3

Recently, I've a problem with the internal http(s) load balancer on GCP, about the timeout of backend service (an instance group). After 300 seconds, the API calling to LB will be failed with 408 HTTP response.

  • I have an internal HTTPS load balancer (LB) on GCP.
  • First, I call a quick API through LB, that worked normally.
  • Then I set timeout for backend service to 10 seconds then call the slow API (say, 500 seconds to complete the request). The error response happens after 10 seconds as expected.
  • But when I set the timeout to 1000 seconds and call the slow API, I received the error timeout response after only exactly 300 seconds.

I also increase the Connection draining timeout to 1000 seconds, but it still doesn't work.

Is the any parameter I need to set to allow API with reply time more than 300 seconds? Thank you.

Anbu Thirugnana Sekar
  • 105
  • 5
txuan
  • 31
  • 1
  • 5

1 Answers1

2

As I understand that you have configured a Backend service timeout to the 1000 seconds which represents the amount of time that the load balancer waits for the backend to return a complete HTTP response.

But it seems that the issue you are getting with error 408 does not appear to be related to Backend service timeout though, but to a TCP session timeout. The TCP session timeout is configurable at the web server software end, used by your backends.

This TCP session timeout is also known as the keepalive timeout and you must configure this timeout at the web server software end, used by your backends. So that its keepalive timeout is longer than 600 seconds to prevent connections from being closed prematurely by the backend services, which can help to keep the TCP session open for a longer duration by reusing the open connection.

Kindly refer the below link to find more detailed information related above solution:

[1] https://cloud.google.com/load-balancing/docs/backend-service#backend_service_settings

[2] https://cloud.google.com/load-balancing/docs/https#timeouts_and_retries

[3] https://cloud.google.com/compute/docs/troubleshooting/general-tips#idle-connections

Ahemad Pasha Mohammed
  • 46
  • 1
  • It's not my web server software issue, I tried to call API directly to the VM of my backend service by its IP, not through the Load Balancer, and the response return properly after 500 seconds. – txuan May 26 '21 at 14:28