From 9fea1530ad3ad8fb89eb3d632a2199d74bd0afd5 Mon Sep 17 00:00:00 2001
From: Claes Mogren <mogren@amazon.com>
Date: Sun, 14 Jun 2020 22:37:10 -0700
Subject: [PATCH] Update probe settings

* Reduce readiness probe startup delay
* Increase liveness polling period
* Reduce shutdown grace period to 10 seconds
---
 config/master/aws-k8s-cni-cn.yaml            | 5 +++--
 config/master/aws-k8s-cni-us-gov-east-1.yaml | 5 +++--
 config/master/aws-k8s-cni-us-gov-west-1.yaml | 5 +++--
 config/master/aws-k8s-cni.yaml               | 5 +++--
 config/master/manifests.jsonnet              | 7 +++++--
 5 files changed, 17 insertions(+), 10 deletions(-)

diff --git a/config/master/aws-k8s-cni-cn.yaml b/config/master/aws-k8s-cni-cn.yaml
index 8ffc831a8f..840cf4c103 100644
--- a/config/master/aws-k8s-cni-cn.yaml
+++ b/config/master/aws-k8s-cni-cn.yaml
@@ -118,7 +118,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 60
         "name": "aws-node"
         "ports":
         - "containerPort": 61678
@@ -128,7 +128,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 1
         "resources":
           "requests":
             "cpu": "10m"
@@ -161,6 +161,7 @@
           "name": "cni-bin-dir"
       "priorityClassName": "system-node-critical"
       "serviceAccountName": "aws-node"
+      "terminationGracePeriodSeconds": 10
       "tolerations":
       - "operator": "Exists"
       "volumes":
diff --git a/config/master/aws-k8s-cni-us-gov-east-1.yaml b/config/master/aws-k8s-cni-us-gov-east-1.yaml
index 11ab723eba..f3e7b6e843 100644
--- a/config/master/aws-k8s-cni-us-gov-east-1.yaml
+++ b/config/master/aws-k8s-cni-us-gov-east-1.yaml
@@ -118,7 +118,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 60
         "name": "aws-node"
         "ports":
         - "containerPort": 61678
@@ -128,7 +128,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 1
         "resources":
           "requests":
             "cpu": "10m"
@@ -161,6 +161,7 @@
           "name": "cni-bin-dir"
       "priorityClassName": "system-node-critical"
       "serviceAccountName": "aws-node"
+      "terminationGracePeriodSeconds": 10
       "tolerations":
       - "operator": "Exists"
       "volumes":
diff --git a/config/master/aws-k8s-cni-us-gov-west-1.yaml b/config/master/aws-k8s-cni-us-gov-west-1.yaml
index cfced1ca33..9f9815bae9 100644
--- a/config/master/aws-k8s-cni-us-gov-west-1.yaml
+++ b/config/master/aws-k8s-cni-us-gov-west-1.yaml
@@ -118,7 +118,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 60
         "name": "aws-node"
         "ports":
         - "containerPort": 61678
@@ -128,7 +128,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 1
         "resources":
           "requests":
             "cpu": "10m"
@@ -161,6 +161,7 @@
           "name": "cni-bin-dir"
       "priorityClassName": "system-node-critical"
       "serviceAccountName": "aws-node"
+      "terminationGracePeriodSeconds": 10
       "tolerations":
       - "operator": "Exists"
       "volumes":
diff --git a/config/master/aws-k8s-cni.yaml b/config/master/aws-k8s-cni.yaml
index 2835d27916..2367bb518b 100644
--- a/config/master/aws-k8s-cni.yaml
+++ b/config/master/aws-k8s-cni.yaml
@@ -118,7 +118,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 60
         "name": "aws-node"
         "ports":
         - "containerPort": 61678
@@ -128,7 +128,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 1
         "resources":
           "requests":
             "cpu": "10m"
@@ -161,6 +161,7 @@
           "name": "cni-bin-dir"
       "priorityClassName": "system-node-critical"
       "serviceAccountName": "aws-node"
+      "terminationGracePeriodSeconds": 10
       "tolerations":
       - "operator": "Exists"
       "volumes":
diff --git a/config/master/manifests.jsonnet b/config/master/manifests.jsonnet
index ba23f39107..7bb8e0f55d 100644
--- a/config/master/manifests.jsonnet
+++ b/config/master/manifests.jsonnet
@@ -100,6 +100,7 @@ local awsnode = {
         },
         spec: {
           priorityClassName: "system-node-critical",
+          terminationGracePeriodSeconds: 10,
           affinity: {
             nodeAffinity: {
               requiredDuringSchedulingIgnoredDuringExecution: {
@@ -143,9 +144,11 @@ local awsnode = {
                 exec: {
                   command: ["/app/grpc-health-probe", "-addr=:50051"],
                 },
-                initialDelaySeconds: 35,
+                initialDelaySeconds: 1,
+              },
+              livenessProbe: self.readinessProbe + {
+                initialDelaySeconds: 60,
               },
-              livenessProbe: self.readinessProbe,
               env_:: {
                 AWS_VPC_ENI_MTU: "9001",
                 AWS_VPC_K8S_CNI_CONFIGURE_RPFILTER: "false",