Hi. I am using Istio ingress gateway (istio version 1.18.2), and creating GCP Network (Passthrough) load balancer by using "networking.gke.io/load-balancer-type: Internal"
annotation in the service.
The load balancer health checks default parameters are Interval 8 seconds, Timeout 1 second, Healthy threshold 1 success, Unhealthy threshold 3 consecutive failures.
When I run load testing, using fortio, and during testing I perform rollout restart for istio ingress gateway deployment, there will be about 20% failures.
If I change the Interval to 4 seconds, in the health checks load balancer parameters, I get a result of 10% errors.
It is also worth noting that changing parameters of istio itself such as
terminationDrainDuration: 30s
or
EXIT_ON_ZERO_ACTIVE_CONNECTIONS: 'true'
have no significant effect on the test result.
So what are the recommendations when using istio ingress gateway and GCP Network load balancer?
And how can I minimize the number of errors during restart of Istio ingress pods?
Thank you,
Oleksandr