meteorhacks / cluster Goto Github PK

Hello,

I tried this in my repo (https://github.com/darkship/test_cluster_client_error) :

if (Meteor.isServer) {
      Cluster.connect("mongodb://localhost:27017/discovery");
      Cluster.register("test-test");
}

But on the client I have an error (that was not expected and killed other connexions):

 Firefox could not connect to ws://localhost:5001/cluster-ddp/8d6a5062c883e6d079209864a3fbeec534018d1e/test-test/866/10ypgijw/websocket.

And sometimes on the server too :

Cluster: no such endpoint for service:test-test

Replacing test-test by test solved the issue.
I tried with test_test and it didn't work either.So what are the service's name limitations ?

I think this is a great concept, especially because it's so easy to implement. But I'm wondering if it should be recommended for serious production use. If you would make that recommendation, it would be nice to see a comparison between cluster and the other solutions in the docs--

1. The docs mention that using cluster is as easy as adding a Meteor package. So I'm guessing that it puts the request handling and load balancing on Meteor applications and uses the application's database? If not, it's a bit more complicated.
2. There are good existing solutions that auto-scale load balancers, and I know you're aware of them--what are they, and why is cluster better for Meteor apps?
3. If DDP connections are dropped as a load balancer is manually scaled (for the very short period of time it takes to restart the load balancer), as long as there are sticky sessions, is that a problem? What happens?
4. The load balancers in question, like HAProxy, are battle-tested in production and have been under development for years. HAProxy supports SSL, HTTP compression, health checks, redispatching, and has a very useful ACL feature, and supports complex redirects, among other features. It's written in C, and can handle upwards to 40K hits per second. At what point does someone decide to use cluster instead of HAProxy, or vice-versa?

These points aren't meant to criticize your package--as I said, it's really cool. And of course, running a real-time Meteor app is much different than running an HTTP request/response-style app.

Are fallback requests go to each of the instances randomly? Do they stick to one instance?

If the apps are running inside a container and are on the same server they do not communicate as it should.

I'm using Docker-compose, and here is a reproduction:
Frontend App: https://bitbucket.org/jadsonlourenco/todo-frontend
Backend App: https://bitbucket.org/jadsonlourenco/todo-backend
Need install Docker and Docker-compose, then:
docker-compose up
(need run this command in each folder, will start two containers, for each app, access "localhost:3031" to see the frontend app running)

In this case I have two Apps, frontend (that subscribe the collection from backend) and the backend (that just publish the collection). The Cluster package works fine and I can subscribe the collection, I have it working in production, using Docker too. But I don't know why not work any more in Docker, I install a new server and did the same installation, and get this problem.

This is not a problem of network, because I did the same test using different server, and various cloud services with valid domain. As I said this only happens if the applications are on the same server (or in the same cluster, when tested with CoreOS).

And in last case I try subscribe the collection in frontend app using the DDP.connect and works, yes! Because of this I think is some thing on Cluster package or on Docker (container network).

PS: @arunoda I ask about it on twitter, but just open it now because I try everything before make sure the cause of problem; thank you!

So, I have two systems: backend.site.com and client.site.com
But in Firefox the Cross-Origin blocks the content of the backend. I configured the package browser-policy but the error continues.

But I can't see this error in: https://bulletproofmeteor.com/
@arunoda What did you do to resolve this?

PS: I'm sorry to open this issue, but after hours trying and searching was not a solution, so I have to ask for those who already have it resolved. Thank You!

Readme says:

We turn off WebSockets since cloudflare does not support SSL with WebSockets yet!

They support both ws:// and wss:// if you have an enterprise account.

Atmosphere perhaps, or https://meteorhacks.com/cluster-a-different-kind-of-load-balancer-for-meteor.html

We can use a tailing cursor to connect to a another connection without doing a poll.

This is just something that came to my head. Do you thing it could be possible to route traffic to the nearest server? Like when you have servers in the US and Europe, the system should prefer near servers. Something like geo based load-balancing.

Think of a hot code push update. You have 1000 clients to be updated, but only use one server/core!!

Does meteorhacks:cluster support cordova meteor builds ?

Hi,

I have a main microservice who request some others microservices, I would like to display an error on my UI when a microservice is unavailable.

For the moment I have the message -> "Cluster: no such endpoint for service:name" in the server console.

How can I do ?

