Migrating from AngularJS to Angular

  • 2019-05-22 02:43 AM
  • 1608

Migrating from AngularJS to Angular a hybrid system architecture running both AngularJS and Angular

Intro

Dealing with legacy code/technologies is never fun and the path to migration isn’t always as straight forward as you want. If you are a small startup, trying to balance business requirements, scarce resources and aggressive deadlines, it becomes even more complicated.

This is the situation one of the startups I was advising was facing.

A bit of background

The startup was developing a SaaS for the last 2 years and (at the time) had around 15 clients worldwide. In these 2 years their code base grew pretty fast and lead to quite a lot of fast/reckless written code. There was nobody to be blame, this is pretty common in the startup world when business needs move way faster than you expect and you start sacrificing code qualify for quantity.

The system architecture was pretty simple. 
• a frontend application written in AngularJS (split into multiple modules that were selected at build time depending on the clients’ configuration)
• a backend application written in Python 2.7 and Django 1.9 using a Mysql database
• Celery for running async tasks

Each client would get their own isolated environment deployed on AWS:
• Apache in front of the Django application (deployed on multiple EC2 instances behind an ELB)
• AngularJS build deployed on individual S3 buckets with CloudFront in front of them

Path to migration

A few months before starting the migration, development was getting very slow, features were not coming out as fast, deadlines were missed and clients were reporting more issues with every update that we were rolling out. It was at this time that we started thinking more seriously about some kind of refactoring or major improvement.

We didn’t know exactly what we were going to “refactor/improve” so we started off by answering three questions (I recommend that anyone who is thinking about a migration/refactoring think really hard about the how to answer them):

1st question: Why is refactoring necessary now ?

This is a very important questions to answer because it helps you understand the value of the migration and also it helps to keep the team focused on the desired outcome. For example because i don’t like the way the code is written isn’t good enough reason. The reason has to have a clear value proposition that somehow directly or indirectly benefits the customers.

For us it was mainly three things:
 1. feature development was becoming painfully slow;
 2. code was unpredictable. we would work in one part of the application and break 3 other parts without realizing;
 3. single point of failure: only 1 engineer knew the FE code base completely and only he could develop new features on the codebase (this is out of a team of only 5 engineers)

So our goal was simple:

improve FE development velocity and remove the simple point of failure by empowering other engineers to develop FE features

2nd question: Who is going to do the migration ?

You can answer this question either now or after the 3rd question. Depending on the size of the company and on the available resources it can be one person, several people, an entire team, etc…

We were in a difficult situation. The only developer who could work on this couldn’t because he was busy building critical features for our customers. Luckily we had one senior backend engineer who wanted to get some FE exposure so he volunteered to take on the task. We also decided to time-box a proof of concept at 2 weeks. We did this because we didn’t know how long it would take to figure out a solution or whether the engineer could actually do this task since he hadn’t worked on FE before.

3rd question: What are we actually going to do ? The answer here usually involves some discovery time, a few tech proposals and a general overview of the options with the entire team while weighing the pros and cons of each.

For us one thing was clear from the start: we didn’t want to invest any resources into learning/on-boarding engineers on AngularJS. AngularJS had already entered Long Term Support and we didn’t want to have our engineers invest time in something that might not benefit them long term. This meant that refactoring the existing AngularJS code was not an option. So we started looking at Angular6 …

The migration

There a multiple approaches on how to have a hybrid app running different frameworks. After reviewing some options we decided that — for us — the best way to move forward was to simply have 2 separate FE applications deployed: the legacy AngularJS one and the new Angular one. This meant that any state on one app could not be transferred to the other application, which wasn’t such a big deal for us since only new modules were going to be developed using Angular and our modules didn’t share state with each other.

From the client’s perspective everything would look like one application, except for a page reload when they would move between the applications.

Pros to this approach

  • speed: get something up and running without untangling legacy code
  • safety: no risk of breaking the current live app since it would be a new code based deployed next to the old one (especially since a developer with no previous exposure to the project was working on it)
  • stop legacy development: we stop adding more code the an already unmanageable codebase

Cons to this approach:

  • maintaining legacy code: it didn’t address feature improvements on existing modules; old modules would still be in AngularJS for an undefined period of time
  • duplicating parts of the code: since the new app had to look and feel like the old one any themes, custom components would have to be written in both places. Also some parts of the layout would have to be duplicated in new app (like header, menu, etc… ) and any changes to those components would have to be done in both apps

We already knew of a new module that we wanted to build so we started form scratch with a new Angular 6 project and we used this new module for our 2 weeks proof of concept.

