template needs at least the following function:

sub vcl_recv {
    set req.backend_hint = RELATIONSHIP_NAME.backend();
}

Where RELATIONSHIP_NAME is the name of the relationship you defined in Step 1. With the example configuration, that would be the following:

sub vcl_recv {
    set req.backend_hint = application.backend();
}

Example VCL template with multiple apps

If you have multiple apps fronted by the same Varnish instance, your VCL templates needs logic to determine where each request is forwarded.

For example, you might have the following configuration for two apps:

varnish:
    type: varnish:7.3
    relationships:
        blog:
            service: blog
            endpoint: http
        main:
            service: app
            endpoint: http
    configuration:
        vcl: !include
            type: string
            path: config.vcl

blog:
    # The location of the application's code.
    source:
        root: "backends/blog"
    # The type of the application to build.
    type: "php:8.3"

app:
    # The location of the application's code.
    source:
        root: "backends/main"
    # The type of the application to build.
    type: "nodejs:22"

You could then define that all requests to /blog/ go to the blog app and all other requests to the other app:

sub vcl_recv {
    if (req.url ~ "^/blog/") {
        set req.backend_hint = blog.backend();
    } else {
        set req.backend_hint = main.backend();
    }
}

3. Route incoming requests to Varnish

Edit your route definitions to point to the Varnish service you just created. Also disable the router cache as Varnish now provides caching.

To forward all incoming requests to Varnish rather than your app, you could have the following:

"https://{default}/":
    type: upstream
    upstream: "varnish:http"
    cache:
        enabled: false

Varnish forwards requests to your app based on the specified VCL template.

Include modules

You can include the following optional modules in your VCL templates to add additional features:

• cookie
• header
• saintmode
• softpurge
• tcp
• var
• vsthrottle
• xkey

To use them, add an import to your template such as the following:

import xkey;

Circular relationships

At this time, Platform.sh doesn’t support circular relationships between services and apps. That means you can’t add a relationship from an app fronted by Varnish to the Varnish service. If you do so, then one of the relationships is skipped and the connection doesn’t work.

Rate limit connections

Sometimes you want to ensure that users, whether human or machine, can’t overload your app with requests. One tool to help is using Varnish to limit the rate at which connections can be made.

For example, say you want to make sure no one can make more than 20 requests within 10 seconds. If they do, you want to block them from any more requests for 2 minutes.

To do so, import the vsthrottle module and add logic similar to the following to your VCL template:

import vsthrottle;

sub vcl_recv {
  # The Platform.sh router provides the real client IP as X-Client-IP
  # This replaces client.identity in other implementations
  if (vsthrottle.is_denied(req.http.X-Client-IP, 20, 10s, 120s)) {
    # Client has exceeded 20 requests in 10 seconds.
    # When this happens, block that IP for the next 120 seconds.
    return (synth(429, "Too Many Requests"));
  }

  # Set the standard backend for handling requests that aren't limited
  set req.backend_hint = application.backend();
}

Clear cache with a push

You may want at times to clear a specific part of your cache when you know the content is outdated. With Varnish, you can clear the content with purging and banning.

The following example shows how to set up purging.

1. Add an access control list to your VCL template:

   .platform/config.vcl

   acl purge {
       "localhost";
       "192.0.2.0"/24;
   }

   This list ensures that only requests from the listed IPs are accepted. Choose which IPs to allow. If you are sending requests from an app, checkout the outbound IPs for the region.

   Alternatively, you could code in a token that must be sent with the request.

2. Add purge handling:

   .platform/config.vcl

   sub vcl_recv {
       if (req.method == "PURGE") {
           # The Platform.sh router provides the real client IP as X-Client-IP
           # Use std.ip to convert the string to an IP for comparison
           if (!std.ip(req.http.X-Client-IP, "0.0.0.0") ~ purge) {
               # Deny all purge requests not from the allowed IPs
               return(synth(403,"Not allowed."));
           }
           # Purge cache for allowed requests
           return (purge);
       }
       ...
   }

3. Set up cache purging to suit your needs. The following cURL call gives an example of how this can work:

   curl -X PURGE "URL_TO_PURGE"

Stats endpoint

The Varnish service also offers an http+stats endpoint, which provides access to some Varnish analysis and debugging tools.

You can’t use it from an app fronted by Varnish because of the restriction with circular relationships. To access the stats, create a separate app (stats-app) with a relationship to Varnish, but not from it. Define app configuration similar to the following:

# The name of the app container. Must be unique within a project.
stats-app:
    # The location of the application's code.
    source:
        root: "stats"
    # The type of the application to build.
    type: "python:3.9"
    # Unique relationship _to_ Varnish from 'stats-app', where no relationship
    #   is defined _from_ Varnish to the same app, to avoid circular relationships.
    relationships:
        varnishstats: 
            service: varnish
            endpoint: "http+stats"
# The name of the app container. Must be unique within a project.
main-app:
    # The location of the application's code.
    source:
        root: "backends/main"
    # The type of the application to build.
    type: "nodejs:22"

# The name of the service container. Must be unique within a project.
varnish:
    type: varnish:7.3
    # Unique relationship _from_ Varnish _to_ 'main-app', where no relationship
    #   is defined _to_ Varnish to the same app, to avoid circular relationships.
    relationships:
        main: 
            service: "main-app"
            endpoint: http
    configuration:
        vcl: !include
            type: string
            path: config.vcl

You choose any valid name and type. When the app is deployed, the app can access the Varnish service over HTTP to get diagnostic information. The following paths are available:

• /: returns any error logged when generating the VCL template failed
• /config: returns the generated VCL template
• /stats: returns the output of varnishstat
• /logs: returns a streaming response of varnishlog

To access the Varnish stats endpoint from the command line:

1. Connect to your stats app using SSH: platform ssh --app stats-app (replace stats-app with the name you gave the app).
2. Display the relationships array with echo $PLATFORM_RELATIONSHIPS | base64 -d | jq .,
3. Query Varnish with curl HOST:PORT/stats, replacing HOST and PATH with the values from Step 2.