Understanding Best Practices for kosh infrastructure on Kubernetes

[dennyabrain]

Infrastructure requirements of Kosh are the following software

For the current version :
1 : elasticsearch : to store and retrieve vectors and full text search
2 : rabbitmq : for queues
3 : flask server which expose certain ML features : this can be large (1-2 GBs) in terms of memory depending on the ML model loaded
4 : nodejs server : basic REST api, nothing special here

For a future version :
5 : redis : to implement caching
6 : mongodb or sql : right now we use a managed service for these but eager to explore if our underutilized k8 cluster could be used for this to save money.

I am opening this discussion up for anyone with experience/expertise to chime in to help us put together k8 files to create and run this deploymnt.

Some doubts to begin with :

1. Is it ok to run a single node elastic search cluster for now since the traffic is less and scale up as we grow.
2. since the flask server can be large, it takes upto 10 seconds to come up, what are best practices on dealing with such pods

[dennyabrain]

Here's the docker compose file for the search server (Feluda). It has services for the backend flask server, elasticsearch and rabbitmq.
https://github.com/tattle-made/tattle-api/blob/4fe2143156c93f6e99a2a1f89a1ecd4e4537f3b5/docker-compose.yml#L1
Hopefully this is useful to create kubernetes config files.