Research and create steps for backend migration to Zappa

[arkid15r]

Describe the solution you'd like

Use https://github.com/zappa/Zappa for OWASP Nest backend.

[AbhayTopno]

@arkid15r can you please assign this to me please

[Naveen-Pal]

@arkid15r , I am familiar with serverless, can i get this one?

[harsh3dev]

@arkid15r @Naveen-Pal
What are the advantages of this migration for our project?

[Naveen-Pal]

@arkid15r @Naveen-Pal
What are the advantages of this migration for our project?

Serverless help to save cost as it is based on pay as you go and FAAS (function as a service)
It also does auto scaling as per the traffic.

In short, if no one is using the website it can automatically downscale to 0 (no cost). And up scale as per demand of service.

[Rajgupta36]

@arkid15r @Naveen-Pal
What are the advantages of this migration for our project?

Serverless help to save cost as it is based on pay as you go and FAAS (function as a service) It also does auto scaling as per the traffic.

In short, if no one is using the website it can automatically downscale to 0 (no cost). And up scale as per demand of service.

I am also curious about Zappa's serverless mechanism.
I have worked with Cloudflare Workers, which use Hono for serverless deployment. However, when the server receives no requests, it triggers a cold start mechanism, which takes some time to respond to the first API request @Naveen-Pal and @arkid15r?

[Rajgupta36]

is it better than ASG's+ load balancers or Kubernetes pods (k8s pods) for server scaling and traffic handling ?

[harsh3dev]

@arkid15r @Naveen-Pal
What are the advantages of this migration for our project?

Serverless help to save cost as it is based on pay as you go and FAAS (function as a service) It also does auto scaling as per the traffic.
In short, if no one is using the website it can automatically downscale to 0 (no cost). And up scale as per demand of service.

I am also curious about Zappa's serverless mechanism. I have worked with Cloudflare Workers, which use Hono for serverless deployment. However, when the server receives no requests, it triggers a cold start mechanism, which takes some time to respond to the first API request @Naveen-Pal and @arkid15r?

@Rajgupta36 I was also curious about the same as in hono, this behavior happens as you told.

[Naveen-Pal]

@Rajgupta36 Yes, you are right about the cold start, as you mention it take time for the first request but also make much impact on cost. Alternatively we can configure to have minimum 1 pods running.

is it better than load balancers or Kubernetes pods (k8s pods) for server scaling and traffic handling ?

In term of cost and scalibility serverless is the best option.
I think load balancers can not autoscale by itself (just distribute traffic among running pods).

[Rajgupta36]

@Rajgupta36 Yes, you are right about the cold start, as you mention it take time for the first request but also make much impact on cost. Alternatively we can configure to have minimum 1 pods running.

is it better than load balancers or Kubernetes pods (k8s pods) for server scaling and traffic handling ?

In term of cost and scalibility serverless is the best option. I think balancers can not autoscale by itself (just distribute traffic among running pods).

I think the combination of ASG's and load balancer is good for horizontal scaling . it's also cost efficient. also in asg we can describe min and max no of active machines.

[abhayymishraa]

@Rajgupta36 @Naveen-Pal provisioned concurrency maybe solve the cold start problem, by setting min no. of machines need to be active !!!!

[Naveen-Pal]

@Rajgupta36 Probably we don't need to use ASG's.
No, you cannot use ASG with Zappa because ASG is designed for managing EC2 instances, while Zappa deploys apps to AWS Lambda, which is a completely serverless service.

[Rajgupta36]

@Naveen-Pal i am not saying use ASG with zappa .
Zappa is a serverless framework, while ASG serves a different purpose. What I’m suggesting is to create Target Groups, then a Launch Template, followed by ASGs and a Load Balancer.
this is just a suggestion and not something I’m insisting on implementing

[Rajgupta36]

i think it's a good discussion on deployment approaches .
it is a good article for all of use :- https://docs.aws.amazon.com/lambda/latest/dg/lambda-runtime-environment.html

[arkid15r]

This approach seems easier than k8s based. But I'm open to assess pros and cons of both approaches before choosing the direction. @Naveen-Pal will you be able to extend the task w/ a comparison of zappa vs k8s approach?

[Naveen-Pal]

One popular method to go serverless with Kubernetes is by using Knative.

Steps Required:

Setup the Environment:

• Install kubectl, minikube, and kn (Knative CLI).
• Set up Knative components (Serving and optionally Eventing).

Containerize the Application:

• Build Docker images for each service(located in the backend directory)..

Create Knative Service Definitions:

• Define a Knative Service for each service in the application using YAML configuration files.

And finally, manage deployment...

(This approach seems overly complicated.)

---

Zappa

For simpler deployments and lower operational overhead, Zappa is a great fit.

Steps:

Setup the Environment:

• Simply run pip install zappa.
• Add a zappa_settings.json file with the required configuration.

Deploy to Production:

• Use zappa deploy production.

We will need to consider how this integrates with our CI/CD pipeline.

---

Personally, I believe Zappa is a better choice for our project since we are using Django. It simplifies the deployment process and reduces complexity.

[arkid15r]

It seems Zappa wins for now. But what would be k8s most important benefits for us if we are willing to deal with the complexity of the solution?

[Naveen-Pal]

one can migrate Kubernetes to any cloud but zappa is tied to AWS