Manual point-in-time restore of a Cloud SQL for PostgreSQL instance. Each run restores the source instance into a new instance, <source>-restore-<random>, at restoreTime (or the latest recoverable time), checks it through the Cloud SQL API, and writes a JSON record of the run to a versioned evidence bucket the job can only add to. The restored instance is left running; each record holds the command to delete it.
The job holds Cloud SQL admin only until accessUntil, so it has no standing access to the source. Set it before each run.
Runs start only from Run Now on the job. GCP only; on other providers nothing is created.
This blueprint accepts 4 configurable inputs to customize the installation for your environment.
sourceInstancestringName of the Cloud SQL for PostgreSQL instance to restore, in this environment's project. Source it from a postgres installation's `instanceName` output.
checkDatabasesstringComma-separated databases that must exist on the restored instance. Empty skips the check.
restoreTimestringRFC 3339 time in UTC (ending in Z) to restore to, for example 2026-01-01T12:00:00Z. Empty means the latest recoverable time.
accessUntilstringRFC 3339 time in UTC (ending in Z) until which the job may start a restore, for example 2026-01-01T15:00:00Z. Until then it holds Cloud SQL admin on the source; empty means no admin access, and runs stop before restoring. Set it about 2 hours after the deploy that opens a run, since the job only needs access to start the restore, and clear it afterwards.
This blueprint exposes 1 output that other services can reference.
evidenceBucketNamestringBucket holding one JSON record per run under `runs/`.