This Caddy module performs substring and regular expression replacements on response bodies, hence the name replace_response.

By default, this module operates in "buffer" mode. This is not very memory-efficient, but it guarantees we can always set the correct Content-Length header because we can buffer the output to know the resulting length before writing the response. If you need higher efficiency, you can enable "streaming" mode. When performing replacements on a stream, the Content-Length header may be removed because it is not always possible to know the correct value, since the results are streamed directly to the client and headers must be written before the body.

Module name: http.handlers.replace_response

Substring substitution:

{
	"handler": "replace_response",
	"replacements": [
		{
			"search": "Foo",
			"replace": "Bar"
		}
	]
}

Regular expression replacement:

{
	"handler": "replace_response",
	"replacements": [
		{
			"search_regexp": "\\s+foo(bar|baz)\\s+",
			"replace": " foo $1 "
		}
	]
}

Same, but with streaming mode (we just set "stream": true in the handler):

{
	"handler": "replace_response",
	"replacements": [
		{
			"search_regexp": "\\s+foo(bar|baz)\\s+",
			"replace": " foo $1 "
		}
	],
	"stream": true
}

This module has Caddyfile support. It registers the replace directive. Make sure to order the handler directive in the correct place in the middleware chain; usually this works well:

{
	order replace after encode
}

Syntax:

replace [stream | [re] <search> <replace>] {
	stream
	[re] <search> <replace>
}

• re indicates a regular expression instead of substring.
• stream enables streaming mode.

Simple substring substitution:

replace Foo Bar

Regex replacement:

replace re "\s+foo(bar|baz)\s+" " foo $1 "

Streaming mode:

replace stream {
	Foo Bar
}

Multiple replacements:

replace {
	Foo Bar
	re "\s+foo(bar|baz)\s+" " foo $1 "
	A B
}

• Regex matches longer than 2kb will not be replaced.

• Compressed responses (e.g. from an upstream proxy which gzipped the response body) will not be decoded before attempting to replace. To work around this, you may send the Accept-Encoding: identity request header to the upstream to tell it not to compress the response. For example:

  reverse_proxy localhost:8080 {
      header_up Accept-Encoding identity
  }