Repository navigation
Workbench and runs
The default workbench is a lightweight JupyterHub KubeSpawner user server with a shared research home. One server supports many browser tabs and JupyterLab workspaces. These are not separate GPU allocations. Named server support is a future, bounded design; older BatchSpawner manifests are optional non-default infrastructure templates.
From the authenticated portal, select Workbench/Notebooks and launch only after a research POSIX identity has been provisioned. The backend expects a configured JupyterHub API, signing key and user UID/GID. A missing UID/GID produces a provisioning error rather than an anonymous workspace.
Keep notebooks under /home/research/<username>. Verify imports/kernel and use small examples interactively. To stop the workbench, use the portal's stop control; stopping it should not delete persisted home files.
The portal's Runs surface wraps ExecutionRecord and a restricted Slurm gateway. At the time of the reviewed source, executable reference offerings include cpu-smoke and bell-aer endpoints. Eligibility, current configuration, walltime, image selection and cluster availability must be checked by the server. GPU and QPU hardware execution require separate explicit entitlement and specialized runners.
portal action -> authenticated execution API -> server-side entitlement
-> restricted SSH Slurm gateway -> job -> Quantum Workflows runner
-> shared result directory / manifest.json / summary.json
-> ExecutionRecord and result view
A submission response is not final completion. Read scheduler state and exit status, verify scientific checks, then inspect reproducibility fields. Avoid running jobs that outlive a workbench kernel through an unmanaged notebook background process.
A workbench launch failure can be JupyterHub configuration, identity or shared storage. A queued Slurm job can be legitimate resource contention. A failed workflow can have successful scheduler dispatch but failed numerical validation. A missing run record is a database/proxy issue; a missing artifact is workflow/permissions/storage.
References: workbench contract, workbench_api.py, execution_api.py.
Quantum Platform
- Home
- Getting started
- Account and access
- Researcher guide
- Workbench and runs
- Agent and Guide
- Administrator guide
- Developer guide
- API and integration
- Troubleshooting
- Security
- Contributing
- Guide integration and support Detailed topics
- Programme approval and resource entitlements
- Authentication MFA and key management
- JupyterHub and SSH identity
- API development and tests
- Research programme administration Guide integration
- Guide knowledge and source ingestion
- Guide support and ticket workflows