sparkrun stop
sparkrun stop [target] [options]Stop a running inference workload across its executor destinations. TARGET can be a recipe name or a workload ID (from sparkrun status output). When --all is used, TARGET is optional and discovered sparkrun workloads on the selected cluster will be stopped.
Confirmed stops release the workload’s model-cache claims on the stopped hosts and restart the cache retention age. If claim release fails, retry normal stop. See cache claim reconciliation when workload metadata is missing; stopping workloads does not itself delete model files.
Options
Section titled “Options”| Option | Description |
|---|---|
--all / -a | Stop every sparkrun workload on the cluster. No target needed. |
--hosts / -H | Comma-separated host list |
--hosts-file | File with hosts |
--cluster | Use a saved cluster by name |
--tp / --tensor-parallel | Match host trimming from run |
--port | Override port (to match run-time override) |
--served-model-name | Override served model name (to match run-time override) |
--dry-run / -n | Show what would be done |
Examples
Section titled “Examples”# Stop a specific recipesparkrun stop qwen3-1.7b-vllmsparkrun stop nemotron3-nano-30b-nvfp4-vllm --tp 2sparkrun stop nemotron3-nano-30b-nvfp4-vllm --cluster mylab
# Stop all sparkrun containers on a clustersparkrun stop --all --cluster mylabWhen selecting by recipe, use the cluster/host scope where it ran and the same
intent-changing overrides, such as --tp 2. Selecting by workload ID uses the
job’s recorded hosts and connection settings; changing the default cluster does
not redirect that connection. Explicit --hosts can override the target host list.
--dry-run resolves the target and executor and can contact hosts for discovery,
but performs no teardown or claim release.