Deployments

A Deployment is a versioned snapshot of your component configuration — the set of activities, workflows, webhook endpoints, and cron tasks that Obelisk should run. Every time you push a new deployment.toml to the server, Obelisk stores it as a new Deployment record in the database.

What a Deployment stores

Each Deployment record contains:

• A unique Deployment ID (e.g. Dep_01KN1NK9A6NMHP31TVJ875X599).
• The verbatim deployment.toml submitted by the client.
• Deployment-owned source files referenced by JS components, exec activities, and backtrace source maps, stored in a content-addressed file store.
• A deployment digest, computed from the stored manifest.
• The status: enqueued, active, or inactive.
• Timestamps (created_at, last_active_at).

Because the manifest is stored verbatim and deployment-owned sources are stored with it, the server can recreate a deployment directory and provide source context to replay, advance, logs, and backtrace inspection even after the active deployment has moved on.

Execution - Deployment relationship

Every execution carries a reference to the Deployment that was active when it was locked for processing. This means:

• You can filter executions by deployment_id in the Web API or Web UI.
• Replay and upgrade operations know exactly which component version handled each event.
• Hot redeploy (see below) switches activity and webhook work to the new deployment immediately, while in-flight workflow executions are replayed and upgraded automatically when using the default auto locking strategy.

Deployment lifecycle

• Enqueued — The deployment is marked as the next one to activate. It will become active on the next server restart (or immediately via hot-redeploy).
• Active — The server is running with this deployment. New executions reference this Deployment ID.
• Inactive — Any new or old deployment stored in the database.

Only zero or one deployment can be enqueued and one can be active at any given time.

Hot redeploy

Hot redeploy activates a new deployment without restarting the server. The behavior differs by component type:

• Activities — Activities default to the by_ffqns locking strategy. Pending executions are selected by the currently active FFQNs and start running on the newly deployed activity workers transparently, without replay or an upgrade.
• Workflows — With the default auto locking strategy, in-progress executions are replayed against the current workflow component after redeployment. Compatible executions are upgraded and associated with the current deployment. If replay detects nondeterminism, the execution stays on its previous component version. With by_ffqns, executions are selected by the functions exported by the current component without an automatic component upgrade. With by_component_digest, executions remain pending until explicitly upgraded (see Upgrading workflow executions below).
• Webhooks — Existing TCP connections continue to be served by the old deployment. New connections use the new deployment.

Hot redeploy is triggered via:

obelisk deployment apply <PATH|ID>           # CLI: submit + hot-redeploy

or via the Web API:

curl "127.1:5005/v1/deployments/Dep_.../switch" \
  -H content-type:application/json -X PUT -d '{"hot_redeploy": true}'

If a configuration change cannot be applied hot (e.g. a component that requires a full restart), the API returns "restart_required" and the deployment is enqueued for the next startup.

Upgrading workflow executions

The default auto locking strategy upgrades compatible in-flight workflow executions automatically. If a workflow uses by_component_digest, existing in-flight executions continue to reference the old component digest. To migrate one to the new version, use the upgrade API:

OLD=$(curl "127.1:5005/v1/executions/${EXECUTION_ID}/status" -H accept:application/json | jq .component_digest)
NEW=$(curl '127.1:5005/v1/components?name=my_workflow' -H accept:application/json | jq .[0].component_id.input_digest)

curl "127.1:5005/v1/executions/${EXECUTION_ID}/upgrade" \
  -H content-type:application/json -X PUT \
  -d "{\"old\": $OLD, \"new\": $NEW}"

The RPC replays the execution history against the new component and refuses to switch if the new workflow code is not replay-compatible — unless skip_determinism_check is set to true.

Managing deployments

CLI:

Web API: see Programmatic Access — Deployments.

gRPC: see the DeploymentRepository service in proto/obelisk.proto.