Design Delivery Batching - System Design | WinJob

Problem Statement: Multi-Order Delivery Batching

Problem Statement: Multi-Order Delivery Batching for a on-demand food and grocery delivery platform (DoorDash / Instacart / Amazon Flex scale): couriers may carry multiple orders when detour, food quality, and customer promise constraints still hold. DoorDash, Instacart, and Amazon interviews probe when to batch, how to score pairs, and how to recover when batching hurts ETA or temperature.

Mechanism (1)

Batch Command service owns batch lifecycle; route geometry is a derived projection keyed by batch_id.

Operational signals

Watch batch_accept_rate, food_degradation_violations, detour_minutes_p95, eta_slip_after_batch, unbatch_churn. Page when food_degradation_violations > 1.5% for 20 minutes in a submarket.

Domain narrative

At dinner peak, submarket SM-441 receives 1,200 ready-to-dispatch orders in ten minutes. The batch scorer proposes pairing order O-9182 (sushi, 12-min food clock) with O-9201 (salad, same strip mall). Validator rejects when projected detour exceeds 8 minutes or degradation score crosses tier threshold—offer never surfaces to courier C-5521.

Deep dive

Peak batch offer rate ≈ 4,500/s globally with partition key submarket_id. Never let analytics consumers mutate batch state; only Batch Command commits transitions. Customer-facing ETA must recompute on every batch mutation, including silent unbatch after accept.

javaOne Dark Pro

1 public final class BatchCommand1 {
2   private final UUID batchId;
3   private final long expectedVersion;
4   private final String submarketId;
5   private final int maxDetourMinutes;
6   public UUID batchId() { return batchId; }
7 }

pythonOne Dark Pro

1 @dataclass(frozen=True)
2 class BatchCommand1:
3     batch_id: str
4     expected_version: int
5     submarket_id: str
6     max_detour_minutes: int

typescriptOne Dark Pro

1 interface BatchCommand1 {
2   batchId: string;
3   expectedVersion: number;
4   submarketId: string;
5   maxDetourMinutes: number;
6 }

Why interviewers care

Delivery Batching interviews reward crisp scope, explicit trade-offs, and failure stories—not generic microservice diagrams.

Interview checkpoint

Name one failure story for Problem Statement: Multi-Order Delivery Batching that proves you understand real outages, not happy-path diagrams.

Key Highlights

•Anchor 1-A: batch command owns writes
•Anchor 1-B: submarket-scored pairs
•Anchor 1-C: detour + food degradation caps
•Anchor 1-D: honest ETA on mutation

1	public final class BatchCommand1 {
2	private final UUID batchId;
3	private final long expectedVersion;
4	private final String submarketId;
5	private final int maxDetourMinutes;
6	public UUID batchId() { return batchId; }
7	}

1	@dataclass(frozen=True)
2	class BatchCommand1:
3	batch_id: str
4	expected_version: int
5	submarket_id: str
6	max_detour_minutes: int

1	interface BatchCommand1 {
2	batchId: string;
3	expectedVersion: number;
4	submarketId: string;
5	maxDetourMinutes: number;
6	}