Thanks :)

I'm wanting to build a status/admin UI page to display information on the cluster (maybe even as a standalone service in the cluster). I'm firstly just wanting to list the endpoints grouped by service. I was hoping to get access to the documents via a find on Cluster.discovery._endpointsColl, but I see the client has a different API limited to discoverConnection. Have would you achieve this?

Hi, I am using Cluster on AWS Opsworks... thanks to the great help of folks Sunsama.com as per https://blog.sunsama.com/meteor-docker-opsworks/

One thing I am not sure on is if I am using 3rd party Oauth (via accounts-facebook etc) the require that I put in the requesting domain and server IP (whitelist).

I have setup cluster to handle requests at mydomain.com ... I have not used the load balancer option. I have two servers

I have setup Facebook app to allow login from mydomain.com. If Cluster redirects me to the actual machine of my domain I can login. But when Cluster sends me to one of the other machines in the Cluster I get the following error:

Exception while invoking method 'login' Error: Failed to complete OAuth handshake with Facebook. failed [400] {"error":{"message":"This IP can't make requests for that application.","type":"OAuthException","code":5}} 

So this makes sense as the second machine in the cluster would have an IP that Facebook does not know.

As the system may autoscale, the IPs may not be known to Facebook.

I there some way to have Cluster have all requests seem like they're from one IP?

Sometimes my app throws that error:

Cluster: Exiting worker 2 with exitCode=7 signalCode=null
Error: write after end
    at writeAfterEnd (_stream_writable.js:133:12)
    at Socket.Writable.write (_stream_writable.js:181:5)
    at Socket.write (net.js:616:40)
    at Socket.Writable.end (_stream_writable.js:341:10)
    at Socket.end (net.js:397:31)
    at App.exports.GenericApp.GenericApp.handle_error (/opt/easydrop/app/programs/server/npm/ddp/node_modules/sockjs/lib/webjs.js:148:13)
    at execute_request (/opt/easydrop/app/programs/server/npm/ddp/node_modules/sockjs/lib/webjs.js:33:30)
    at Object.req.next_filter (/opt/easydrop/app/programs/server/npm/ddp/node_modules/sockjs/lib/webjs.js:105:18)
    at Listener.webjs_handler (/opt/easydrop/app/programs/server/npm/ddp/node_modules/sockjs/lib/webjs.js:107:13)
    at Listener.handler (/opt/easydrop/app/programs/server/npm/ddp/node_modules/sockjs/lib/sockjs.js:147:12)

Btw How if I am removing my packages data from my main app collections? With this tutorial the data will be not synced between "packages collection" in main app and search app. I think it will be an issue. How to get it work (I mean the data between all collections are remain same/synced)? Thanks

Hi, I'm using meteor-cluster in the following scenario:

• AWS
• 2 instances running meteor-cluster as normal app-servers (e.g. not balancers)
• Deployment using mup
• ELB
• SSL termination on ELB
• DISABLE_WEBSOCKETS (since ELB+SSL)
• The site itself (if you want to test) is https://gormim.com
  Most of the time things work well, but since I have an alert on ELB looking for 4xx I noticed that every once in awhile the calls to POST /cluster-ddp are 404ing

Here's just one screenshot of the 404s on my ELB.
http://cl.ly/image/17083P0j2x07
And I recorded the access logs, so here's an example request that 404s

2015-02-28T17:41:57.787697Z web 66.249.78.7:60042 172.31.38.90:80 0.00012 0.003284 0.000023 404 404 417 26 "POST https://gormim.com:443/cluster-ddp/f4a987ea4fac27bdc2f00df9dd01a691e23f29b8/web/352/etyjoyb4/xhr_send HTTP/1.1"

Most of the /cluster-ddp calls end with 200 (or 204), but occasionally they 404.
Here are just a few examples or 2xx lines:

2015-02-28T17:41:18.573685Z web 109.64.50.27:53781 172.31.12.222:80 0.000073 30.001697 0.000045 200 200 0 24 "POST https://gormim.com:443/cluster-ddp/f4a987ea4fac27bdc2f00df9dd01a691e23f29b8/web/607/lqhhw6rh/xhr HTTP/1.1"
2015-02-28T17:41:49.567741Z web 109.64.50.27:53761 172.31.12.222:80 0.000116 0.004414 0.000032 204 204 22 0 "POST https://gormim.com:443/cluster-ddp/f4a987ea4fac27bdc2f00df9dd01a691e23f29b8/web/607/lqhhw6rh/xhr_send HTTP/1.1"

As far as I can tell, this does not interfere in the sense that users are bothered by it, but it does make monitoring really hard (b/c it's hard to say what's a real error and that's just 404 from /cluster-ddp).

