Compressor

Compressor

Compressor is an HTTP filter which enables Envoy to compress dispatched data from an upstream service upon client request. Compression is useful in situations when bandwidth is scarce and large payloads can be effectively compressed at the expense of higher CPU load or offloading it to a compression accelerator.

Configuration

v3 API reference
This filter should be configured with the name envoy.filters.http.compressor.

How it works

When compressor filter is enabled, request and response headers are inspected to determine whether or not the content should be compressed. The content is compressed and then sent to the client with the appropriate headers, if response and request allow.

Currently the filter supports gzip, brotli and zstd compression only. Other compression libraries can be supported as extensions.

An example configuration of the filter may look like the following:

http_filters:
- name: envoy.filters.http.compressor
  typed_config:
    "@type": type.googleapis.com/envoy.extensions.filters.http.compressor.v3.Compressor
    response_direction_config:
      common_config:
        min_content_length: 100
        content_type:
          - text/html
          - application/json
      disable_on_etag_header: true
    request_direction_config:
      common_config:
        enabled:
          default_value: false
          runtime_key: request_compressor_enabled
    compressor_library:
      name: text_optimized
      typed_config:
        "@type": type.googleapis.com/envoy.extensions.compression.gzip.compressor.v3.Gzip
        memory_level: 3
        window_bits: 10
        compression_level: BEST_COMPRESSION
        compression_strategy: DEFAULT_STRATEGY

By default request compression is disabled, but when enabled it will be skipped if:

A request does not contain a content-type value that matches one of the selected mime-types, which default to application/javascript, application/json, application/xhtml+xml, image/svg+xml, text/css, text/html, text/plain, text/xml.
content-length header is not present in the request.
A request contains a content-encoding header.
A request contains a transfer-encoding header whose value includes a known compression name.

By default response compression is enabled, but it will be skipped when:

A request does NOT contain accept-encoding header.
A request includes accept-encoding header, but it does not contain “gzip” or “*”.
A request includes accept-encoding with “gzip” or “*” with the weight “q=0”. Note that the “gzip” will have a higher weight then “*”. For example, if accept-encoding is “gzip;q=0,*;q=1”, the filter will not compress. But if the header is set to “*;q=0,gzip;q=1”, the filter will compress.
A request whose accept-encoding header includes any encoding type with a higher weight than “gzip“‘s given the corresponding compression filter is present in the chain.
A response contains a content-encoding header.
A response contains a cache-control header whose value includes “no-transform”.
A response contains a transfer-encoding header whose value includes a known compression name.
A response does not contain a content-type value that matches one of the selected mime-types, which default to application/javascript, application/json, application/xhtml+xml, image/svg+xml, text/css, text/html, text/plain, text/xml.
Neither content-length nor transfer-encoding headers are present in the response.
Response size is smaller than 30 bytes (only applicable when transfer-encoding is not chunked).

Please note that in case the filter is configured to use a compression library extension other than gzip it looks for content encoding in the accept-encoding header provided by the extension.

When response compression is applied:

The content-length is removed from response headers.
Response headers contain “transfer-encoding: chunked”, and “content-encoding” with the compression scheme used (e.g., gzip).
The “vary: accept-encoding” header is inserted on every response.

Also the “vary: accept-encoding” header may be inserted even if compression is not applied due to incompatible “accept-encoding” header in a request. This happens when the requested resource still can be compressed given compatible “accept-encoding”. Otherwise, if an uncompressed response is cached by a caching proxy in front of Envoy, the proxy won’t know to fetch a new incoming request with compatible “accept-encoding” from upstream.

When request compression is applied:

content-length is removed from request headers.
content-encoding with the compression scheme used (e.g., gzip) is added to request headers.

Using different compressors for requests and responses

If different compression libraries are desired for requests and responses, it is possible to install multiple compressor filters enabled only for requests or responses. For instance:

http_filters:
# This filter is only enabled for responses.
- name: envoy.filters.http.compressor
  typed_config:
    "@type": type.googleapis.com/envoy.extensions.filters.http.compressor.v3.Compressor
    request_direction_config:
      common_config:
        enabled:
          default_value: false
          runtime_key: request_compressor_enabled
    compressor_library:
      name: for_response
      typed_config:
        "@type": type.googleapis.com/envoy.extensions.compression.gzip.compressor.v3.Gzip
        memory_level: 3
        window_bits: 10
        compression_level: BEST_COMPRESSION
        compression_strategy: DEFAULT_STRATEGY
# This filter is only enabled for requests.
- name: envoy.filters.http.compressor
  typed_config:
    "@type": type.googleapis.com/envoy.extensions.filters.http.compressor.v3.Compressor
    response_direction_config:
      common_config:
        enabled:
          default_value: false
          runtime_key: response_compressor_enabled
    request_direction_config:
      common_config:
        enabled:
          default_value: true
          runtime_key: request_compressor_enabled
    compressor_library:
      name: for_request
      typed_config:
        "@type": type.googleapis.com/envoy.extensions.compression.gzip.compressor.v3.Gzip
        memory_level: 9
        window_bits: 15
        compression_level: BEST_SPEED
        compression_strategy: DEFAULT_STRATEGY

Statistics

Every configured Compressor filter has statistics rooted at <stat_prefix>.compressor.<compressor_library.name>.<compressor_library_stat_prefix>.<direction_prefix>.* with the following:

Name	Type	Description
compressed	Counter	Number of requests compressed.
not_compressed	Counter	Number of requests not compressed.
total_uncompressed_bytes	Counter	The total uncompressed bytes of all the requests that were marked for compression.
total_compressed_bytes	Counter	The total compressed bytes of all the requests that were marked for compression.
content_length_too_small	Counter	Number of requests that accepted the compressor encoding but did not compress because the payload was too small.

In addition to the statics common for requests and responses there are statistics specific to responses only:

Name	Type	Description
no_accept_header	Counter	Number of requests with no accept header sent.
header_identity	Counter	Number of requests sent with “identity” set as the accept-encoding.
header_compressor_used	Counter	Number of requests sent with filter’s configured encoding set as the accept-encoding.
header_compressor_overshadowed	Counter	Number of requests skipped by this filter instance because they were handled by another filter in the same filter chain.
header_wildcard	Counter	Number of requests sent with “” set as the accept-encoding*.
header_not_valid	Counter	Number of requests sent with a not valid accept-encoding header (aka “q=0” or an unsupported encoding type).
not_compressed_etag	Counter	Number of requests that were not compressed due to the etag header. disable_on_etag_header must be turned on for this to happen.