HLD-412·Coordination & IDs
Expert · 2,000 ptsDurable Stateful AI Agent Checkpointing & Sandboxed Tool Runner
Design a Temporal / LangGraph-scale durable AI agent orchestrator executing multi-step ReAct loops with exactly-once side-effect tool idempotency and instant crash resumption.
Max p99
≤ 95 ms
Availability
≥ 99.5%
Budget
≤ $11,300
Functional Requirements
- FR-1POST /agents/{run_id}/step — Atomically checkpoints ReAct state and dispatches idempotent tool call.
- FR-2GET /agents/{run_id}/state — Streams live reasoning trace and tool outputs to clients.
- FR-3DELETE /agents/{run_id}/step — Releases canceled or expired leases/holds atomically and immediately refreshes cached availability.
- FR-4Worker/DB crashes must never execute external mutating tools twice or lose committed checkpoints.
Workload Profile
Steady Autonomous Runs58,000 State Read RPS / 13,000 Checkpoint Write RPS
Enterprise Workflow Trigger Burst155,000 Read RPS / 35,000 Step Write RPS
Active Working Set40 GB active ReAct state graphs & tool idempotency locks
Consistency GateStrict serializable step transitions + RPO = 0 checkpointing
Consistency Model
Strict serializable step-index CAS + durable write-ahead queue so crashed agent runs resume from exact step without duplicate tool side effects.
Relevant practice for
Solutions Locked
Sign in and either achieve AC or complete 3 official submissions on this question to inspect community diagrams and writeups.
Loading solutions...
Submission History (0)Select to replay
No submissions yet
Click Submit in the top bar to evaluate against all 7 judge cases and save each run here.
Click an item in Inventory above, or press B to add a component, double-click to name, C to connect.
Design Write-up
Write under ### Architecture for Diagram to take that up directly
Compulsory section · if not written yet, Diagram reads from the start of your write-up.
Architecture: required·FRs mentioned: 0/4·0 boxes · 0 edges
0 words