When does Rails respond with 'transfer-encoding' vs. 'content-length'?

Question

I'm building an API on Rails version 4.1.7/Nginx that responds to request from an iOS app. We're seeing some weird caching on the client and we think it has something to do with a small difference in the response that Rails is sending back. My questions...

1) I want to understand why, for the exact same request (with only the Authorization header value changed), Rails sends back transfer-encoding: chunked sometimes and Content-Length: <number> sometimes? I thought that maybe it had something to do with the response size, but in the example responses whose headers I've pasted below, the data returned in the body is EXACTLY the same.

2) Is there a way to force it to use Content-Length? We think that that will fix our caching issues in our iOS app.

Response #1

HTTP/1.1 200 OK
Cache-Control: max-age=0, private, must-revalidate
Content-Type: application/json; charset=utf-8
Date: Wed, 18 Mar 2015 00:59:31 GMT
ETag: "86f277ea63295460d4f3bed9a073eaa2"
Server: nginx/1.6.2
Status: 200 OK
X-Content-Type-Options: nosniff
X-Frame-Options: SAMEORIGIN
X-Request-Id: dd36f139-1986-4da6-9645-4438d41e74b0
X-Runtime: 0.123865
X-XSS-Protection: 1; mode=block
transfer-encoding: chunked
Connection: keep-alive

Request #2

HTTP/1.1 200 OK
Cache-Control: max-age=0, private, must-revalidate
Content-Type: application/json; charset=utf-8
Date: Wed, 18 Mar 2015 00:59:36 GMT
ETag: "86f277ea63295460d4f3bed9a073eaa2"
Server: nginx/1.6.2
Status: 200 OK
X-Content-Type-Options: nosniff
X-Frame-Options: SAMEORIGIN
X-Request-Id: 0cfd7705-157b-41b5-aa36-739bc6f8302e
X-Runtime: 0.092672
X-XSS-Protection: 1; mode=block
Content-Length: 2234
Connection: keep-alive

Elsey

Did Rails send the second response back, or did nginx? That is, is nginx doing its own caching? – Endogen

@MichaelHampton hmm... I'll look into it and report back... – Elsey

Answer 1

Both responses are valid according to HTTP 1.1, so you need to fix your client code that it can handle both. It is a bad idea to try to fix the server so that that it behave in a way that it does not trigger a bug in the client. The next version of nginx may behave differently, you users may even have proxies that change the transfer, maybe only when they do roaming and use a different provider.

If you want to do some finger-printing on the header, the ETag-header may help you, as the ETag should stay constant when the content of the response is not changed, regardless of the transfer.

The server typically sends in chunks when it calls a dynamic page, because it then does not need to create a buffer for the whole page and wait till all of the page is generated.

The server often send the response in one go if it has the buffer already, for example because it is in cache or the content is on a file and is not to big. Sending in one go is more efficient, on the other hand, an extra copy of the data to buffer the output needs more memory and is less efficient. So the server may even decide this according to the available memory.

Basrelief