From fa5e57c6e15a87e2a6f14b6613bd99ce85d25967 Mon Sep 17 00:00:00 2001
From: maslow <wangfugen@126.com>
Date: Wed, 2 Feb 2022 00:04:42 +0800
Subject: [PATCH] feat(k8s): add startupProbe & readinessProbe support to
 services;

---
 .vscode/settings.json                         |  1 +
 deploy/kubernetes/fs.yml                      | 13 +++++++++
 deploy/kubernetes/sys.yml                     | 21 ++++++++++++---
 packages/system-server/src/index.ts           | 12 ++++++---
 .../src/lib/service-driver/kubernetes.ts      | 27 +++++++++++++++++--
 5 files changed, 64 insertions(+), 10 deletions(-)

diff --git a/.vscode/settings.json b/.vscode/settings.json
index e235b5c8ca..e107a5aa6d 100644
--- a/.vscode/settings.json
+++ b/.vscode/settings.json
@@ -79,6 +79,7 @@
     "lafyun",
     "millicores",
     "objs",
+    "Referer",
     "signin",
     "uids",
     "upsert",
diff --git a/deploy/kubernetes/fs.yml b/deploy/kubernetes/fs.yml
index 0a0e79cd8a..872d2357c4 100644
--- a/deploy/kubernetes/fs.yml
+++ b/deploy/kubernetes/fs.yml
@@ -44,6 +44,19 @@ spec:
         - image: docker.io/lafyun/storage-service:0.7.3
           imagePullPolicy: IfNotPresent
           name: storage-service
+          startupProbe:
+            httpGet:
+              path: /health-check
+              port: http
+            periodSeconds: 3
+            timeoutSeconds: 3
+            failureThreshold: 20
+          readinessProbe:
+            httpGet:
+              path: /health-check
+              port: http
+            periodSeconds: 10
+            timeoutSeconds: 3
           resources:
             requests:
               memory: "256Mi"
diff --git a/deploy/kubernetes/sys.yml b/deploy/kubernetes/sys.yml
index 8ee17ed97d..4d7e0182c3 100644
--- a/deploy/kubernetes/sys.yml
+++ b/deploy/kubernetes/sys.yml
@@ -51,11 +51,23 @@ spec:
             limits:
               memory: "1024Mi"
               cpu: "1000m"
-          livenessProbe:
+          startupProbe:
             httpGet:
               path: /health-check
-              port: 9000
-            initialDelaySeconds: 20
+              port: http
+              httpHeaders:
+                - name: Referer
+                  value: startupProbe
+            periodSeconds: 3
+            timeoutSeconds: 3
+            failureThreshold: 30
+          readinessProbe:
+            httpGet:
+              path: /health-check
+              port: http
+              httpHeaders:
+                - name: Referer
+                  value: readinessProbe
             periodSeconds: 10
             timeoutSeconds: 3
           command: ["sh", "/app/start.sh"]
@@ -147,4 +159,5 @@ spec:
             - name: LOG_LEVEL
               value: debug
           ports:
-            - containerPort: 9000
+            - name: http
+              containerPort: 9000
diff --git a/packages/system-server/src/index.ts b/packages/system-server/src/index.ts
index 298cf74011..fc5bcf2659 100644
--- a/packages/system-server/src/index.ts
+++ b/packages/system-server/src/index.ts
@@ -1,7 +1,7 @@
 /*
  * @Author: Maslow<wangfugen@126.com>
  * @Date: 2021-07-30 10:30:29
- * @LastEditTime: 2022-01-23 19:46:34
+ * @LastEditTime: 2022-02-01 22:55:54
  * @Description: 
  */
 
@@ -58,9 +58,13 @@ process.on('SIGTERM', gracefullyExit)
 process.on('SIGINT', gracefullyExit)
 
 async function gracefullyExit() {
-  logger.info('exiting: removing system extension service')
-  await ServiceDriver.create().removeService({ appid: Constants.SYSTEM_EXTENSION_APPID } as any)
-  logger.info('exiting: system extension service has been removed')
+  
+  // NOT remove system extension app service if service driver is 'kubernetes', 
+  if (Config.SERVICE_DRIVER === 'docker') {
+    logger.info('exiting: removing system extension service')
+    await ServiceDriver.create().removeService({ appid: Constants.SYSTEM_EXTENSION_APPID } as any)
+    logger.info('exiting: system extension service has been removed')
+  }
 
   logger.info('exiting: closing db connection')
   await DatabaseAgent.sys_accessor.close()
diff --git a/packages/system-server/src/lib/service-driver/kubernetes.ts b/packages/system-server/src/lib/service-driver/kubernetes.ts
index 8e81528b9d..fe3146cd72 100644
--- a/packages/system-server/src/lib/service-driver/kubernetes.ts
+++ b/packages/system-server/src/lib/service-driver/kubernetes.ts
@@ -120,6 +120,7 @@ export class KubernetesServiceDriver implements ServiceDriverInterface {
             labels: labels
           },
           spec: {
+            terminationGracePeriodSeconds: 15,
             automountServiceAccountToken: app.appid === Constants.SYSTEM_EXTENSION_APPID,
             containers: [
               {
@@ -136,7 +137,7 @@ export class KubernetesServiceDriver implements ServiceDriverInterface {
                   { name: 'RUNTIME_IMAGE', value: app.runtime?.image },
                   { name: 'FLAGS', value: `--max_old_space_size=${max_old_space_size}` }
                 ],
-                ports: [{ containerPort: 8000 }],
+                ports: [{ containerPort: 8000, name: 'http' }],
                 resources: {
                   requests: {
                     memory: `${req_memory}Mi`,
@@ -146,7 +147,29 @@ export class KubernetesServiceDriver implements ServiceDriverInterface {
                     memory: `${limit_memory}Mi`,
                     cpu: `${limit_cpu}m`
                   }
-                }
+                },
+                startupProbe: {
+                  httpGet: {
+                    path: '/health-check',
+                    port: 'http',
+                    httpHeaders: [{ name: 'Referer', value: 'startupProbe' }]
+                  },
+                  initialDelaySeconds: 0,
+                  periodSeconds: 3,
+                  timeoutSeconds: 3,
+                  failureThreshold: 40
+                },
+                readinessProbe: {
+                  httpGet: {
+                    path: '/health-check',
+                    port: 'http',
+                    httpHeaders: [{ name: 'Referer', value: 'readinessProbe' }]
+                  },
+                  initialDelaySeconds: 0,
+                  periodSeconds: 60,
+                  timeoutSeconds: 5,
+                  failureThreshold: 3
+                },
               }
             ],
           }