Set timeout for the whole request process (including retries)?

1

For each incoming request, we want Kong to return a response after a maximum of 60 seconds.
Therefore, we are using the Service entity default values for connect_timeout, write_timeout and read_timeout.
The service retries attribute is also set to the default of 5.

However, if the upsteam is not responding within 60 seconds, Kong will try again - up to 5 times. In this case, Kong returns a response to the client only after 60*6 seconds…

We can set the retries attribute to 0 to fix this but then we will miss the benefit of the retry in case of poor connection etc’.

Is there an option to set up a timeout for the whole request process including retries? Or maybe to restrict the retries mechanism only to failures that are not timeout?

1 Like
2

No.

The point to note is that by default Kong only retries for connection timeouts. So you can do the math and set this up accordingly.

I’ve a question for you, what did you expect Kong to rety on other than timeouts?

3

Thanks for the answer.

I have created an upstream server with an endpoint that responds only after 30s (using sleep) and set Kong’s ‘read_timeout’ for this service to 15s.

I noticed that once the ‘read_timeout’ ended, Kong sent another request and so on - according to the specified retries attribute.

Therefore, I concluded that Kong also retries on ‘read_timeout’. Is that correct? If not, what is the explanation for this behavior?

4

Kong will retry for connect and read timeouts (as defined on Service object in Kong) while communicating with upstreams.
I’m not sure what is the behavior of write timeouts.