I wonder if you had seen this before or could guide me towards a solution?

Thanks!

HI! I had a problem. When a document is added to the collection, my observers are triggered on all servers (2 clusters with 4 cores = 8 workers). How can i prevent this behavior?

For example:

Items.find().observeChanges({
  added: function(id, fields) {
    HTTP.get(fields.url, function() {
      // DO SOMETHING
    })
  }
})

Its fired on ALL workers, but i want it to only run on one worker (most are not loaded).

Is this something cluster and/or mup is doing, or is it a problem with some other package?

Accounts.onLogin (info) ->
  console.log('onLogin', info.connection.clientAddress)
  console.log(info.connection.httpHeaders)

[parlay.io] onLogin 127.0.0.1[parlay.io] 
[parlay.io] { host: 'parlay.io',[parlay.io] 
[parlay.io]   'user-agent': 'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_10_2) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/42.0.2311.135 Safari/537.36',[parlay.io] 
[parlay.io]   'accept-language': 'en-US,en;q=0.8' }[parlay.io]

Now we are doing an info hit to explore balancers and endpoints. If we could use DNS, we can support external apps without much issue.

May be we don't need to go with this approach.

I managed to deploy my app after adding cluster and I set "CLUSTER_WORKERS_COUNT":2 in my mup.json.

My app is working fine but when I check the logs, cluster seems to be trying to initialize new workers all the time because they are exiting immediately for some reason:

...
Cluster: Initializing worker 55 on port 4810
Cluster: Exiting worker 55 with exitCode=0 signalCode=null
Cluster: Initializing worker 56 on port 4828
Cluster: Exiting worker 56 with exitCode=0 signalCode=null
Cluster: Initializing worker 57 on port 15815
Cluster: Exiting worker 57 with exitCode=0 signalCode=null
...

Hello,

I like the idea of mini-services using cluster. It makes my application more modular and as an added benefit I can scale out DDP traffic.
Now, I actually made a separate build process using gulp to separate my front-end from my back-end (as we have some specific requirements on how to serve the front-end).
Therefore I do not need webapp (since I do not want autoupdate etc either).
Clusters package.js specifies it needs webapp. Therefore my first question is how hard this dependency is between cluster and webapp.

My second question is how session data is handled. If I make mini services using cluster and use the accounts package is session and authentication data shared between these services?

Also,

Hi, I posted this question on SO:

Basically is meteor:cluster able yet to route different dns to the right meteor app ? This would involve the main meteor service to know the URL origin and then to redirect to the right meteor app instance. A bit like what we are doing with a reverse proxy.

Hi @arunoda,
first of all, thanks for all the great work you are doing for the meteor community ;)

We are trying to build an app and would like to use the cluster package for service-discovery.

As soon as we run the app in the ios simulator or device we get following error:

Uncaught Error: SyntaxError: JSON Parse error: Unrecognized token '<':2087:http://meteor.local/packages/ddp.js?41b62dcceb3ce0de6ca79c6aed088cccde6a44d8

Basically the ddp call returns the raw html.

I have created a small demo project here https://github.com/banglashi/meteor-cordova-cluster to reproduce the issue. Just edit the CLUSTER_DISCOVERY_URL in the dev and mobile shell scripts.
Than lunch the ./dev script for the search service and the ./mobile script for the web service.

It could be possible that I am doing/setting something wrong. Could it be related to the endpoint being localhost?

Any hint or suggestions is really appreciated.

Best regards

Hi @arunoda,

if you have a lot of serivces and one changes from 1.0.0 to 2.0.0 including breaking "API" changes. How can I control that?

By naming conventions of the service?

Cluster.register('ServiceA.v1')
Cluster.register('ServiceA.v2')
...

Or would it be an idea to register like

// register service
Cluster.register('ServiceA', {version: '1.0.1'})

// connect to service
serviceConnection = Cluster.discoverConnection("mlServiceTest", {version: '1.0.1'});

I'm running to an issue that I have an Mongo update query in startup code of server:

if (Meteor.isServer) {
 Meteor.startup(function(){
    Posts.find().observeChanges({
        added: function(id, fields) {
            Authors.update({_id: fields.authorId}, {$inc: {count: 1}});
        }
    })
 });
}

I run app with 4 cores server and this added event run 4 times which I expected just 1.
Is there a way in app that we can detect current running core? Or any other best practices to run observerChanges in multicore environment?

