feat: monitor add max retries to avoid network fluctuation

src/client/components/monitor/MonitorInfoEditor.tsx

@@ -19,6 +19,7 @@ const defaultValues: Omit<MonitorInfoEditorValues, 'payload'> = {
   type: monitorProviders[0].name,
   active: true,
   interval: 60,
+  maxRetries: 0,
 };
 
 interface MonitorInfoEditorProps {
@@ -90,6 +91,14 @@ export const MonitorInfoEditor: React.FC<MonitorInfoEditorProps> = React.memo(
             />
           </Form.Item>
 
+          <Form.Item
+            label="Max Retries"
+            name="maxRetries"
+            tooltip="Maximum retries before the service is marked as down and a notification is sent"
+          >
+            <InputNumber min={0} max={10} defaultValue={0} />
+          </Form.Item>
+
           {formEl}
 
           <Form.Item label="Notification" name="notificationIds">

src/server/model/_schema/index.ts

@@ -1,4 +1,5 @@
 import { z } from 'zod';
+import { MonitorModelSchema } from '../../prisma/zod';
 
 // Match prisma `JsonValue`
 export const jsonFieldSchema = z.union([
@@ -53,18 +54,7 @@ export const websiteInfoSchema = z.object({
   deletedAt: z.date().nullable(),
 });
 
-export const monitorInfoSchema = z.object({
-  id: z.string(),
-  workspaceId: z.string(),
-  name: z.string(),
-  type: z.string(),
-  active: z.boolean(),
-  interval: z.number(),
-  payload: jsonFieldSchema,
-  createdAt: z.date(),
-});
-
-export const monitorInfoWithNotificationIdSchema = monitorInfoSchema.and(
+export const monitorInfoWithNotificationIdSchema = MonitorModelSchema.and(
   z.object({
     notifications: z.array(z.object({ id: z.string() })),
   })

src/server/model/monitor/manager.ts

@@ -5,7 +5,7 @@ import { logger } from '../../utils/logger';
 
 export type MonitorUpsertData = Pick<
   Monitor,
-  'workspaceId' | 'name' | 'type' | 'interval'
+  'workspaceId' | 'name' | 'type' | 'interval' | 'maxRetries'
 > & {
   id?: string;
   active?: boolean;

src/server/model/monitor/runner.ts

@@ -15,6 +15,7 @@ import { createAuditLog } from '../auditLog';
 export class MonitorRunner {
   isStopped = false;
   timer: NodeJS.Timeout | null = null;
+  retriedNum = 0;
 
   constructor(public monitor: Monitor & { notifications: Notification[] }) {}
 
@@ -23,7 +24,7 @@ export class MonitorRunner {
    */
   async startMonitor() {
     const monitor = this.monitor;
-    const { type, interval, workspaceId } = monitor;
+    const { type, interval, workspaceId, maxRetries } = monitor;
 
     const provider = monitorProviders[type];
     if (!provider) {
@@ -58,30 +59,41 @@ export class MonitorRunner {
           value = -1;
         }
 
-        // check event update
-        if (value < 0 && currentStatus === 'UP') {
-          await this.createEvent(
-            'DOWN',
-            `Monitor [${monitor.name}] has been down`
-          );
-          await this.notify(`[${monitor.name}] 🔴 Down`, [
-            token.text(
-              `[${monitor.name}] 🔴 Down\nTime: ${dayjs().format(
-                'YYYY-MM-DD HH:mm:ss (z)'
-              )}`
-            ),
-          ]);
-          currentStatus = 'DOWN';
-        } else if (value > 0 && currentStatus === 'DOWN') {
-          await this.createEvent('UP', `Monitor [${monitor.name}] has been up`);
-          await this.notify(`[${monitor.name}] ✅ Up`, [
-            token.text(
-              `[${monitor.name}] ✅ Up\nTime: ${dayjs().format(
-                'YYYY-MM-DD HH:mm:ss (z)'
-              )}`
-            ),
-          ]);
-          currentStatus = 'UP';
+        if (this.retriedNum < maxRetries) {
+          // can be retry
+          this.retriedNum++;
+        } else {
+          // check event update
+          if (value < 0 && currentStatus === 'UP') {
+            // UP -> DOWN
+            await this.createEvent(
+              'DOWN',
+              `Monitor [${monitor.name}] has been down`
+            );
+            await this.notify(`[${monitor.name}] 🔴 Down`, [
+              token.text(
+                `[${monitor.name}] 🔴 Down\nTime: ${dayjs().format(
+                  'YYYY-MM-DD HH:mm:ss (z)'
+                )}`
+              ),
+            ]);
+            currentStatus = 'DOWN';
+          } else if (value > 0 && currentStatus === 'DOWN') {
+            // DOWN -> UP
+            this.retriedNum = 0;
+            await this.createEvent(
+              'UP',
+              `Monitor [${monitor.name}] has been up`
+            );
+            await this.notify(`[${monitor.name}] ✅ Up`, [
+              token.text(
+                `[${monitor.name}] ✅ Up\nTime: ${dayjs().format(
+                  'YYYY-MM-DD HH:mm:ss (z)'
+                )}`
+              ),
+            ]);
+            currentStatus = 'UP';
+          }
         }
 
