Make NodeJs handle 5x request with 99.9% uptime adding 10 lines of code

Scale your NodeJs Server to Utilize Full Resources

Node.js works in single-threaded, non-blocking performance, working as a single process in CPU. No matter how powerful server is used and the resources utilized, what a single-threaded process can do is limited. Node.js is designed for building distributed applications with multiple nodes, hence the name Node.js.

Workload is one of the main reasons we start scaling our application, including availability and fault tolerance among others. Scaling can be done in multiple ways, one of the easiest available solution is Cloning. We can perform cloning using Cluster Module provided by Node.js.

Before we start handling requests with our resource-utilized Node.Js server, let’s take the basics of how the Cluster module works.

How Cluster Module Works?

The cluster module has got two types of processes, Master and Worker. All incoming requests are handled by Master process and Master process decides which Worker should handle the incoming requests. Worker process can be thought of as normal Node.Js single instance server which serves the requests.

How does Master process distribute the incoming connections?

  1. The first method (and the default one on all platforms except Windows), is the round-robin approach, where the master process listens on a port, accepts new connections, and distributes them across the workers in a round-robin fashion, with some built-in smarts to avoid overloading a worker process.

2. The second approach is where the master process creates the listen socket and sends it to interested workers. The workers then accept incoming connections directly.

The second approach should, in theory, give the best performance. In practice, however, distribution tends to be very unbalanced due to operating system scheduler vagaries. Loads have been observed where over 70% of all connections ended up in just two processes, out of a total of eight.

Creating a Simple Node.js Server

Let’s create a basic Node.js server that handles the request:

/*** server.js ***/
const http = require(“http”);
// get the process ID of Node Server
const processId = process.pid;
// Creating server and handling request
const server = http.createServer((req, res) => {
// Simulate CPU Work
for (let index = 0; index < 1e7; index++);

res.end(`Process handled by pid: ${processId}`);
});
// start server and listen the request
server.listen(8080, () => {
console.log(`Server Started in process ${processId}`);
});

The server gives us the following response:

Load Testing this Simple Node.Js Server:

We are going to use ApacheBench Tool, there are other tools that perform similar testing. They can be used according to personal choice.

We are going to hit our Node.js server with 500 concurrent requests for a span of 10 seconds.

➜ test_app ab -c 500 -t 10 http://localhost:8080/
This is ApacheBench, Version 2.3 <$Revision: 1843412 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking localhost (be patient)
Finished 3502 requests
Server Software:
Server Hostname: localhost
Server Port: 8080
Document Path: /
Document Length: 29 bytes
Concurrency Level: 500
Time taken for tests: 11.342 seconds
Complete requests: 3502
Failed requests: 0
Total transferred: 416104 bytes
HTML transferred: 116029 bytes
Requests per second: 308.76 [#/sec] (mean)
Time per request: 1619.385 [ms] (mean)
Time per request: 3.239 [ms] (mean, across all concurrent requests)
Transfer rate: 35.83 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 6 3.7 5 17
Processing: 21 1411 193.9 1412 2750
Waiting: 4 742 395.9 746 1424
Total: 21 1417 192.9 1420 2750
Percentage of the requests served within a certain time (ms)
50% 1420
66% 1422
75% 1438
80% 1438
90% 1624
95% 1624
98% 1624
99% 1625
100% 2750 (longest request)

This Simple server, on the level of 500 Concurrent requests, a Total of 3502 Requests were served. And 308 requests per second with Time Per Request of 1619(ms) was done.

The number of requests handled is good and it should work for most small to medium scale applications. But we haven’t fully utilized the resources and most of the available resources are sitting idle.

Implementing the Cluster Module

Now that we are up and running, let’s implement the Cluster for our server.

/** cluster.js **/
const os = require(“os”);
const cluster = require(“cluster”);
if (cluster.isMaster) {
const number_of_cpus = os.cpus().length;

console.log(`Master ${process.pid} is running`);
console.log(`Forking Server for ${number_of_cpus} CPUs\n`);
// Create a Worker Process for each Available CPU
for (let index = 0; index < number_of_cpus; index++) {
cluster.fork();
}
// When Worker process has died, Log the worker
cluster.on(“exit”, (worker, code, signal) => {
console.log(`\nWorker ${worker.process.pid} died\n`);
});
} else {
// if Worker process, master is false, cluster.isWorker is true
// worker starts server for individual cpus
//
the worker created above is starting server
require(“./server”);
}

My personal PC with i7 8th Generation has got 8 processor Cores. Considering that most of the CPUs available nowadays have a minimum of dual-core processor, the resources for the 7 remaining cores were sitting idle*.

Now, let’s run created cluster.js file, the server gives us the following response:

If you implemented the above cluster, you will have utilized your full CPU/Server power. The request is handled by Master process from same server port, which will be served by any one of the 8 Worker Processes (Servers).

