Server Fault is a question and answer site for system and network administrators. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I use nginx 1.2.3 to proxy to a script:

proxy_set_header Host $host;
proxy_buffering off;
proxy_read_timeout 300s;
gzip off;

The scripts sends both Transfer-encoding: chunked and Content-Length: 251:

HTTP/1.0 307 Temporary Redirect
Content-length: 251
Pragma: no-cache
Location: /...
Cache-control: no-cache
Transfer-encoding: chunked

I need both, but nginx automatically removes the Content-Length:

HTTP/1.1 302 Found
Server: nginx/1.2.3
Content-Type: application/json; charset=utf-8
Content-Length: 58
Connection: keep-alive
Location: /...

As a result, the clients do not wait for the chunks to be sent. This used to work with an earlier version of nginx.

share|improve this question
What do the headers look like from the nginx proxy? – hrunting Feb 28 '13 at 4:10
which version did it used to work with? – cnst Mar 5 '13 at 9:35
It used to work with nginx 0.9.8 – Julien Mar 6 '13 at 3:53
You're violating HTTP protocol. It works with nginx 0.9.8, because till the 1.1.4 version it doesn't support chunked encoding at all. – VBart Sep 30 '13 at 22:05
up vote 7 down vote accepted

Unfortunately, I can't comment on cnst's post - so I'm going to answer here.

The nginx_http_proxy module by default talks with the upstream in HTTP/1.0. This can be changed with the directive proxy_http_version 1.1.

This might also be the cause for your script to return a HTTP/1.0 answer, although chunked coding and status code 307 don't exist in this version.

You shouldn't use chunked coding with a redirect either, as this doesn't really make sense.

Additionally, it seems like nginx doesn't pass chunks from the upstream to the client one by one, but it buffers the upstream's response. The Content-Length header field is ignored because it is against the definition. I had to look at the source code of the module because all this appears to be undocumented.

You may want to try out the nginx_tcp_proxy_module to proxy the chunked content as raw TCP data: Module at Github

UPDATE (10.04.14)
The nginx_http_proxy module has support for X-Accel-* headers, of which one (X-Accel-Buffering: yes|no) controls whether the response should be buffered or not.

Adding this header (X-Accel-Buffering: no) to the backend's response will cause nginx to directly pass chunks to the client.

This header allows to control buffering on an per-request basis.

The module also has a configuration directive proxy_buffering to enable or disable response buffering (not buffering means sending chunks will work).

Proxy buffering (both header and directive based) is documented here.

share|improve this answer
He should not do that even with nginx_tcp_proxy_module. It works with some browsers only because they are very error tolerant. – VBart Sep 30 '13 at 22:08
because all this appears to be undocumented Wrong. It's documented in RFC 2616. See 13.5.1. – VBart Sep 30 '13 at 22:10
@VBart Sure there are standards - but there's only very few information about how far particularly nginx implements them.The TCP proxy module is hust a suggested workaround. – Lukas Sep 30 '13 at 23:27

You have not specifically elaborated why your script needs chunked encoding in the first place, especially with a redirect response.

I see a multitude of problems here.

  • Transfer-Encoding: chunked is an HTTP/1.1 feature (and your script seems to be replying with an HTTP/1.0 header)

  • there is no 307 in HTTP/1.0

  • the whole purpose of chunked is that you don't know what your Content-Length would have been, so, chunked is used in place of providing the length within Content-Length, where instead lengths are provided within the body of the response, intermixed with the actual content; it would be pointless for a script to be generating both headers upfront

I'm not personally familiar with chunked, but as per the basic info at and also, I would guesstimate that your script's whole handling of the chunked encoding may be completely wrong.

If the above still doesn't cover it, and in all actuality otherwise, it is also unclear why a reply with a 307 or 302 http status code should be provided with a "weird" encoding. There was recently a similar discussion in the nginx mailing list about 410 Gone and other error pages always excluded from gzip compression, and I think the sentiment would equally apply here. (

share|improve this answer
I use it to make the user wait: I send chunks every seconds, so that the user will wait for the redirection for X seconds without getting a timeout – Julien Mar 6 '13 at 3:53
I'd advise you to first fix HTTP/1.0 to HTTP/1.1 (these things do make a difference), and make sure your chunked encoding is not improper. The newer version of nginx probably discards some headers on which you depend because they're wrong. – cnst Mar 7 '13 at 3:00

As Lukas alludet to, HTTP 1.1 prohibits Content-Length if there's a Transfer-Encoding set.


   3.If a Content-Length header field (section 14.13) is present, its
     decimal value in OCTETs represents both the entity-length and the
     transfer-length. The Content-Length header field MUST NOT be sent
     if these two lengths are different (i.e., if a Transfer-Encoding
     header field is present). If a message is received with both a
     Transfer-Encoding header field and a Content-Length header field,
     the latter MUST be ignored.
share|improve this answer

I had the same issue streaming mp4 file through html5 video tag.

Safari and Firefox behaved normally whereas Chrome was triggering ERR_CONTENT_LENGTH_MISMATCH at some point (but it allowed me to wach several minutes of the video before failing).

The problem didn't reproduce after I turned off cache control for mp4 files.

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.