NSURLErrorDomain Code=-1004 for few seconds after app start up

Question

I'm getting "NSURLErrorDomain Code=-1004" error with Alamofire API calls, but only for few seconds after app started (or took a rest for few minutes while the app's opened and make a call after that)

If I try to make the same call after few seconds, everything works fine. I searched all the Stack Overflow questions and checked all the possible causes below:

1. No problem with Internet connection
2. "App Transport Security Settings" are correct and server's using https (I also tried "NSAllowsArbitraryLoads=true" but that didn't help)
3. APIs working fine

My gut feeling is that getting the network settings takes few seconds and when I make an API call before that's done, it just fails immediately. OR.. I'm using a Websocket in background that could be related?

FAILURE: Error Domain=NSURLErrorDomain Code=-1004 "Could not connect to the server." UserInfo={NSUnderlyingError=0x137d39380 {Error Domain=kCFErrorDomainCFNetwork Code=-1004 "(null)" UserInfo={NSErrorPeerAddressKey={length = 16, capacity = 16, bytes = 0x100201bb341a9f540000000000000000}, _kCFStreamErrorCodeKey=-2200, _kCFStreamErrorDomainKey=4}}, NSErrorFailingURLStringKey=[FILTERED], NSErrorFailingURLKey=[FILTERED], _kCFStreamErrorDomainKey=4, _kCFStreamErrorCodeKey=-2200, NSLocalizedDescription=Could not connect to the server.}

Any suggestions?

UPDATED

Found that app makes 4 requests on launch, and 1 or 2 of them randomly fails, and I checked Nginx access and error log and there's no log for the failed calls at all.

Answer 1

We have the same issue here with Nginx 1.10.0 (and 1.9.15), iOS 9.3.1 using HTTP/2 with TLS 1.2.

The issue goes away with HTTP/1.1 and it's also working with HTTP/2 in Nginx version up to 1.9.14.

Answer 2

Nginx 1.11.0 Mainline is now available with the fix included mentioned earlier in this topic;

Change: HTTP/2 clients can now start sending request body immediately; the "http2_body_preread_size" directive controls size of the buffer used before nginx will start reading client request body.

I tested it and for me this release now works correctly again.

Answer 3

This seems like a confirmed bug in nginx 1.10. An issue about it can be found on the bug tracker of nginx at https://trac.nginx.org/nginx/ticket/979. The actual issue can be found at https://trac.nginx.org/nginx/ticket/959

You might want to consider switching to the 1.9 branch which has releases which do work. Hopefully nginx will release a 1.10.1 version soon which does not have this bug.

The issue actually only occurs on iOS; Android, Windows and OSX itself seem to have no problems negotiating a valid http2 connection.

Answer 4

I can also confirm that the nginx 1.9.15 not works correctly. Some calls always got "Could not connect to the server", and after revert to nginx 1.9.12 everything works fine.