Close the Upgrade, Not Just the Change Ticket
A change ticket should describe what happened to the service, not merely repeat a supplier announcement.
Google made MySQL 8.0.46 the Cloud SQL default for MySQL 8.0 on 10 October 2026. That is a managed-service change, not the upstream release date. Google’s current documentation, checked on 15 October, retains the following routes.
Which instances follow the default?
| Instance state | Upgrade route |
|---|---|
| New MySQL 8.0 instance, no minor specified | Default minor; automatic upgrades enabled. |
| Existing 8.0.35 or later, automatic upgrades on | Default minor during regular scheduled maintenance. |
| Existing 8.0.35 or later, automatic upgrades off | No automatic opt-in from this announcement. |
| Existing 8.0.34 or earlier | Manual minor upgrade first; enable automatic upgrades after reaching 8.0.35 or later. |
Automatic minor upgrades use regular scheduled maintenance, not self-service maintenance. A fleet-wide statement therefore cannot replace an instance-level decision.
What the maintenance window does not settle
Maintenance restarts instances and interrupts connectivity. Edition, workload, dataset, replicas and client settings affect impact. Enterprise Plus offers near-zero downtime only with prerequisites and constraints, not a universal guarantee.
Windows and rollout timing are configurable; email notices require opt-in. An ongoing instance operation cancels maintenance; disk use above 97 per cent skips it. Connection pools, timeouts and query behaviour influence recovery. Replicas need separate checks.
Those qualifications belong in the change record alongside region, edition, installed minor, upgrade setting, window, notices, capacity and overlapping work. A green planning approval should not silently become a green completion result.
Give the next owner an answer they can use
Consider a hypothetical payroll service whose overnight import succeeds but whose payment export does not. The database team could close its work while finance still has an unfinished pay run. That is a handover failure even if neither team’s individual status report is false.
BlackTree recommends agreeing the acceptance evidence before the change. Name the business operation, its expected output and the person who can accept it. For the payroll example, retain the import total, export total and reconciliation result together, with any exception assigned to an owner. A successful import alone would not authorise closing the export issue.
Test drivers, authentication, connection pools, representative queries, scheduled jobs and background workers, including restart recovery, in a representative environment. Afterwards, record the running minor, application health, replication, jobs and monitoring. Keep unresolved application exceptions separate from the database completion decision.
This makes the record useful to someone who was not present for the change. They should be able to distinguish the planned work, the observed result and the remaining obligation without reconstructing a conversation across several teams.
Sources
- Cloud SQL release notes, 10 October default change.
- Database version policies.
- Maintenance guidance.
Related: PostgreSQL 14 Has Different Support Deadlines in the Cloud.
