notation destroy
notation destroy removes all Notation-managed cloud resources in a stage in reverse dependency order and deletes their state records. Every run logs an execution ID upfront so an interrupted teardown can resume with --execution-id.
Usage
Run notation destroy from your project root to tear down defaultStage, or pass --stage and --config to target another deployment:
notation destroy
notation destroy --stage prod
notation destroy --config apps/orders/notation.config.ts --stage staging run destroy
run destroy --stage prodBefore deleting resources, destroy compiles your entry file and evaluates the graph without declaration checks so validation errors never block teardown.
It then prints <app>:<stage>, an execution UUID, and the resume command:
Compiling infrastructure /project/infra/orders.ts
Compiling runtime modules
Destroying orders:dev
Execution ID 7f3e9b1a-4c2d-4a8e-9b1c-2d5e6f7a8b9c
To resume after a crash, run notation destroy --execution-id 7f3e9b1a-4c2d-4a8e-9b1c-2d5e6f7a8b9c
… delete lambda:create-order:role-policy aws/lambda/LambdaRolePolicy
✓ delete lambda:create-order:role-policy aws/lambda/LambdaRolePolicy
… delete lambda:create-order aws/lambda/LambdaFunction
✓ delete lambda:create-order aws/lambda/LambdaFunction
… delete sqs-queue:orders aws/sqs/Queue
✓ delete sqs-queue:orders aws/sqs/QueueUnlike notation deploy, notation destroy always compiles first and bypasses resource loss-protection checks (rejectChange and rejectReplace).
Options
notation destroy accepts project, output, and execution flags (see CLI Flags for syntax and validation rules):
| Flag | Purpose |
|---|---|
--config <path> | Loads a specific config file instead of the nearest notation.config.ts |
--stage <stage> | Targets a specific stage instead of defaultStage |
--json | Emits versioned NDJSON events on stdout |
--pretty | Forces readable terminal lines |
--execution-id <id> | Resumes an interrupted destroy execution or assigns the execution UUID |
If neither --stage nor defaultStage is set, the command exits with an error before compiling or acquiring a stage hold.
JSON output
In JSON mode, destroy emits "version": 1 NDJSON events to stdout and sends banners to stderr.
Stream structured deletion events in CI or scripts with --json:
notation destroy --stage prod --jsonEach line on stdout represents one reconciler event:
{"version":1,"level":"info","event":"reconciler.operation.lifecycle","operation":"delete","status":"start","resourceId":"lambda:create-order","resourceType":"aws/lambda/LambdaFunction"}
{"version":1,"level":"info","event":"reconciler.operation.lifecycle","operation":"delete","status":"success","resourceId":"lambda:create-order","resourceType":"aws/lambda/LambdaFunction"}Deletion order
destroy acquires an exclusive hold on <app>:<stage> and deletes orphans first, then declared resources in reverse dependency order. destroy also deletes any retired instances from earlier replacements before deleting the active resource.
Interrupted creates
If an earlier deploy crashed mid-create, destroy checks whether the remote resource exists before cleaning up state. Missing remote resources or unknown keys clean up state without failing.
Dependency cycles
Resources inside a dependency cycle tear down in reverse deploy order so cross-references detach cleanly during teardown.
Unregistered resource types
If an orphaned resource's provider package or type cannot be loaded, destroy logs a warning and preserves that record and its dependencies. Credential source records are removed without warnings.
Provider delete behavior
Most resources execute their provider delete APIs directly, with these exceptions:
| Resource type | Behavior |
|---|---|
aws/eventBridge/DefaultEventBus | Clears only local state and leaves the default bus intact |
cloudflare/secrets-store/SecretsStore | Clears only local state while deleting individual SecretsStoreSecret records remotely |
cloudflare/containers/ContainerImage | Clears only local state and keeps pushed registry tags intact |
cloudflare/workers/WorkerScript | Fails deletion if external Workers or Pages projects still bind to it |
cloudflare/r2/R2Bucket | Fails deletion if objects remain in the bucket |
Resuming an interrupted destroy
If notation destroy stops mid-run due to a crash or network failure, pass the printed execution ID to --execution-id:
notation destroy --execution-id 7f3e9b1a-4c2d-4a8e-9b1c-2d5e6f7a8b9cOmit --stage when resuming so Notation reads the saved stage from state, and note that only destroy execution IDs can resume notation destroy. Failed executions cannot resume – fix the blocker and run notation destroy without --execution-id, or see Recovery if a crashed run left a stale stage hold.
Exit codes
notation destroy exits with 0 when all tracked resources and state records are removed and 1 on compilation, admission, credential, or provider errors.