Tronic
commented
8 months ago I am not against this but there are complications. First of all, the compression is not async but would need extra threads to avoid blocking the server. Brotli would need to be an additional (optional) dependency, while gzip is available in the standard library. Checking and doing this for every request also costs, possibly more than the gain (even just checking/setting the headers and of course the compression itself with slow CPU and fast network).
Ideally static data (files or other common responses) should be precompressed - I wrote my own static file handler to replace Sanic's for this (it also keeps all static files and their compressed versions in RAM - loaded and compressed at each server startup to avoid storing .br on disk, because a few megabytes of RAM "wasted" don't cost anything).
Dynamic data compression per request can also be useful for large JSON data and such, and if added as a feature in Sanic, and this of course should not be done in case apps or other facilities already do compression. Streaming responses would be difficult to support.
Both types of compression can also be handled very efficiently by a proxy such as Nginx or Caddy. At least the latter supports static file and dynamic compression in gzip, brotli and zstd formats with a simple config directive.
ahopkins
robd003
dmckeone
Is there an existing issue for this?
Is your feature request related to a problem? Please describe.
When serving larger responses it would be faster to send a compressed response as long as the client supports it.
Describe the solution you'd like
When a response from Sanic is over a certain threshold (2kb for Brotli / 512 bytes - 1kb for Gzip) and the HTTP client's accept headers permit it choose either Brotli or Gzip to compress the response
Additional context
It would be great to do this natively in Sanic rather than having to rely on something like nginx