servatus
Run resumable work through Slurm and atomically publish validated outputs.
Resumable Slurm work and atomic publication.
// overview
Servatus is zero-runtime-dependency Python infrastructure for resumable application work. Campaign submits immutable ordered tasks to Slurm through native OpenSSH and Apptainer, while Workspace preserves private work and atomically publishes application-validated outputs.
Resilience is concentrated at distributed submission and POSIX publication. Durable intent and receipt state prevents silent duplicate submission; identity locks and no-replace commits prevent concurrent corruption or overwrite. Applications retain task meaning, schemas, validation, and completion decisions.
// how it works
Campaign freezes task order, arguments, and payload digests; plans balanced homogeneous allocations with exact CPU, memory, and GPU resources; renders quote-safe Slurm scripts; then records intent before submission and receipts after acceptance.
Workspace binds hidden state to an application destination and opaque identity. Draft assembles same-filesystem files, syncs content, and uses a kernel-enforced no-replace rename to expose one absent-or-complete immutable directory.
// capabilities
Durable Slurm campaigns
Packs ordered tasks into exact resource allocations and executes them through native OpenSSH, Slurm, and Apptainer.
Submission recovery
Uses durable intents and receipts, explicit retries, and bounded ambiguity resolution to avoid silent duplicate work.
Resumable workspaces
Preserves private checkpoints across failures while binding each workspace to one opaque request identity.
Atomic publication
Syncs and commits application-validated outputs without exposing partial or overwritten destinations.