Slurm adapter: real HPC cluster integration

[jeremymanning]

Description

adapters/slurm/src/main.rs (135 lines) has a CLI scaffold and config struct but no actual Slurm API integration. The status command returns "not yet implemented".

Requirements

• Connect to Slurm head node via slurmrestd REST API or SSH+sbatch fallback
• Advertise aggregate cluster capacity to World Compute broker
• Dispatch incoming tasks as Slurm batch jobs (sbatch)
• Monitor job status via sacct/squeue
• Report results back to World Compute data plane
• Handle Slurm-specific errors (PENDING, TIMEOUT, NODE_FAIL)
• Implement install/configure/status CLI commands

Success Criteria

• Adapter connects to real Slurm cluster
• Jobs dispatched via sbatch and results collected
• Cluster capacity advertised accurately
• Error handling for common Slurm failure modes
• Integration test on real 2+ node Slurm testbed
• Adapter appears as aggregate node in worldcompute cluster peers

Testing (Principle V)

• Deploy on real Slurm cluster (even 2-node test setup)
• Submit SHA-256 test job → verify correct result
• Simulate node failure → verify job rescheduled
• Verify resource reporting matches sinfo output

[jeremymanning]

Partially addressed by spec 004 / PR #59 — not closing. SlurmClient struct + request/response parsers exist. HOWEVER, no integration test exercises real HTTP calls to slurmrestd; adapter has not been deployed against a real Slurm cluster.

Remaining work tracked here.

[jeremymanning]

Spec 005 progress — PR #61

STATUS: No change beyond spec 004 — live integration deferred.

Slurm adapter code (parsers + HTTP client) exists from spec 004. Spec 005 did not exercise it against a live slurmrestd instance or add CI integration. Task T060 (docker-compose-based containerized Slurm fixture for CI) not done.

Remains as written in the spec-005 task list; deferred to a future session.

See notes/session-2026-04-19-spec-005-kickoff.md (§ Task completion state).