502 Bad Gateway from Nginx for large GitLab fork

Question

I'm running GitLab 6.0.0 through Nginx and can fork small repositories, but when I try to fork a large repository (2GB) I see a "502 Bad Gateway" page after about one minute.

/var/log/nginx/gitlab_error.log shows:

2013/08/29 12:21:33 [error] 25098#0: *221 upstream prematurely closed connection while reading response header from upstream, 
  client: 12.34.56.78, 
  server: myserver, 
  request: "POST /mygroup/myproject/fork HTTP/1.1", 
  upstream: "http://unix:/home/git/gitlab/tmp/sockets/gitlab.socket:/mygroup/myproject/fork", 
  host: "myserver", 
  referrer: "http://myserver/mygroup/myproject/fork"

Answer 1

Issue 1527 suggests a memory issue (and that memory requirement in mentioned in the doc).

It can also be because of an initial timeout:

I found this error would also occur because the unicorn workers would sometimes take 33 seconds to start and they're configured to timeout after 30 seconds.

You can modify unicorn config file /home/git/gitlab/config/unicorn.rb :

timeout 300

In your NGiNX config, you can also add:

proxy_connect_timeout 300;
proxy_read_timeout 300;

If you have a /etc/nginx/fastcgi_params file with your NGiNX, you can add:

fastcgi_connect_timeout 60;
fastcgi_send_timeout 180;
fastcgi_read_timeout 180;
fastcgi_buffer_size 156 16k;
fastcgi_buffers 256 16k;
fastcgi_busy_buffers_size 256k;
fastcgi_temp_file_write_size 256k;
fastcgi_intercept_errors on;
fastcgi_max_temp_file_size 0

fastcgi_pass            unix:/tmp/fpm.sock;

---

Note that after a 502, and after making the fixes mentioned above, it is a good idea to clear the browser cache before trying again to access gitlab.

Answer 2

I did the same as your guide. But have little bit change on the fastcgi configuration.

fastcgi_pass  unix:/home/git/gitlab/tmp/sockets/gitlab.socket;

Thank you very much for your guide!

TWINQ78