Aegis Gateway 2.0

PostgreSQL shared-state validation

Scope and classification

On 2026-10-08, PostgresStateStoreIT and TwoGatewayProcessesPostgresIT were reported run against a short-lived, dedicated managed PostgreSQL test instance. The engine version was not retained, so this record does not name one. The first suite opened independent PostgresStateStore objects. The second started two gateway JVMs against the same database. No container, mock driver, or in-memory database was used.

The result is historical recorded-unattested evidence. No immutable raw Maven/Surefire artifact, commit/tree manifest, or content-addressed run bundle was retained. The temporary managed PostgreSQL instance was deleted after the run, so this document must not be promoted to verified-current without a new, candidate-bound run.

Commands

The state-store module command was:

export AEGIS_POSTGRES_URL='jdbc:postgresql://<endpoint>:5432/<database>?sslmode=require'
export AEGIS_POSTGRES_USER='<user>'
export AEGIS_POSTGRES_PASSWORD='<password>'
mvn -B -Ppostgres-integration -pl aegis-state-postgres -am verify

That command produced the reported PostgresStateStoreIT result only. The gateway integration test was selected separately from the gateway module:

export AEGIS_POSTGRES_URL='jdbc:postgresql://<endpoint>:5432/<database>?sslmode=require'
export AEGIS_POSTGRES_USER='<user>'
export AEGIS_POSTGRES_PASSWORD='<password>'
mvn -B -Ppostgres-integration -pl aegis-gateway -am install \
  -Dtest=TwoGatewayProcessesPostgresIT \
  -Dsurefire.failIfNoSpecifiedTests=false \
  -Djacoco.skip=true

The placeholders are intentional. This evidence does not retain or disclose an AWS account, endpoint, secret, public IP, or resource identifier.

Reported result

PostgresStateStoreIT reported 5/5 integration tests passed, and TwoGatewayProcessesPostgresIT reported 1/1 passed, with zero failures and zero skips. These are reported outcomes, not candidate attestation. The run covered:

  1. Two independent stores admitted one nonce winner and no more than the shared rate-window capacity.
  2. Values persisted across connections; expired values disappeared; a rejected cost did not consume rate capacity.
  3. Escaped prefix scanning, bounded scans, and invalid rate requests behaved correctly against PostgreSQL.
  4. The factory resolved credential environment-variable names rather than placing credential values in configuration.
  5. The signed serial high-water mark advanced from 9 to 10 and then refused serial 9 across independent PostgreSQL connections.
  6. Two independent gateway JVMs shared approval and receipt single-use state and rate windows.

Test environment

A dedicated, network-restricted test instance holding only generated test tables and namespace rows was created for this run and deleted after it. It held no production or customer data. The connection required PostgreSQL TLS through sslmode=require.

The run establishes only the reported state-store and two-process behavior. A production deployment should use encrypted storage, private network placement, least-privilege roles, managed credential handling, a backup and recovery policy, and a repeat of these tests against the release candidate.

Evidence boundary

This record does not establish PostgreSQL replication, failover, backup restoration, multi-region behavior, catalog input sharing, circuit-breaker sharing, or operational security of a production database deployment. Relevant source and tests have changed since the run, so current behavior requires a new candidate-bound execution and retained evidence bundle.