Skip to main content
The scale section of squid.yaml controls deployment placement and resources. It is available in Professional organizations. Resource changes take effect on redeployment and change billing immediately. Review the current profile prices and specifications instead of copying resource figures into application configuration.

Dedicated or collocated

With a scale section, dedicated defaults to true.
  • Dedicated services reserve their requested resources. Use them for production and workloads that need predictable performance.
  • Collocated services share capacity. They cost less, but performance is not guaranteed and they are not covered by the production SLA.
Playground organizations are collocated and do not accept a scale section.

Processor profile

Allowed profiles are small, medium, large, xlarge, and 2xlarge. If a manifest declares multiple named processors, the processor profile applies to those processor services. Each processor runs and is billed separately.
Increase the processor profile only after checking logs, indexing speed, and database load. A faster processor cannot fix a slow or incomplete upstream RPC endpoint.

API profile and replicas

Allowed API profiles are small, medium, large, xlarge, and 2xlarge. replicas accepts 1 through 5. Each replica is billed independently. Replicas improve request concurrency and service redundancy; they do not speed up the processor or database.

Postgres resources

Allowed Postgres profiles are small, medium, large, xlarge, and 2xlarge. Storage values are a number followed by G, Gi, T, or Ti. The default is 10Gi.
Increasing storage is not a substitute for controlling table growth. Monitor storage, remove data you no longer need, and tune update-heavy tables before they accumulate excessive bloat.
See Postgres for database versions, timeouts, connection limits, and direct access.

Hasura resources

Hasura supports small, medium, and large profiles, with 1 through 5 replicas.

Complete example

squid.yaml

Cost checklist

Before redeploying, count every running resource:
  • each slot is an independent deployment;
  • each named processor is a separate processor service;
  • each API replica is billed separately;
  • Postgres compute and storage are separate charges;
  • old and new slots both accrue charges during zero-downtime cutovers.
Use Billing and Usage to filter usage by squid, then compare it with the cost calculator.