We are using nginx to load balance requests to our application. We have found that nginx switches to a different upstream server when requests time out (good). However it does this for PUT and POST requests which can cause undesirable results (data stored twice). Is it possible to configure nginx to only retry GET requests on timeout? Or is there another way to solve the problem?

Our configuration is as follows:

upstream mash {

server {
    location / {
        proxy_pass http://mash/;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;        

It's become default behaviour starting since version 1.9.13

To change it back manually you can use:

proxy_next_upstream error timeout non_idempotent;

I know i'm pretty late to the game, but for me this is the top result when searching for this problem, so i wanted to share my solution.

This uses the if directive (with one of the few valid use cases) combined with the custom error handler:

upstream backend {
    server backend1;
    server backend2;

server {
    server_name proxy;

    location / {
        error_page 598 = @retry;
        error_page 599 = @no_retry;
        if ($request_method = POST) {
            return 599;
        return 598;

    location @retry {
        proxy_pass http://backend;

    location @no_retry {
        proxy_pass http://backend;
        proxy_next_upstream off;

Please see here for doc: proxy_next_upstream

Please note this is an untested gist


  • Actually it didn't work: Nginx says "proxy_next_upstream not allowed here". I tried moving the if blocks into location and got the same error. Using "proxy_next_upstream error" in either location on its own works. – David Tinker Aug 6 '13 at 9:33
  • that is very odd since there documentation clearly says it works in the location context – WojonsTech Aug 6 '13 at 20:17
  • it seems to be the if (...) { } around proxy_next_upstream that nginx doesn't like – David Tinker Aug 7 '13 at 9:18
  • Has anyone tested this? 4 upvotes but it doesn't seem to adhere to valid use cases here: nginx.com/resources/wiki/start/topics/depth/ifisevil – EoghanM Feb 5 '16 at 12:58

use proxy_method directive

refer to: http://nginx.org/en/docs/http/ngx_http_proxy_module.html#proxy_method

  • 2
    It's generally recommended to include the useful information from a link in your answer such that it does not explicitly rely on the link to be useful – BE77Y Jun 17 '15 at 15:52
  • 1
    Welcome to Server Fault! Whilst this may theoretically answer the question, it would be preferable to include the essential parts of the answer here, and provide the link for reference. – Mark Henderson Jun 17 '15 at 22:51

I have same problem in my tomcat server. proxy time out when long request occur. i solved my problem by using proxy_read_timeout. when increase timeout then my request never time outed & not occurred any problem. default time out 60s. reference

location / {
    proxy_pass  http://xxxxxxxxxx.com;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header Host $http_host;
            proxy_set_header X-Forwarded-Proto https;
            proxy_redirect off;
            proxy_connect_timeout      800;
            proxy_send_timeout         800;
            proxy_read_timeout         240;     
  • 1
    This doesn't answer the question at all. Your problems are nothing alike. – Sven Jun 8 '15 at 21:29

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

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