Upgrade Planning Post Upgrade Detail

Statistics refresh (CRITICAL)

A major-version upgrade does NOT recompute table statistics, and stale statistics against the new optimizer are a top cause of post-upgrade plan regressions. You MUST refresh statistics before trusting query plans post-upgrade:

• Run ANALYZE TABLE on hot tables to rebuild index/column statistics for the new optimizer.

  sqlCopy

  ANALYZE TABLE critical_table_1, critical_table_2;

• OPTIMIZE TABLE is more aggressive (it rebuilds the table and reclaims fragmentation) but takes a table lock — avoid on production unless you specifically need to defragment, and run it in a maintenance window.

Plugin / component checks (Aurora MySQL)

Aurora MySQL has no PostgreSQL-style extension catalog to update post-upgrade, but verify any server components or audit plugins are still loaded and compatible with the new major version:

If you used mysql_native_password auth, confirm application drivers work with the 8.0 default caching_sha2_password (or that the plugin is still available).

Query plan verification

Optimiser changes across major versions are one of the top causes of post-upgrade regression. For each critical query that was in the “hot queries” set before the upgrade, capture a fresh plan and compare:

Use EXPLAIN FORMAT=JSON (estimated plan) or EXPLAIN ANALYZE (MySQL 8.0+; runs the query and reports actual vs estimated row counts and timing):

Look for:

• Different join ordering or join algorithm (nested-loop ↔ hash join; MySQL 8.0.18+ adds hash joins).
• Index usage changes (previously used index now not chosen).
• Large gaps between estimated and actual rows (selectivity estimates degraded — refresh stats with ANALYZE TABLE).

Aurora parameter group family migration (commonly missed)

Aurora cluster parameter groups are pinned to a specific major version family (e.g. aurora-mysql5.7, aurora-mysql8.0, aurora-mysql8.4). The upgrade process creates a new parameter group in the target family OR requires you to assign one — you cannot reuse an aurora-mysql5.7 parameter group on an 8.0 cluster.

Verify the cluster is actually using a target-family parameter group:

Risk: if the pre-upgrade cluster had custom parameters (e.g. innodb_buffer_pool_size, max_connections, innodb_log_file_size, custom logging settings), those MUST be re-applied to the new-family parameter group — they are NOT carried across automatically. Mis-applied parameter groups are the second-largest source of post-upgrade regressions after optimiser changes.

Pre-upgrade snapshot — rollback window (CRITICAL)

If you took a pre-upgrade manual snapshot (you should have — it’s a pre-upgrade-checklist item), note that:

• A snapshot taken on the old major version restores to the old major version — NOT to the post-upgrade new version. A 3.04 snapshot → 3.04 restore. You cannot “upgrade by restoring from a snapshot.”
• This snapshot is your rollback path for the first 7–14 days post-upgrade. Do NOT delete it until you’ve confirmed the new version is stable under full production load. 7–14 days of stable production is the industry norm; longer for regulated workloads.
• If you need to rollback, you restore the pre-upgrade snapshot into a new cluster, then cut traffic back over. There is no in-place downgrade.

Monitoring window (24–72 hours)

Watch these CloudWatch metrics for the first 24–72 hours and compare to pre-upgrade baselines:

• CPUUtilization — a 5–15% change is normal; > 25% indicates a plan regression.
• DatabaseConnections — should be stable; sudden rise can mean connection-pool re-auth loops on engine changes.
• ReadLatency, WriteLatency, DMLLatency, SelectLatency — p95 should return to baseline within 2 hours; sustained elevation indicates query-plan issues.
• FreeableMemory — especially important if the cluster uses a custom innodb_buffer_pool_size; freezing at a different level indicates a parameter-group-family migration issue.
• BufferCacheHitRatio (the Aurora MySQL buffer-pool hit ratio) — should be ≥95% for OLTP; drop below 90% means statistics or cache warmup issue.
• AuroraReplicaLag, AuroraReplicaLagMaximum — as above, should settle below 100 ms for readers.
• Deadlocks, LoginFailures — both should be near pre-upgrade baseline; a spike can signal an authentication-plugin change (caching_sha2_password default in MySQL 8.0) or a reserved-word conflict.

What NOT to do post-upgrade

• Do NOT suggest a rollback as the first response to a minor issue. The rollback path destroys the timeline and takes significant effort. Debug regressions on the new version first.
• Do NOT delete the pre-upgrade snapshot in the first week. That is the rollback path.
• Do NOT run modify-db-cluster --engine-version to downgrade — downgrades are not supported in-place. (Downgrades are not supported in-place by Aurora.)
• Do NOT skip the ANALYZE TABLE pass on hot tables even if the optimizer seems to be running fine. A one-time post-upgrade manual statistics refresh is insurance.