         // insert into data
@@ -93,9 +105,6 @@ export class MonitorRunner {
         });
 
         subscribeEventBus.emit('onMonitorReceiveNewData', workspaceId, data);
-
-        // Run next loop
-        nextAction();
       } catch (err) {
         logger.error('[Monitor] Run monitor error,', monitor.id, String(err));
         createAuditLog({
@@ -104,6 +113,9 @@ export class MonitorRunner {
           relatedType: 'Monitor',
           content: `Run monitor(id: ${monitor.id}) error: ${String(err)}`,
         });
+      } finally {
+        // Run next loop
+        nextAction();
       }
     };
 

src/server/prisma/migrations/20240209124945_add_monitor_maxretries/migration.sql

@@ -0,0 +1,2 @@
+-- AlterTable
+ALTER TABLE "Monitor" ADD COLUMN     "maxRetries" INTEGER NOT NULL DEFAULT 0;

src/server/prisma/schema.prisma

@@ -265,6 +265,7 @@ model Monitor {
   type        String   @db.VarChar(100)
   active      Boolean  @default(true) @db.Boolean
   interval    Int      @default(20) @db.Integer
+  maxRetries  Int      @default(0)
   // TODO
   // maxRetry      Int       @default(0) @db.Integer
   // retryInterval Int       @default(0) @db.Integer

src/server/prisma/zod/monitor.ts

@@ -15,6 +15,7 @@ export const MonitorModelSchema = z.object({
   type: z.string(),
   active: z.boolean(),
   interval: z.number().int(),
+  maxRetries: z.number().int(),
   /**
    * [CommonPayload]
    */

src/server/trpc/routers/monitor.ts

@@ -16,13 +16,15 @@ import {
 import dayjs from 'dayjs';
 import {
   monitorEventSchema,
-  monitorInfoSchema,
   monitorInfoWithNotificationIdSchema,
   monitorStatusSchema,
 } from '../../model/_schema';
 import { OPENAPI_TAG } from '../../utils/const';
 import { OpenApiMeta } from 'trpc-openapi';
-import { MonitorStatusPageModelSchema } from '../../prisma/zod';
+import {
+  MonitorModelSchema,
+  MonitorStatusPageModelSchema,
+} from '../../prisma/zod';
 import { runCodeInVM } from '../../model/monitor/provider/custom';
 import { createAuditLog } from '../../model/auditLog';
 import {
@@ -124,11 +126,12 @@ export const monitorRouter = router({
         type: z.string(),
         active: z.boolean().default(true),
         interval: z.number().int().min(5).max(10000).default(20),
+        maxRetries: z.number().int().min(0).max(10).default(0),
         notificationIds: z.array(z.string()).default([]),
         payload: z.object({}).passthrough(),
       })
     )
-    .output(monitorInfoSchema)
+    .output(MonitorModelSchema)
     .mutation(async ({ input }) => {
       const {
         id,
@@ -137,6 +140,7 @@ export const monitorRouter = router({
         type,
         active,
         interval,
+        maxRetries,
         notificationIds,
         payload,
       } = input;
@@ -148,6 +152,7 @@ export const monitorRouter = router({
         type,
         active,
         interval,
+        maxRetries,
         notificationIds,
         payload,
       });
@@ -166,7 +171,7 @@ export const monitorRouter = router({
         monitorId: z.string().cuid2(),
       })
     )
-    .output(monitorInfoSchema)
+    .output(MonitorModelSchema)
     .mutation(async ({ input }) => {
       const { workspaceId, monitorId } = input;
 
@@ -239,7 +244,7 @@ export const monitorRouter = router({
         active: z.boolean(),
       })
     )
-    .output(monitorInfoSchema)
+    .output(MonitorModelSchema)
     .mutation(async ({ input, ctx }) => {
       const { workspaceId, monitorId, active } = input;
       const user = ctx.user;