From 13d1f2ae242cfc445d0c96e9ac4c3b8c8bbe27fd Mon Sep 17 00:00:00 2001
From: Sylvain Desbureaux <sylvain.desbureaux@orange.com>
Date: Mon, 6 Dec 2021 11:33:11 +0100
Subject: [DMAAP][MR] Use startup probes

Instead of using long intial delay for readiness and liveness probes,
use startup Probes with aggressive checks in order to detect start as
fast as possible AND handles slow start cases.

Issue-ID: OOM-1
Signed-off-by: Sylvain Desbureaux <sylvain.desbureaux@orange.com>
Change-Id: Icb19ef62a4a8c52ac8800aa1dd4b71e350a6eeca
(cherry picked from commit 8fade99d8fefc0c8d09a6e3df74fe68c70a07458)
---
 .../components/message-router/templates/statefulset.yaml | 14 +++++++++++++-
 kubernetes/dmaap/components/message-router/values.yaml   | 16 ++++++++++++++--
 kubernetes/onap/resources/overrides/environment.yaml     |  9 ++-------
 3 files changed, 29 insertions(+), 10 deletions(-)

diff --git a/kubernetes/dmaap/components/message-router/templates/statefulset.yaml b/kubernetes/dmaap/components/message-router/templates/statefulset.yaml
index 9bad341792..f0832add80 100644
--- a/kubernetes/dmaap/components/message-router/templates/statefulset.yaml
+++ b/kubernetes/dmaap/components/message-router/templates/statefulset.yaml
@@ -103,13 +103,25 @@ spec:
             initialDelaySeconds: {{ .Values.liveness.initialDelaySeconds }}
             periodSeconds: {{ .Values.liveness.periodSeconds }}
             timeoutSeconds: {{ .Values.liveness.timeoutSeconds }}
-          {{ end -}}
+            successThreshold: {{ .Values.liveness.successThreshold }}
+            failureThreshold: {{ .Values.liveness.failureThreshold }}
+          {{ end }}
           readinessProbe:
             tcpSocket:
               port: {{ .Values.readiness.port }}
             initialDelaySeconds: {{ .Values.readiness.initialDelaySeconds }}
             periodSeconds: {{ .Values.readiness.periodSeconds }}
             timeoutSeconds: {{ .Values.readiness.timeoutSeconds }}
+            successThreshold: {{ .Values.readiness.successThreshold }}
+            failureThreshold: {{ .Values.readiness.failureThreshold }}
+          startupProbe:
+            tcpSocket:
+                port: {{ .Values.startup.port }}
+            initialDelaySeconds: {{ .Values.startup.initialDelaySeconds }}
+            periodSeconds: {{ .Values.startup.periodSeconds }}
+            timeoutSeconds: {{ .Values.startup.timeoutSeconds }}
+            successThreshold: {{ .Values.startup.successThreshold }}
+            failureThreshold: {{ .Values.startup.failureThreshold }}
           env:
           - name: enableCadi
             value: "{{ .Values.global.aafEnabled }}"
diff --git a/kubernetes/dmaap/components/message-router/values.yaml b/kubernetes/dmaap/components/message-router/values.yaml
index 44697d7c92..7aa31769f4 100644
--- a/kubernetes/dmaap/components/message-router/values.yaml
+++ b/kubernetes/dmaap/components/message-router/values.yaml
@@ -85,18 +85,30 @@ affinity: {}
 
 # probe configuration parameters
 liveness:
-  initialDelaySeconds: 70
+  initialDelaySeconds: 10
   periodSeconds: 10
   timeoutSeconds: 1
+  successThreshold: 1
+  failureThreshold: 3
   # necessary to disable liveness probe when setting breakpoints
   # in debugger so K8s doesn't restart unresponsive container
   port: api
   enabled: true
 
 readiness:
-  initialDelaySeconds: 70
+  initialDelaySeconds: 10
   periodSeconds: 10
   timeoutSeconds: 1
+  successThreshold: 1
+  failureThreshold: 3
+  port: api
+
+startup:
+  initialDelaySeconds: 10
+  periodSeconds: 10
+  timeoutSeconds: 1
+  successThreshold: 1
+  failureThreshold: 70
   port: api
 
 service:
diff --git a/kubernetes/onap/resources/overrides/environment.yaml b/kubernetes/onap/resources/overrides/environment.yaml
index 278a2da156..50703fbf4a 100644
--- a/kubernetes/onap/resources/overrides/environment.yaml
+++ b/kubernetes/onap/resources/overrides/environment.yaml
@@ -82,10 +82,10 @@ appc:
       periodSeconds: 60
 cassandra:
   liveness:
-    initialDelaySeconds: 120
+    timeoutSeconds: 30
     periodSeconds: 120
   readiness:
-    initialDelaySeconds: 120
+    timeoutSeconds: 30
     periodSeconds: 60
 clamp:
   liveness:
@@ -120,11 +120,6 @@ dmaap:
       initialDelaySeconds: 120
     readiness:
       initialDelaySeconds: 120
-  message-router:
-    liveness:
-      initialDelaySeconds: 120
-    readiness:
-      initialDelaySeconds: 120
   dmaap-dr-prov:
     liveness:
       initialDelaySeconds: 120
-- 
cgit 1.2.3-korg