It would be nice if we could run code on certain events. For example, if the average load is more than 90%, launch a new server (I think DigitalOcean & friends allow this using their APIs). Or at least notify the admin. Or remove a server, if the load goes below a certain value. And having a number of current servers would be nice, so we could scale but check before removing servers so that we keep at least three servers or so.

I'm not sure what events and data would be helpful, but basically a cluster API would be nice :)

Hi,

Does Cluster clean itself up or do I need to delete old entries that in the discovery database?

I am worried that there are some "old" IPs in there that are still working as part of a separate test environment and cluster is redirecting to it... I just wanted to re-initialize the DB and make sure. Do I need to do this or will Cluster clean itself up?

thanks

Hi, Is it possible
to separate the microservices by folder naming convention?
Or by a service.json file, inside a folder, which would describe the (to compile) code scope and limit the execution only to a special service environment.

It could be something like the already implemented balancer solution:
"Every node can be a balancer"
This would be:
"Every Node can be any of the possible services inside one project"

This could also simplifie the code base, because we could share code between services without copy & paste of files between different meteor projects, because we would have only one project. And it would be very easy to start using services for existing projects.

So my question is, is it possible and are there some limitations to solve it?
This would also make possible of a MotherShip solution, because we would know about all the microservices in a project ;)

And a lot of thanks, to Arunoda and all the MeteorHacks Guys which are helping a lot to improve meteor.
Really great work all that stuff!

Hey! I'm loving all the things you guys are building for meteor!
I was using cluster up until a few days ago when I realized it breaks my current job que!

I have a custom job handler running on my server that looks like this:

Jobs = new Mongo.Collection('jobs')

var throttle = [];
var running = false;

Meteor.setInterval(function(){
    if (!(running)) {
        running = true;
        var now = new Date();
        var time = now.getTime();
        var unfinishedJobs = Jobs.find({_id: { $nin: throttle }, date: {$lt: time} });
        unfinishedJobs.map(function (job) {
            throttle.push(job._id);
            if (job.name === 'setTime') {
                setTime(job.campaign);
            }else if (job.name === 'matchLimit') {
                matchLimitReached(job.match);
            }

            /* kill throttle */
            var index = throttle.indexOf(job._id);
            if (index > -1) {
                throttle.splice(index, 1);
            }
            Jobs.remove({_id: job._id});
        });
        running = false;
    }
}, 500);

The issue is that this code gets ran on both cores... so a second one can run before the first one is completed and cause problems.

Is there a way for the separate instances to share a variable?
I want them to be able to share var running = false; so that way this isn't an issue!

Is this possible?
Can I just set the environment variables to do this or?

@arunoda this sounds really interesting. I noticed that you are not running with Websockets. Anyway, I was wondering if you have any high-level guidance on what sort of performance impact could we expect on falling back to HTTP (vs TCP/Websockets), please?

Thank you for this great meteor package! I also noticed your great (a bit older) node library node-usage.

So, is there any way of integrating these two? In the way that the system-resources are being taken into account for the balancing logic? I think that would be the ultimate goal, pointing the visitor to the most idle server, and ultimately alerting some system administrators when all servers become to heavily loaded (so he can spawn a new server).

WebSocket connection to 'wss://domain/sockjs/163/h0c16r5q/websocket' failed: WebSocket is closed before the connection is established.

I suspect cluster could be the blame, but have no proof just yet.

This happens 3-4 times per hour in meteor run mode

Can anyone shed some light on what I need to do to correct this?

I'm attempting to deploy in own server environment (using mup), I assume I have misconfigured something.

I'm experimenting with cluster and what I want to achieve is to be able to round robin requests between multiple instances of a particular micro service.

E.g. I have a "log" service which provides logging functionality to the cluster:

Cluster.register('log');

and a web service which connects to that log service:

var logConn = Cluster.discoverConnection('log');

As a test I start 2 instances of the log service (let's call them log1 and log2), both on a different port:

~/apps/demo-log1$ MONGO_URL=mongodb://localhost:27017/demo meteor --port 2001

~/apps/demo-log2$ MONGO_URL=mongodb://localhost:27017/demo meteor --port 2002

And 1 instance of the web service:

~/apps/demo-web$ MONGO_URL=mongodb://localhost:27017/demo meteor --port 2000

The web service can send log messages to the log service, but it's always the same instance (let's say log1) that's receiving the messages.
When I stop this instance (log1), the messages are being sent to the other one (log2).
So the failover is working, but in addition to this I also want some kind of load balancing between the two log instances so that they can both accept messages from the web service in a round robin fashion.

