One workspace, one manual execution
An Azure Container Apps job is a resident runtime: each workspace is one manual-trigger execution of a job you created. Grove starts, reads and stops executions, and the job and its history stay yours.
sequenceDiagram
participant D as Grove daemon
participant R as Container Apps job driver
participant A as az and Container Apps API
participant X as Job execution
participant W as grove-native-worker
D->>R: up(workspace spec)
R->>A: az containerapp job start, with the full container
A->>X: start one replica
A-->>R: execution name
R-->>D: handle saved on the workspace record
X->>W: start the worker
W->>D: dial out, GET /mailboxes/connection
D-->>W: mailbox messages and steer frames
W-->>D: acks, reads and the branch it pushes
How Grove drives it
The job is the deployment, an execution is the workspace, and Grove reads every execution by its name.
- You create the job once. Grove emits a manual-trigger template for
az containerapp job create --yaml. The daemon token is a job secret whose value you set, and the template only names it. - Each start carries the whole container. A start override replaces the full template for that execution, so Grove always sends the complete container rather than a delta.
- Status is read by execution name. Container Apps keeps only the latest hundred executions in its history, so reading by position would drift onto a stranger's execution.
- An Azure Files share can be mounted into each execution when you name one, for anything that should outlive the replica.
- Nothing dials in. The worker dials the daemon, so the job needs no ingress.
Freeze, thaw and eviction
Container Apps has no suspend and offers no delete for an execution, so stop, freeze and remove all end it the same way.
- Freeze stops the execution. The worker pushes the branch first, and the branch is the saved state.
- Thaw starts a new execution from the frozen one's own template, read back from Azure and marked as a resume. The handle changes, and Grove saves the new one.
- Remove stops the execution and leaves it in history. A removed workspace may still read as stopped afterwards, never as running.
- One execution runs until the replica timeout, a day by default. When the platform ends it, the branch is what survives.
- A quota or capacity refusal is final. Grove surfaces it on the first answer and does not retry.
Configuration
.grove/config.json
{
"runtimes": {
"azure": {
"kind": "aca_job",
"subscription": "00000000-0000-0000-0000-000000000000",
"resource_group": "grove-rg",
"environment": "grove-env",
"job": "grove-workspaces",
"image": "acme.azurecr.io/grove/runtime:latest"
}
}
}
client_secret_env names the variable holding the service principal secret. Every field is in the Configuration Reference.