---
title: "Reserving and releasing order capacity with an event stream"
language: en
topics: ["data", "ecommerce", "lemmid-store", "payments", "server", "work"]
full_article_url: https://willem.com/en/2020-12-02_designing-a-multi-dimensional-capacity-queue/
---
# Reserving and releasing order capacity with an event stream

*Immutable events and negative orders instead of counters.*

> Store every order as an immutable timestamped event and reserve capacity the moment the order is placed. If payment fails or the order is abandoned, append a negative order with the same dispatch time to release the capacity.

I built this for my food ordering app, where three capacity dimensions interact: kitchen, pickup and delivery, each with its own timespan granularity, say 10 dishes per 15 minutes against 3 deliveries per 30 minutes.
The design follows the Lambda Architecture: an append-only stream of immutable, timestamped events processed by a realtime speed layer and a batch precomputation layer. Every order is an event; cancelling an order of three pizzas appends a negative order of minus three pizzas with the same dispatch timestamp, which releases the reserved capacity without ever mutating a record.
Metrics are computed from the real stream data in RAM. The first query initialises a model from historic data in batch, realtime updates keep it current while it is in demand, and answers arrive within milliseconds.