What am I doing wrong?

If I go through the "deploying a highly available meteor cluster" tutorial and enter the url in with https the page loads but the cluster does not work. If I open the console i'm getting this in Chrome:

Mixed Content: The page at 'https://mydomain.net/' was loaded over HTTPS, but requested an insecure XMLHttpRequest endpoint 'http://162.xxx.xx.xxx/cluster-ddp/3ad40d3dc87940b5724b52c633d860b3be071630/search/631/nkhc327e/xhr'. This request has been blocked; the content must be served over HTTPS.

If I set the "CLUSTER_BALANCER" to "//162.xxx.xx.xx" (or https) I get an error:

POST https://45.xx.xxx.xxx/cluster-ddp/9807e97fb619e150c77a30f653b521d75033eac1/web/494/lk53znrn/xhr net::ERR_CONNECTION_REFUSED

Is there an easy way to accept https on all of the nodes that are not balancers?

Below problems are occured:

• Logged in user logged out autmatically.
• Method not found 404 on
• Uncaught SyntaxError: Unexpected token <

If I remove below cluster code section, everything is ok. I think velocity client integration test and cluster can't work together.

Cluster.connect('mongodb://localhost/discovery-conn');
Cluster.register('main');

var conn = Cluster.discoverConnection('service');

Meteor.methods({
  getMessage: function (name) {
    check(name, String);
    return conn.call('greet', name);
  }
});

Reproducing repo: https://github.com/boldkhuu/cluster-with-velocity

Create a way to load balance services. Come up with a pattern to completely ignore Nginx

After an error happened. It may have send the headers. If so, we need to stop sending them again.
See: https://ui.kadira.io/et/3e584969-9068-4552-9ec0-e1395615f0f7

Also, we may need think about re-try logic again. May be we don't need to retry if we send some data to the client.

Has anyone tried that before?

I built my meteor-ionic app for IOS and Android, which works just fine when I run it in the browser ( Chrome ) on mac and when I run it in Chrome from my android phone.

I am running the ionic version on port 5000, and web version on port 3000. All the publishers are part of the web version and so is the database. Ionic app at 3000 connects to the server and database using Meteor Cluster solution.

When I launch the app as an Android app, it does not connect to the server and does not show any data.

Command to run meteor-ionic mobile app -
sudo MONGO_URL=mongodb://localhost:3001/meteor CLUSTER_ENDPOINT_URL=http://localhost:3000 meteor run android-device --verbose --port 5000 --settings settings.json

discovering connection from mobile app -
SearchConn = Cluster.discoverConnection('qollserver');

Declaring the mongo-collection in the mobile app -
QbSummary = new Mongo.Collection('qbank_summary', SearchConn);

Subscribing to the server side data from mobile app -
SearchConn.subscribe('QBANK_SUMMARY_PUBLISHER', {userId: userId});

Query for the data in the mobile app -
QbSummary.find();

On the server side, I have the following in app.js
Cluster.connect("mongodb://localhost:3001/meteor");
Cluster.register("qollserver");

Run the web version server running the following command -
sudo meteor --port 3000

how can I know which is the master process to perform actions that should only be performed once ?
thx

I use multicore cluster mode. When reload page ( i use ironrouter) the web enter in a loop reload infinite.
When disable cluster workers, the web work fine

We now send heartbeats for every 15 secs. We need to make it bit random.

Is there a way to be able to launch multiple instances of the application on the same server that has multiple cores with meteor-up and utilize cluster to load balance? If so, how should this be setup?

Hi!

This package has binary dependencies, and we wanted to give you a heads up that Meteor 1.1, with support for Windows will be released in a few days.

Now is the time to publish your package for Windows!

It should be very straightforward -- we've added new build machines that run Windows that you can access with the meteor admin get-machine command. This is the same as you've done before for the other architectures we support.

For complete directions on how to publish your package for Windows, visit https://github.com/meteor/meteor/wiki/Publishing-a-package-with-binary-dependencies-for-the-Windows-preview

On behalf of the Meteor team,
@stubailo, @Slava and @avital.

P.S. This is an automated message based on our database of package authors.

Add a cluster wide authentication system based on DDP.

When I turn on cluster package connection.clientAddress becomes 127.0.0.1. How to fix that?

