Merge pull request #3811 from Dokploy/3332-zombie-processes-from-failed-log-cleanup

refactor: improve error handling and logging in cron job setup for ba…

Author: Siumauricio

packages/server/src/utils/access-log/handler.ts

@@ -1,3 +1,5 @@
+import fs from "node:fs";
+import path from "node:path";
 import { paths } from "@dokploy/server/constants";
 import {
 	getWebServerSettings,
@@ -12,19 +14,24 @@ export const startLogCleanup = async (
 	cronExpression = "0 0 * * *",
 ): Promise<boolean> => {
 	try {
-		const { DYNAMIC_TRAEFIK_PATH } = paths();
-
 		const existingJob = scheduledJobs[LOG_CLEANUP_JOB_NAME];
 		if (existingJob) {
 			existingJob.cancel();
 		}
 
 		scheduleJob(LOG_CLEANUP_JOB_NAME, cronExpression, async () => {
 			try {
+				const { DYNAMIC_TRAEFIK_PATH } = paths();
+				const accessLogPath = path.join(DYNAMIC_TRAEFIK_PATH, "access.log");
+
+				if (!fs.existsSync(accessLogPath)) {
+					console.error("Access log file does not exist");
+					return;
+				}
+
 				await execAsync(
-					`tail -n 1000 ${DYNAMIC_TRAEFIK_PATH}/access.log > ${DYNAMIC_TRAEFIK_PATH}/access.log.tmp && mv ${DYNAMIC_TRAEFIK_PATH}/access.log.tmp ${DYNAMIC_TRAEFIK_PATH}/access.log`,
+					`tail -n 1000 ${accessLogPath} > ${accessLogPath}.tmp && mv ${accessLogPath}.tmp ${accessLogPath}`,
 				);
-
 				await execAsync("docker exec dokploy-traefik kill -USR1 1");
 			} catch (error) {
 				console.error("Error during log cleanup:", error);

packages/server/src/utils/backups/index.ts

@@ -30,34 +30,42 @@ export const initCronJobs = async () => {
 	const webServerSettings = await getWebServerSettings();
 
 	if (webServerSettings?.enableDockerCleanup) {
-		scheduleJob("docker-cleanup", CLEANUP_CRON_JOB, async () => {
-			console.log(
-				`Docker Cleanup ${new Date().toLocaleString()}]  Running docker cleanup`,
-			);
+		try {
+			scheduleJob("docker-cleanup", CLEANUP_CRON_JOB, async () => {
+				console.log(
+					`Docker Cleanup ${new Date().toLocaleString()}]  Running docker cleanup`,
+				);
 
-			await cleanupAll();
+				await cleanupAll();
 
-			await sendDockerCleanupNotifications(admin.user.id);
-		});
+				await sendDockerCleanupNotifications(admin.user.id);
+			});
+		} catch (error) {
+			console.error("[Backup] Docker Cleanup Error", error);
+		}
 	}
 
 	const servers = await getAllServers();
 
 	for (const server of servers) {
 		const { serverId, enableDockerCleanup, name } = server;
 		if (enableDockerCleanup) {
-			scheduleJob(serverId, CLEANUP_CRON_JOB, async () => {
-				console.log(
-					`SERVER-BACKUP[${new Date().toLocaleString()}] Running Cleanup ${name}`,
-				);
-
-				await cleanupAll(serverId);
-
-				await sendDockerCleanupNotifications(
-					admin.user.id,
-					`Docker cleanup for Server ${name} (${serverId})`,
-				);
-			});
+			try {
+				scheduleJob(serverId, CLEANUP_CRON_JOB, async () => {
+					console.log(
+						`SERVER-BACKUP[${new Date().toLocaleString()}] Running Cleanup ${name}`,
+					);
+
+					await cleanupAll(serverId);
+
+					await sendDockerCleanupNotifications(
+						admin.user.id,
+						`Docker cleanup for Server ${name} (${serverId})`,
+					);
+				});
+			} catch (error) {
+				console.error(`[Backup] ${error}`);
+			}
 		}
 	}
 
@@ -87,11 +95,15 @@ export const initCronJobs = async () => {
 	}
 
 	if (webServerSettings?.logCleanupCron) {
-		console.log(
-			"Starting log requests cleanup",
-			webServerSettings.logCleanupCron,
-		);
-		await startLogCleanup(webServerSettings.logCleanupCron);
+		try {
+			console.log(
+				"Starting log requests cleanup",
+				webServerSettings.logCleanupCron,
+			);
+			await startLogCleanup(webServerSettings.logCleanupCron);
+		} catch (error) {
+			console.error("[Backup] Log Cleanup Error", error);
+		}
 	}
 };