gzip

gzip

Compresses context.response.body with gzip. It is a response-phase node — place it after upstream (and after any body-mutating node such as response-rewrite), before client. The node performs both the request's Accept-Encoding check and the compression itself, in process.

Configuration

All keys are optional.

Key	Type	Default	Description
types	array of strings, or "*"	["text/html"]	Response content types to compress. "*" matches any. The response content-type is compared with parameters (;charset=...) stripped. A response without a content-type header is never compressed.
min_length	integer ≥ 1	20	Bodies shorter than this many bytes are not compressed.
comp_level	integer 1-9	1	gzip compression level.
vary	bool	false	When true, appends Vary: Accept-Encoding to compressed responses.

type: gzip
config:
  types: ["text/html", "application/json"]
  min_length: 20
  comp_level: 6
  vary: true

Out-of-range levels, min_length: 0, an empty types array, or a types string other than "*" are rejected at config load.

Behavior

This plugin never fails at execution time — it always exits through the success port. The response passes through unchanged when any of these hold:

• the request's Accept-Encoding does not allow gzip — the token (or *) must be listed with a non-zero q value; a missing header means no compression;
• the response already carries a content-encoding (never compress twice, and never compress opaque pre-encoded bytes — even for encodings featherbit does not know);
• the response content-type is absent or does not match types;
• the body is shorter than min_length.

Otherwise the body is gzip-compressed at comp_level, content-encoding: gzip is set, the stale content-length is removed (the server layer recomputes it from the compressed body), and Vary: Accept-Encoding is appended when vary is set. A codec failure logs a warning and leaves the response untouched.

Behavior notes: http_version and buffers are not supported (featherbit responses are fully buffered), and min_length is compared against the actual buffered body length rather than the upstream Content-Length header. Unlike the brotli node, the gzip node does not weaken ETag headers.

The plugin does not read or write context.message or context.errors.