Step 1 - make a monolithic architecture Step 2 - convert it to serverless, telling that it will make working with it better Step 3 - move back to monolithic and write articles how you made it 90% fast and saved 69.7% cost 😂
Hey mehul, great video. Can we get a video on how you handle splitting the node processes to run on each server available on your machine. Like what would production way of doing the things would be.
I want to understand how you manage multi tenancy for fermion , is it seperated by database or is it seperated by servers, Because both have its pros and cons to discuss. Please tell us if possible
Well it was expected in case of fermion l that serverless might end up over spending after a threshold due to white label url offering because addition of a single school can suddenly increase the no. of request but congrats for the anticipated cost saving btw just curious that now you will change your db architecture as well because as far as I remember fermion uses Postgres with an ec2 sitting in between to resolve that infinite connection issue due to lambda creating a pool and now since you moved away from lambda that issue will get resolved by default, right? So need of that db proxy ec2 and now AWS also announced a good price reduction for their on demand RDS offering ( i guess for provisioned too but not sure) so looks like decent cost saving 🔥🔥
majority of places I've seen k8s they could have ECS/Fargate, container lambdas, or just run them on EC2 instances and avoided teams worth of devops platform engineering people and the teams spawned by the consequences of having those teams and so on pretending they're at Google
Aside from devops complexity as others mentioned, for a bare minimum prod cluster you are looking at these extra VMs and the associated cost for just spinning up a K8s cluster in prod. K8s is a terrible idea for startups. 3 VMs for the control plane (to ensure HA). 2 or more worker VMs, depending on workload. 1-2 additional VMs for monitoring, logging, ingress, or other support services.
how do you design your fermion app is it like unadecdemy design like this ==> 21. Unacademy adminsArr:[adminsName,adminsvdosArr:[vdo,likes,comments:[],pubclictionDate:[],],AdminHistory:[userName,vdo ]] usersArr:[userName,usersHistory:[adminsName,vdo] ]
Step 1 - make a monolithic architecture
Step 2 - convert it to serverless, telling that it will make working with it better
Step 3 - move back to monolithic and write articles how you made it 90% fast and saved 69.7% cost 😂
🤣🤣🤣
I was going to write the same thing 😂
Developers during midlife crisis :
Moves from monoliths to serverless to monoliths...
AWS serverless is really good for validating your idea. Once you have that you can go to dedicated machines.
Hey so you are using elastic load balancer or nginx for load balancing??
Hey mehul, great video.
Can we get a video on how you handle splitting the node processes to run on each server available on your machine. Like what would production way of doing the things would be.
Next Video - How we fix high memory usage with Node. (The soln is to migrate to rust)
haha now I think fermion is build for teaching us good stufss
Which backend framework are you using?
Back to basics
I want to understand how you manage multi tenancy for fermion , is it seperated by database or is it seperated by servers,
Because both have its pros and cons to discuss.
Please tell us if possible
Well it was expected in case of fermion l that serverless might end up over spending after a threshold due to white label url offering because addition of a single school can suddenly increase the no. of request but congrats for the anticipated cost saving btw just curious that now you will change your db architecture as well because as far as I remember fermion uses Postgres with an ec2 sitting in between to resolve that infinite connection issue due to lambda creating a pool and now since you moved away from lambda that issue will get resolved by default, right? So need of that db proxy ec2 and now AWS also announced a good price reduction for their on demand RDS offering ( i guess for provisioned too but not sure) so looks like decent cost saving 🔥🔥
Why didn't you go with kubernetes and it's auto scalable architecture and maintenance
Mostly you don't need that much it is for big companies like Google to manage huge user base
majority of places I've seen k8s they could have ECS/Fargate, container lambdas, or just run them on EC2 instances and avoided teams worth of devops platform engineering people and the teams spawned by the consequences of having those teams and so on pretending they're at Google
Aside from devops complexity as others mentioned, for a bare minimum prod cluster you are looking at these extra VMs and the associated cost for just spinning up a K8s cluster in prod. K8s is a terrible idea for startups.
3 VMs for the control plane (to ensure HA).
2 or more worker VMs, depending on workload.
1-2 additional VMs for monitoring, logging, ingress, or other support services.
How do you write backend api design for fermion is it like api design of unacadmey
thanks for the video, but it is confusing in someparts.
how do you design your fermion app is it like unadecdemy design like this ==>
21. Unacademy
adminsArr:[adminsName,adminsvdosArr:[vdo,likes,comments:[],pubclictionDate:[],],AdminHistory:[userName,vdo ]]
usersArr:[userName,usersHistory:[adminsName,vdo] ]