Step 1— same domain

Have both apps running on the same domain so that they have access to the same cookies and local data. This was extremely important since only the AngularJS app would continue handing authentication & authorization.

Step 2— look and feel

Both apps The goal was to make the new app look the same as the original application. So we: 
 • copied over all the stylesheets
 • implemented the base layout of the application (header & menu drawer)

Step 3 — authentication & authorization

We had to duplicate the authorization logic in the Angular6 app and make sure the correct session tokens were available to allow access to the module

Step 4— routing between apps

Since our main navigation links would take you to either app, we decided do move all that logic to a backend service called menu-service. This would eliminate the need to write any navigation changes in both apps and also would allow for greater runtime control over what navigation buttons we show.

Example:

HEADER: Authorization: Bearer xxxxx
GET menu-service/v1/menu/?type=0|1 (0: legacy, 1: new)
[{
  "slug": "refresh",
  "name" : "Refresh",
  "icon" : "fa-refresh",
  "type" : 1  
 }, {
  "slug": "module1",
  "name" : "Module1",
  "icon" : "fa-module1",
  "type" : 1
}, {
  "slug": "module2",
  "name" : "Module2",
  "icon" : "fa-module2",
  "type" : 0
}, {
  "slug": "logout",
  "name" : "Logout",
  "icon" : "fa-logout",
  "type" : 0
}]

In the above example based on the type value we identify that the module1 and refresh are links towards the new application while module2 and logout are links in the old application.
This information allows each application to decide whether to use the internal routing mechanism or do a window.location redirect

Example of routing in the Angular app (AngularJS does something similar):

export class MenuService {
  constructor(private router: Router) {  }
  onMenuItemClicked(menuItem): void {
    if (menuItem.type === 1) {
      this.router.navigate([menuItem.slug])    
    } else {   
      const url = `${legacy_endpoint}/${menuItem.slug}`;
      window.location.href = url      
    } 
  }
}

Step 5— building/deployment on a real environment

Like i mentioned in the beginning the AngularJS application was deployed to an AWS S3 bucket and exposed through Cloudfront to take advantage of the massively scaled and globally distributed infrastructure offered by AWS.

The result we wanted was the following: anything that has the url [https://hostname/v2](https://hostname/v2)/ is routed to the Angular application and everything else is routed to the legacy AngularJS app.

We used base-href and to make sure our Angular6 application builds accordingly

ng build --base-href /v2/ --deploy-url /v2/

Unfortunately we didn’t manage to achieve the desired routing behavior with AWS Cloudfront. This was a big disappointment since we had to pivot to a less optimal solution. (if anyone has any suggestion on how to do this in Cloudfront i’d love to hear it)

We ended up with the following structure:
• each app deployed in a NGINX Docker container

# AngularJS — Dockerfile:
FROM nginx:alpine
COPY dist /usr/share/nginx/html
--------------------------------------------------------------------
# Angular6 — Dockerfile:
FROM nginx:alpine
COPY dist /usr/share/nginx/html/v2

• AWS ALB with path routing

Step 6: Local development

Local development for the AngularJS application didn’t have to change. However in order to develop on the Angular6 app you had to also run the AngularJS application to be able to authenticate and get the appropriate session tokens.

We were already using Docker for deploying our application as containers. So we added a Makefile target to run the latest from our Docker repository

# Angular6 — Makefile:
AWS_REPOSITORY = xxx.dkr.ecr.eu-central-1.amazonaws.com
JS_APP_NAME = angular-js
...
run-local: 
  docker run -p 8080:80 $(AWS_REPOSITORY)/$(JS_APP_NAME):latest

Conclusion

This might not be the cleanest or optimal solution, however it was the fastest way towards our goals. And this was the most important thing to us.

The goal of this post isn’t to teach you how to do a AngularJS to Angular6 migration but instead is to showcase our path when dealing with such a task.

Further reading:

An in-depth look at Angular’s ng template

Angular 8 Data (Event & Property) Binding Tutorial: Build your First Angular App

Angular 8 Forms Tutorial - Reactive Forms Validation Example