While using internal load balancer, scale out event happens (due to traffic metric activation), a new instance is created.
At this point I noticed that all existing instances do not get traffic AT ALL from the load balancer for at least 10 seconds (traffic drops to zero).
This cause a sever traffic failures to the backend and only after a few minutes when the scale out event completes the old instanced get traffic back.
Traffic drops to zero on existing instances when scale out event happens
Hello
Is this a TCP or HTTP/S load balancer? The backend is GKE or a managed instance group?
Let me know if you want to have a Google Meet session to talk about the issue.
Hi Yuval,
As for your questions:
1. This is an HTTPS load balancer
2. The backend is Managed instance group
Can we schedule a Google Meet for next week so I can show a live demo of this issue?
Yuval, can we have a short session to show you this issue and maybe get some insights?
This prevents us from using any scaling topology in our solution as we are traffic affected each time scale out event happens .
Seeing the same behavior in my lab. A short while after my MIG scales out, traffic to all existing instances is temporarily drained (for about 10-20 seconds). No clue why it happens yet.
Reply
Enter your username or e-mail address. We'll send you an e-mail with instructions to reset your password.