Why the Platform Is Fast, Reliable, and Secure

Stable

Which SPG99 infrastructure decisions are responsible for cold-start speed, storage reliability, and the safety of the new autoscale handoff.

Updated: March 21, 2026

The speed, reliability, and security of SPG99 come not from one setting, but from a combination of architectural decisions.

Why the platform is fast

Soft basebackup instead of a heavy local restore

Compute receives a thin startup image from Pageserver and starts from a minimal set of files.

Minimal pod disk

There is no unnecessary local user-data layer on compute. This speeds up startup and simplifies the pod lifecycle.

Local write-back cache

Hot work remains fast because compute still keeps a local working cache.

Gateway as a single entry point

The user connects to one stable PostgreSQL endpoint instead of manually finding the active writer.

Why the platform is reliable

Durable state is separated from one compute instance

Data does not depend only on the life of a particular worker.

WAL durability through Safekeeper

Confirmed writes rely on WAL quorum, not on the local disk of one compute instance.

Pageserver and the storage chain

The database can safely go to sleep and then start again without losing state.

The autoscaler uses generation handoff

The platform does not risk a live resize of the active writer. It prepares a new generation and switches traffic only at a safe moment.

Why the platform is secure

Compact public surface

Only the necessary user-facing domains and services are published externally.

TLS and managed certificates

Console and the API work over HTTPS, and PostgreSQL connections go through a TLS-enabled Gateway.

Separation of secrets and roles

The API key and PostgreSQL credentials are used for different tasks and are not mixed together.

Controlled cutover

Freeze/drain/pinned guards make handoff predictable and safe for production workloads.

Practical conclusion for the user

All of these decisions reduce to very clear benefits:

  • the database starts quickly;
  • data does not depend on the life of one compute instance;
  • the writer profile changes safely;
  • the user works through clear public domains and protected protocols.