Load Testing with Cluster Implemented:

➜  test_app ab -c 500 -t 10  http://localhost:8080/
This is ApacheBench, Version 2.3 <$Revision: 1843412 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking localhost (be patient)
Completed 5000 requests
Completed 10000 requests
Completed 15000 requests
Completed 20000 requests
Finished 20374 requests
Server Software:
Server Hostname: localhost
Server Port: 8080
Document Path: /
Document Length: 29 bytes
Concurrency Level: 500
Time taken for tests: 10.000 seconds
Complete requests: 20374
Failed requests: 0
Total transferred: 2118896 bytes
HTML transferred: 590846 bytes
Requests per second: 2037.39 [#/sec] (mean)
Time per request: 245.412 [ms] (mean)
Time per request: 0.491 [ms] (mean, across all concurrent requests)
Transfer rate: 206.92 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 0 1.3 0 12
Processing: 6 242 15.6 241 369
Waiting: 6 242 15.5 241 368
Total: 18 242 15.5 241 371
Percentage of the requests served within a certain time (ms)
50% 241
66% 244
75% 246
80% 247
90% 251
95% 259
98% 283
99% 290
100% 371 (longest request)

Remember the above simple Node.js server which handled 308 Requests Per Seconds on our load test now that number has increased to 2037 Requests Per Seconds that’s an outstanding 6X Increase in the number of requests handled. Also, previously Time per Request was 1619 ms now it has been decreased to 245ms. We were serving a Total of 3502 Requests before, now it has increased to a Total of 20374 Requests (That’s 5.8X increase). If you look at the implementation above this great improvement is caused by 10 lines of code. And we also didn’t have to refactor our existing server code.

Availability and Zero Down Time

Excited with the progress we have made so far, now it gets better.

When we have a single instance of server and that server crashes. There will be downtime when the server has to be restarted. Even if the process is automated, there will be delay and not a single request can be served in that time.

/*** server.js ***/
const http = require(“http”);
// get the process ID of Node Server
const processId = process.pid;
// Creating server and handling request
const server = http.createServer((req, res) => {
// Simulate CPU Work
for (let index = 0; index < 1e7; index++);

res.end(`Process handled by pid: ${processId}`);
});
// start server and listen the request
server.listen(8080, () => {
console.log(`Server Started in process ${processId}`);
});
// Warning: Only For Testing and Visualization Purpose
// Don't add the code below in production
// Let's simulate Server Randomly Crashing using process.exit()
setTimeout(() => {
process.exit(1);
}, Math.random() * 10000);

Now for simulation purposes if we add highlighted code above to our server code. And start our server we can see that one by one all server is crashing and eventually the whole process exists. Master Process also exists due to no available Worker. The server could crash and these scenarios could exist due to any problem.

Now, remember this is the case when we have 8 servers created by our Cluster Module. When we have a single instance of server and that crashes, no requests can be served in that time.

➜  test_app node cluster.js
Master 63104 is running
Forking Server for 8 CPUs
Server Started in process 63111
Server Started in process 63118
Server Started in process 63112
Server Started in process 63130
Server Started in process 63119
Server Started in process 63137
Server Started in process 63142
Server Started in process 63146
Worker 63142 died
Worker 63112 died
Worker 63111 died
Worker 63146 died
Worker 63119 died
Worker 63130 died
Worker 63118 died
Worker 63137 died
➜ test_app

When we have multiple instances of a server, the availability of the server could be easily increased.

Let’s open up our cluster.js file and add the highlighted code to cluster.js:

/** cluster.js **/
const os = require(“os”);
const cluster = require(“cluster”);
if (cluster.isMaster) {
const number_of_cpus = os.cpus().length;

console.log(`Master ${process.pid} is running`);
console.log(`Forking Server for ${number_of_cpus} CPUs\n`);
// Create a Worker Process for each Available CPU
for (let index = 0; index < number_of_cpus; index++) {
cluster.fork();
}
// When Worker process has died, Log the worker
cluster.on(“exit”, (worker, code, signal) => {
/**
* The condition checks if worker actually crashed and
* wasn't manually disconnected or killed by master process.
*
* The condition can be changed by desired error code,
* and condition.
*/

if (code !== 0 && !worker.exitedAfterDisconnect) {
console.log(`Worker ${worker.process.pid} died`);
cluster.fork();
}
});
} else {
// if Worker process, master is false, cluster.isWorker is true
// worker starts server for individual cpus
// the worker created above is starting server
require(“./server”);
}

Load Testing with Server Restart Implemented:

Let’s run the server with cluster changes implemented (Run: node cluster.js). Now let’s open up our benchmarking tool and start benchmarking our server.

➜  test_app ab -c 500 -t 10 -r http://localhost:8080/
This is ApacheBench, Version 2.3 <$Revision: 1843412 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/
Benchmarking localhost (be patient)
Completed 5000 requests
Completed 10000 requests
Completed 15000 requests
Completed 20000 requests
Finished 20200 requests
Server Software:
Server Hostname: localhost
Server Port: 8080
Document Path: /
Document Length: 29 bytes
Concurrency Level: 500
Time taken for tests: 10.000 seconds
Complete requests: 20200
Failed requests: 12
(Connect: 0, Receive: 4, Length: 4, Exceptions: 4)
Total transferred: 2100488 bytes
HTML transferred: 585713 bytes
Requests per second: 2019.91 [#/sec] (mean)
Time per request: 247.536 [ms] (mean)

Time per request: 0.495 [ms] (mean, across all concurrent requests)
Transfer rate: 205.12 [Kbytes/sec] received
Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 0 1.5 0 13
Processing: 13 243 15.7 241 364
Waiting: 0 243 16.0 241 363
Total: 22 243 15.5 241 370
Percentage of the requests served within a certain time (ms)
50% 241
66% 245
75% 248
80% 250
90% 258
95% 265
98% 273
99% 287
100% 370 (longest request)
➜ test_app

In the above load test on the level of 500 Concurrent requests and 2019 requests per second. With a Total of 20200 Requests, only 12 requests failed which means we had 99.941% uptime of the server, even with servers crashing one by one and restarting. That’s really sweet considering we only added 3 extra lines of code.

Meet me LinkedInTwitter

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store