From 3d8d14a4f1c30e0258297b97c45f29816656be98 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            | 4 +++-
 config/master/aws-k8s-cni-us-gov-east-1.yaml | 4 +++-
 config/master/aws-k8s-cni-us-gov-west-1.yaml | 4 +++-
 config/master/aws-k8s-cni.yaml               | 4 +++-
 config/master/manifests.jsonnet              | 9 ++++++++-
 5 files changed, 20 insertions(+), 5 deletions(-)

diff --git a/config/master/aws-k8s-cni-cn.yaml b/config/master/aws-k8s-cni-cn.yaml
index ec2c9e14ef9..62aface3dd6 100644
--- a/config/master/aws-k8s-cni-cn.yaml
+++ b/config/master/aws-k8s-cni-cn.yaml
@@ -112,6 +112,7 @@
             - "/app/grpc-health-probe"
             - "-addr=:50051"
           "initialDelaySeconds": 35
+          "periodSeconds": 60
         "name": "aws-node"
         "ports":
         - "containerPort": 61678
@@ -121,7 +122,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 10
         "resources":
           "requests":
             "cpu": "10m"
@@ -154,6 +155,7 @@
           "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 26386d68f58..862d37a133b 100644
--- a/config/master/aws-k8s-cni-us-gov-east-1.yaml
+++ b/config/master/aws-k8s-cni-us-gov-east-1.yaml
@@ -112,6 +112,7 @@
             - "/app/grpc-health-probe"
             - "-addr=:50051"
           "initialDelaySeconds": 35
+          "periodSeconds": 60
         "name": "aws-node"
         "ports":
         - "containerPort": 61678
@@ -121,7 +122,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 10
         "resources":
           "requests":
             "cpu": "10m"
@@ -154,6 +155,7 @@
           "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 9a8cd99c596..4eccde11e7f 100644
--- a/config/master/aws-k8s-cni-us-gov-west-1.yaml
+++ b/config/master/aws-k8s-cni-us-gov-west-1.yaml
@@ -112,6 +112,7 @@
             - "/app/grpc-health-probe"
             - "-addr=:50051"
           "initialDelaySeconds": 35
+          "periodSeconds": 60
         "name": "aws-node"
         "ports":
         - "containerPort": 61678
@@ -121,7 +122,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 10
         "resources":
           "requests":
             "cpu": "10m"
@@ -154,6 +155,7 @@
           "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 a4af8724b95..2292a99bf0e 100644
--- a/config/master/aws-k8s-cni.yaml
+++ b/config/master/aws-k8s-cni.yaml
@@ -112,6 +112,7 @@
             - "/app/grpc-health-probe"
             - "-addr=:50051"
           "initialDelaySeconds": 35
+          "periodSeconds": 60
         "name": "aws-node"
         "ports":
         - "containerPort": 61678
@@ -121,7 +122,7 @@
             "command":
             - "/app/grpc-health-probe"
             - "-addr=:50051"
-          "initialDelaySeconds": 35
+          "initialDelaySeconds": 10
         "resources":
           "requests":
             "cpu": "10m"
@@ -154,6 +155,7 @@
           "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 9a24843ffcf..b09e7e9f9b2 100644
--- a/config/master/manifests.jsonnet
+++ b/config/master/manifests.jsonnet
@@ -95,6 +95,7 @@ local awsnode = {
         },
         spec: {
           priorityClassName: "system-node-critical",
+          terminationGracePeriodSeconds: 10,
           affinity: {
             nodeAffinity: {
               requiredDuringSchedulingIgnoredDuringExecution: {
@@ -135,12 +136,18 @@ local awsnode = {
               }],
               name: "aws-node",
               readinessProbe: {
+                exec: {
+                  command: ["/app/grpc-health-probe", "-addr=:50051"],
+                },
+                initialDelaySeconds: 10,
+              },
+              livenessProbe: {
                 exec: {
                   command: ["/app/grpc-health-probe", "-addr=:50051"],
                 },
                 initialDelaySeconds: 35,
+                periodSeconds: 60,
               },
-              livenessProbe: self.readinessProbe,
               env_:: {
                 AWS_VPC_ENI_MTU: "9001",
                 AWS_VPC_K8S_CNI_CONFIGURE_RPFILTER: "false",