Scalable

Saleor was designed as a hyperscale commerce platform. You can run it on your own hardware, but it thrives in the cloud, where it can take advantage of dynamic scaling and elasticity.

Horizontal Scaling

Where vertical scaling means adding more resources to a single powerful machine, horizontal scaling means adding more machines to the pool of resources. Saleor's engine was built to allow for massive parallelism with many small, stateless containers. The entire state, including locking mechanisms, is handled by a shared database layer.

This is an excellent match for container orchestration platforms like Kubernetes. The resulting architecture can scale dynamically to suit demand with minute resolution, removing the need to commit to Black Friday-level capacity throughout the entire year or quarter. It allows upgrades to be rolled out without interrupting services and lets you distribute the service among multiple availability zones for increased resilience.

Of course, for stateless workers to scale well, the persistence layers must be able to keep up. All file storage is delegated to cloud-native solutions like AWS S3 and GCP Cloud Storage.

The database of our choice, PostgreSQL, does not offer a direct horizontal scaling solution, but Saleor is following the best practices of PostgreSQL's high availability by separating database writes from pure reads and using read replicas for the latter. Managed services like Amazon Aurora can take this further by automating writer failovers and allowing the read replica pool to scale automatically.

Even if you intend to run only a single instance of Saleor, all of the above means you can use patterns like blue/green deployment to avoid downtime when moving to a more powerful host.

Buffered Outputs

A big challenge in scaling a composable architecture is avoiding the situation where all systems must scale up in response to a single system experiencing a spike in traffic. This can not only get very costly but may be outright impossible, as some enterprise-grade ERP systems can only scale to a few requests per second, while a big sale can result in thousands of orders being placed every second. For this reason, Saleor divides events into two groups.

Events required by the delegation APIs are sent synchronously, requiring integrations to keep up. If an app breaks down and is unable to respond, an appropriate fallback mechanism is used to prevent the failures from cascading and causing an outage. Repeated failures will cause a circuit breaker to temporarily disable the integration entirely to prevent further resource loss on repeated reattempts and give the app some breathing room to recover.

All the other events are buffered and delivered asynchronously, allowing external applications to consume them at the rate they can comfortably keep up. This may result in temporary backlogs forming and some delays in processing, but it means a sudden scale-up of Saleor won't overwhelm and wipe all of its installed extensions.

Handling Large Datasets

We've personally built stores that offered millions of SKUs and ones that handled thousands of orders per second.

While there are cases where the user interface is not up to our standard, like managing hundreds of warehouses in the dashboard, or where we occasionally discover an SQL query that needs to be optimized, the entire product was built to tackle and gracefully handle copious amounts of data.

It's not data lake-scale, but you don't need a data lake for your storefront, and we have yet to see a case in which our architects cannot recommend a reasonable solution.

Scaling for the Future

Saleor is a living project that is constantly evolving. Its composability and use of GraphQL make it a good fit for today's needs, including native mobile storefronts, point-of-sale kiosks, and virtual- or mixed-reality experiences. It's also ready for some of the tomorrow's tech, like ubiquitous AI agents speaking cryptic MCP languages.

This project is never "done," and we do our best to ensure that no matter where the future takes you, Saleor will not be holding you back.