Application crashes when droplet CPU usage reaches 40%

deployment
1

I am using meteor version 2.7.1 and using mup to deploy to digital ocean. I bought a droplet with 8 core cpu and 16GB ram to deploy the application, but when deploying it to users, it crashes every time there are many users requesting the application. Every time it crashes, I check the cup droplet and container in docker, the cup droplet is up to 40%, the application is crashing, while the cup container in docker is 100%. Why is the cup droplet not up to 80% or 90%, why is the application crashing like this? What is the problem? Please help me solve this problem, thanks in advance :pray: .

module.exports = {
  servers: {
    one: {
      host: 'xxx.xxx.xxx.xxx',
      username: 'root',
      password: 'server-password'
    },
  },

  app: {
    // TODO: change app name and path
    name: 'app',
    path: '../',
    volumes: {
      '/data/file_uploads': '/data/file_uploads',
    },

    servers: {
      one: {},
    },

    env: {
      // TODO: Change to your app's url
      ROOT_URL: 'http://xxx.xxx.xxx.xxx',
      MONGO_URL: 'mongodb://mongodb/dbName',
      MONGO_OPLOG_URL: 'mongodb://mongodb/local',
    },

    docker: {
      image: 'zodern/meteor:root',
      prepareBundle: true,
      useBuildKit: true,
    },
    deployCheckWaitTime: 120,
    enableUploadProgressBar: true,
  },

  mongo: {
    version: '5.0.5',
    servers: {
      one: {},
    },
  },
}

photo_2025-03-23 09.20.09
photo_2025-03-23 09.20.091280×846 89.9 KB

2

Do you have the server logs by any chance?

3

When it crashed, I checked the container logs, but there were no errors in the container logs.

4

Unless you use Meteor (Node) in multithreading mode, if you have memory allocated per CPU (like in AWS), what you actually run your Meteor on is a 1 core with 2GB memory. The rest is … redundant.

In AWS, I run 2 cores with 1GB. When my app reaches ~500MB memory … booom… the application crashes. I only have the autoscaling set for CPU% and going to have to add memory as a trigger too, probably something around 35-40% (when one CPU memory that Node uses gets to ~80%).

Meteor CPU and memory scale cheaper and better horizontally. You start more 1 - 2 CPU boxes but in theory, anything over 1 CPU is a waste of money. I know … pretty hard to find 1 CPU (1 core) boxes in the clouds.

5

You may consider this: Running meteor with pm2 for auto recovery of your server.
Also might consider this: PM2 - Cluster Mode

6

I checked pm2-meteor but it is out of maintenance, never updated in the past 5 years.