CosmicAC Logo

Upgrade the CosmicAC stack

Upgrade a running CosmicAC deployment to a new image release.

Upgrade a running CosmicAC deployment to a new image release.

In-cluster job images upgrade separately

task update upgrades only the Compose platform stack, not the GPU and inference images that run as jobs in your Kubernetes cluster. To change those, update their tags in the Kubernetes worker's config/common.json.

Prerequisites

You need the following before you start:

  • A running CosmicAC deployment. See Deploy CosmicAC.
  • The deployment repository on your host.
  • The new image TAG you want to deploy.

Steps

Back up the current state

Save a snapshot you can restore if the upgrade fails:

task backup

Set the new tag

Set TAG=<newtag> in .env, or pass it inline in the next step.

Run the update

Pull the new images, add any new config keys, and recreate the stack at the new tag:

task update

To skip editing .env, pass the tag inline with task update TAG=<newtag>. To scope the upgrade to one service, add SERVICES="cosmicac-app-node".

Change a config key's value (optional)

Do this step only if you want a different value for one of the new keys. task update sets each new key to the release default and leaves your existing values unchanged.

Find the key in the service's config/common.json, then set it in .env. This example sets pagination.limit to 100 for app-node:

APPNODE_COMMON_CONFIG__pagination__limit=100

Apply the change, then restart the service:

task apply-app-node-common-config
task restart SERVICES="cosmicac-app-node"

See Task deployment commands for each service's prefix and apply command, and for the other variable forms, including the ones that set values in config/facs/.

task update restarts the services it upgrades, so this restart applies only to your .env change. Any later change you make in .env needs both commands, because the apply command writes the config file without starting containers.

Verify the deployment

Confirm every service is up:

task ps

For the full health checks, see Verify the deployment.

If a worker won't connect, refresh the shared runtime keys, then recreate the worker so it picks them up:

task wire
task restart SERVICES="cosmicac-wrk-server-k8s-nvidia"

Roll back

To roll back, deploy the previous tag. The rollback preserves your state, but back up first for a snapshot:

task backup
task update TAG=<previousTag>

Next steps

On this page