Before you begin: Complete Validate your image, including the Checklist: What must pass before a container can auto-deploy. An image that fails any of those three gates will not deploy.
Deployment in Index Cloud is artifact-driven. You push a correctly named image to Docker Hub, and optionally upload a correctly named model folder to your S3 location. Index detects it, scans it, validates it, and rolls it out. There is no deploy button, no ticket, and no release window to book.
That makes the naming convention the most important thing on this page. Your artifact name is the entire interface: it identifies the version and controls which regions the artifact deploys to.
Who does what
| You | Index |
|---|---|
Name the artifact to the convention, push the image, and upload the model folder if you use one. | Detects the upload, scans it for vulnerabilities, validates it, and promotes it through to the data centers. |
Watch the rollout in the Deployment Tracker dashboard and decide whether your change is behaving the way you want. | Confirms the rollout was accepted in each targeted data center and that your health and readiness endpoints report healthy. |
Roll forward with a new version if something looks wrong. | Sends optional lifecycle notifications to the shared Slack channel as the deployment progresses. |
Four things to be aware of
These are the deployment behaviors most likely to catch you out on a first or second deploy. Each one is covered in full in Deploying your services to Index Cloud.
A badly named artifact produces silence, not an error. If the tag does not match the required format, no deployment event is generated at all. The upload is ignored rather than rejected. If you push something and "nothing happens," check the name first.
Images and models deploy together. A single sync deploys an image and its model as a unit. Update only the image and your current model is redeployed alongside it, and vice versa.
A global upload overwrites region-specific deployments. Omitting the region suffix deploys everywhere, including over a region that was running a region-specific artifact. Pick one approach per artifact and stay with it.
"Rolled out" is not a health signal. It confirms the release was accepted in a data center. Whether your logic is doing the right thing is a question for your own metrics.
Note: Uploading a model folder to your S3 location triggers a deployment, exactly like pushing an image. Only upload a folder when you intend to deploy it.
Topics in this section
Deploying your services to Index Cloud. The naming convention, how to follow a rollout in Grafana, notifications and alerting, and what happens across the trial, optimization, and production phases.
Updating models in containers without rebuilding. Optional. Iterate on model data independently of your code, using versioned folders in S3 mounted at
MODEL_PATH.
For the exact version format, the optional annotation, and the _AMS, _EMEA, and _APAC region suffixes, see Artifact naming conventions and deployment regions.
Next
Once your container is live and taking traffic, the work shifts to Monitor your performance.