Randomly all of the workers will kill them selves with an error similar to these:

[104.236.27.101] Error: write after end
    at writeAfterEnd (_stream_writable.js:133:12)[104.236.27.101] 
    at Socket.Writable.write (_stream_writable.js:181:5)
    at Socket.write (net.js:616:40)
    at Socket.Writable.end (_stream_writable.js:341:10)
    at Socket.end (net.js:397:31)
    at App.exports.GenericApp.GenericApp.handle_error (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/webjs.js:148:13)
    at execute_request (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/webjs.js:33:30)
    at Object.req.next_filter (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/webjs.js:105:18)
    at Listener.webjs_handler (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/webjs.js:107:13)
    at Listener.handler (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/sockjs.js:147:12)
Cluster: Exiting worker 7 with exitCode=7 signalCode=null

[104.236.27.101] TypeError: Cannot read property 'writeHead' of undefined
    at Listener.webjs_handler (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/webjs.js:78:21)
    at Listener.handler (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/sockjs.js:147:12)
    at Listener.handler (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/sockjs.js:6:61)
    at Server.<anonymous> (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/sockjs.js:154:24)
    at Server.new_handler (/opt/meteor/app/programs/server/npm/ddp/node_modules/sockjs/lib/utils.js:86:19)
    at packages/ddp/stream_server.js:133:1
    at Array.forEach (native)
    at Function._.each._.forEach (packages/underscore/underscore.js:105:1)
    at Server.newListener (packages/ddp/stream_server.js:132:1)
    at packages/meteorhacks:cluster/lib/server/utils.js:11:1

This error may be unrelated

[104.236.27.101]     at Object.Meteor._nodeCodeMustBeInFiber (packages/meteor/dynamics_nodejs.js:9:1)
    at [object Object]._.extend.get (packages/meteor/dynamics_nodejs.js:21:1)
    at [object Object].RouteController.lookupOption (packages/iron:router/lib/route_controller.js:66:1)
    at new Controller.extend.constructor (packages/iron:router/lib/route_controller.js:26:1)
    at [object Object].ctor (packages/iron:core/lib/iron_core.js:88:1)
    at Function.Router.createController (packages/iron:router/lib/router.js:201:1)
    at Function.Router.dispatch (packages/iron:router/lib/router_server.js:39:1)
    at Object.router (packages/iron:router/lib/router.js:15:1)
    at next (/opt/meteor/app/programs/server/npm/webapp/node_modules/connect/lib/proto.js:190:15)
    at Object.Package [as handle] (packages/cfs:http-methods/http.methods.server.api.js:420:1)

It is causing instability on our website is there a known fix for this?

My microservices allocated on DigicalOcean droplets. Sometimes cluster throw this errors.

stream error Network error: ws://10.133.235.201:3000/websocket: connect ECONNREFUSED

I'm using 4 EC2 instances with Meteor-Up + Cluster for setup. This is my Meteor-up configuration:

"servers": [
    {
      "host": "server-1.myserver.com",
      "username": "ubuntu",
      "pem": "./ssh.pem",
      "env": {
        "CLUSTER_BALANCER_URL": "https://server-1.myserver.com",
        "CLUSTER_ENDPOINT_URL": "https://server-1.myserver.com"
      }
    },
    {
      "host": "server-2.myserver.com",
      "username": "ubuntu",
      "pem": "./ssh.pem",
      "env": {
        "CLUSTER_BALANCER_URL": "https://server-2.myserver.com",
        "CLUSTER_ENDPOINT_URL": "https://server-2.myserver.com"
      }
    },
    {
      "host": "server-3.myserver.com",
      "username": "ubuntu",
      "pem": "./ssh.pem",
      "env": {
        "CLUSTER_BALANCER_URL": "https://server-3.myserver.com",
        "CLUSTER_ENDPOINT_URL": "https://server-3.myserver.com"
      }
    },
    {
      "host": "server-4.myserver.com",
      "username": "ubuntu",
      "pem": "./ssh.pem",
      "env": {
        "CLUSTER_BALANCER_URL": "https://server-4.myserver.com",
        "CLUSTER_ENDPOINT_URL": "https://server-4.myserver.com"
      }
    }
]

But when I check the 4 instances, the instance-1 CPU usage is much more than instance-2 (34% compared to 5%), instance-3 & 4 consumes about 10%.

I've also checked the server log and see that log from instance-1 is much more than the other's.

Is there anything wrong with my configuration?
Thank you so much for your help.