http-proxy with gzip chunked server responses

[wtritch]

Node Version: 4.4.4
http-proxy version: 1.13.2

When proxying a browser request to a JSON RESTful endpoint that returns gzip content-encoding and chunked transfer-encoding the browser is receiving invalid bytes back from the proxy. This causes a browser error: ERR_CONTENT_DECODING_FAILED

It looks like the server is returning valid bytes to the proxy, but when http-proxy pipes the data from the proxy response to the browser response, the bytes are being UTF-8 encoded. To work around this I added an onProxyRes handler that hijacks the browser response "write" and "end" methods to force "binary" encoding when receiving a "content-encoding" header that contains gzip.

This feels like a really ugly solution:

function handleGzip(proxyRes, req, res) {
    var gzipped = /gzip/.test(proxyRes.headers["content-encoding"]);
    if (gzipped) {
        res.write = (function(override) {
            return function(chunk, encoding, callback) {
                override.call(res, chunk, "binary", callback);
            };
        })(res.write);
        res.end = (function(override) {
            return function(chunk, encoding, callback) {
                override.call(res, chunk, "binary", callback);
            };
        })(res.end);
    }
}

Oddly enough, this is only an issue when the Accept request header contains "text/html" or "application/xhtml+xml" When no Accept request header is specified, the response is properly proxied.

I was hoping to use ServerResponse#setDefaultEncoding("binary") but it appears ServerResponse doesn't implement that Writable method.

I'm not sure if this is an issue with http-proxy or the underlying node http implementation, but please help.

[marmelo]

Hi all,

The workaround I am using is disabling compression in the DevServer:

module.exports = {
  ...
  devServer: {
    ...
    compress: false
  }
};

This results in all responses to the browser not being compressed (usually gzip) which is not efficient.
Do you know a more efficient workaround?

What about a real fix for this?
@johnyb did you sent a pull request with your fix?

Thanks.

[vansen]

I just encountered this issue as well. Since that post is already a bit older - are there any recent suggestions on how to handle this properly? In my case it only happens when Transfer-Encoding is chunked. Thanks