what do they mean and can you provide an demo

I do not think we should expose a runtime key here. Also for how many services do you have to configure this? And does it differ from service to service?

I do not think we should expose a runtime key here. Also for how many services do you have to configure this? And does it differ from service to service?

As far as I remember runtime_key parameter is mandatory if we want to use an aggression parameter (which is the one we really need).

what do they mean and can you provide an demo

We're using Java microservices and pods need a warmup phase in order to have full performance.

In practice the goal is to avoid giving 100% of the traffic to a new READY pod. Leveraging a slow start allow us to give first a certain % of the traffic then ramp-up progressively to 100%.

For this first attempt we tried to use the LoadBalancerSettings feature from Istio. This allow us to specify the duration of the warmup. However in this config we can’t configure 2 important options because they are not exposed by Istio API :

min_weight_percent : specifies the initial percent of origin load, if not present, it is default to 10%.

aggression : will defined the evolution of the % of traffic sent to the pods from min_weight_percent to 100%, by default the the ramp-up curve is linear, but by customising it we can achieve exponential type of curve.

The result (sorry I don't have a picture to illustrate that) is that 10% of traffic still too much : our latency increase a lot and impact our users.

To check if the 2 parameters mentioned above impact our traffic, we used an EnvoyFilter that we applied to 3 clients of our app.

Deploying this config from only a portion of our traffic (roughly 75%) with a slow_start_window of 3 minutes and a min_weight_percent of 1% we have been able to observe an impact were we can see the progressive ramp-up of the traffic.

I do not think runtime is must https://www.envoyproxy.io/docs/envoy/latest/api-v3/config/core/v3/base.proto#envoy-v3-api-msg-config-core-v3-runtimedouble

it is actually mandatory if we want to use the agression parameter. If I try this EnvoyFilter :

apiVersion: networking.istio.io/v1alpha3
kind: EnvoyFilter
metadata:
  name: h2-control
spec:
  configPatches:
    - applyTo: CLUSTER
      match:
        cluster:
          name: "outbound|8080||http-echo.infra.svc.cluster.local"
      patch:
        operation: MERGE
        value:
          name: "outbound|8080||http-echo.infra.svc.cluster.local"
          lbPolicy: LEAST_REQUEST
          leastRequestLbConfig:
            slowStartConfig:
              min_weight_percent: { value: 99 }
              slow_start_window: "12s"
              aggression: { default_value: 2  }
  workloadSelector:
    labels:
      app: landing-f.gaudet

Then I have this warning in the logs :

landing-f.gaudet istio-proxy {"level":"warning","time":"2024-06-04T04:41:49.406026Z","scope":"envoy config","msg":"gRPC config for type.googleapis.com/envoy.config.cluster.v3.Cluster rejected: Proto constraint validation failed (ClusterValidationError.LeastRequestLbConfig: embedded message failed validation | caused by LeastRequestLbConfigValidationError.SlowStartConfig: embedded message failed validation | caused by SlowStartConfigValidationError.Aggression: embedded message failed validation | caused by RuntimeDoubleValidationError.RuntimeKey: value length must be at least 1 characters):

and the config is not applied. However, if I setup the runtime key

          leastRequestLbConfig:
            slowStartConfig:
              min_weight_percent: { value: 99 }
              slow_start_window: "12s"
              aggression: { default_value: 2, runtime_key: "(" }

Then the config is successfully applied :

istioctl pc cluster landing-f.gaudet.infra --fqdn http-echo.infra.svc.cluster.local -ojson | jq ".[].leastRequestLbConfig"


{
  "slowStartConfig": {
    "slowStartWindow": "12s",
    "aggression": {
      "defaultValue": 2,
      "runtimeKey": "("
    },
    "minWeightPercent": {
      "value": 99
    }
  }
}