cd - run dev_core locally via nomad as a prereq for CD to aws

[noah edward hall]

C

• terraform should be setup to manage the infrastructure, now we need another pipeline for code deployments
• before going to far with nomad integration, ensure the entire stack runs locally via nomad jobs consuming docker image artifacts output from the normal development cycle

T

• nomad
  • create a server configuration for dev
  • continue to use compose to build images, and use those tagged images in nomad task configs
  • extract values needed for nomad out of the compose file and into an .env file for consumption by nomad as an input var file
    • you can use yq to convert docker compose convert into json and consume the json
      • save json: docker compose convert | yq -r -o=json > ./apps/nirvai-core-nomad/dev/.env.json
      • save yaml: docker compose convert > ./apps/nirvai-core-nomad/dev/compose.yaml
      • read the levant template docs
        /docs/templates.md)
    • read the nomad pack docs
    • you can use hcl2 for jobspecs
    • set up local registry for use by nomad to pull images from
    • nomad vault task
      • convert to build + dockerfile instead of image so we can tag it via compose
        • ${PROJECT_NAME}-${SERVICE_NAME}:${NODE_ENV:-development}
    • nomad postgres task in core-db group
    • nomad bff task in core-bff group
      • fix node bff sleep recursion waiting on vault to be ready
      • use esbuild define logic to inject env vars into build and remove dotenv (which requires a file)
    • nomad proxy task in core-proxy group
      • convert to build + dockerfile instead of image so we can tag it via compose
        • ${COMPOSE_PROJECT_NAME}-${SERVICE_NAME}:${NODE_ENV:-development}
      • might as well handle the server conf on this ticket

A

• nomad successfuly runs all cotnainers (sans UI)

  • skipping UI as the architecture is completely different (lambda, no server)
  • while nomad runs each service, there not connected
    • the work to translate docker networks into compose architecture would be futile, as theres a different set of requirements of usecases

• all that work on levant to only find this

• check how hashicorp does their userdata scripts

• new workflow

  • docker based: continue to run nirvai-core via docker compose;
    • anticipate 0 changes to this workflow, however preference should be for nomad to have closer parity with prod
  • nomad based
    • set NODE_ENV for tagging all images
    • docker compose build: build images, will auto tag based on current NODE_ENV value
    • start nomad: will find the images tagged with the current NODE_ENV and use them for each task group
    • accordsing to @danielabar there may be issues around caching with nomad
      • tips and tricks by daniela

• fixed bff sleep logic: now bff wont automatically throw (removed throw ZombieError()) and will just recursively sleep and try forever