A Silent Hang That Could Derail Your 26ai Upgrade Plans
If you’re planning or actively executing an upgrade to Oracle AI Database 26ai, stop and read this before you proceed. Mike Dietrich, Oracle’s Vice President of Product Management for Database Upgrade; and the undisputed authority on all things Oracle upgrade; disclosed on July 15, 2026, that an internal change in 26ai can cause upgrades to hang indefinitely due to block checking events. The issue was uncovered during two very large customer upgrades and has prompted Oracle Support to publish a dedicated knowledge base article with workarounds and a permanent fix.
This is not a theoretical edge case. It’s a real-world production issue that has already bitten enterprise customers, and given that AutoUpgrade is the only supported upgrade method for 26ai, virtually every organization upgrading to the latest release could be at risk.
What Exactly Happens?
The root cause lies in an internal change introduced in Oracle AI Database 26ai related to block checking behavior. Under specific conditions, the upgrade process, or the POSTFIXUPS phase that follows it, can stall completely. The database doesn’t crash. It doesn’t throw a dramatic error. It simply hangs silently, leaving DBAs staring at an unresponsive upgrade session with no obvious indication of what went wrong.
According to Dietrich’s disclosure, the hang is triggered by block checking events that interact poorly with the upgrade catalog scripts. The issue is particularly insidious because it doesn’t always manifest immediately. In some cases, the upgrade appears to progress normally through early phases before grinding to a halt during later operations, making it difficult to diagnose without prior awareness of the bug.
Who Is Affected?
The bug does not affect every upgrade; but the conditions that trigger it are far from uncommon in enterprise environments. Specifically, the issue affects databases that underwent any of the following operations prior to the upgrade:
- Database cloning: a standard practice for creating upgrade test environments or pre-production copies
- PDB refreshable clones: widely used in multitenant architectures for maintaining synchronized copies of production pluggable databases
- Database copy or restore operations: including RMAN-based restores to new hosts or standby conversions used as upgrade targets
If any of these operations were performed on the database before initiating the 26ai upgrade, the block checking event issue can be triggered. This is a critical detail because many organizations follow best-practice upgrade methodologies that involve cloning production databases to test the upgrade process first; the very workflow that can activate this bug.
The two customer cases where the issue was discovered involved very large databases, suggesting that scale may be an amplifying factor. However, Oracle has not confirmed that smaller databases are immune, so caution is warranted regardless of database size.
The Fix and Workarounds
Oracle Support has responded promptly with MOS Note KB914929, which documents the full pattern, diagnostic steps, and available workarounds. DBAs with active Oracle Support contracts should review this note immediately if they are planning or conducting 26ai upgrades.
The permanent fix is included in Release Update 23.26.3, which is part of the July 2026 quarterly release update cycle. For organizations that cannot wait for or immediately apply the RU, the MOS note provides interim workarounds that can be applied to prevent the hang from occurring during upgrade execution.
Dietrich’s public disclosure serves as both an alert and a reminder of the value of the Oracle upgrade community’s transparency. By surfacing the issue quickly, along with actionable guidance, Oracle has given DBAs the information they need to avoid what could otherwise be hours or even days of unexplained downtime during critical upgrade windows.
Why This Matters for Your Upgrade Strategy
The implications of this bug extend beyond the immediate technical fix. Consider the operational context: upgrade windows for large enterprise databases are meticulously planned, often negotiated weeks in advance with business stakeholders, and carry significant pressure to complete within tight timeframes. A silent hang during upgrade or POSTFIXUPS can blow through those windows entirely, triggering rollback procedures, missed SLAs, and uncomfortable conversations with leadership.
Moreover, because AutoUpgrade is the sole supported upgrade path to 26ai, there is no alternative method that sidesteps this issue. Every upgrading customer flows through the same pipeline; making awareness of this bug universally relevant.
Practical Takeaway for Oracle Professionals
Here’s what you should do right now:
- Review MOS Note KB914929 in full; understand the diagnostic indicators and workaround steps before your next upgrade attempt
- Audit your pre-upgrade workflow; determine whether your upgrade target databases were created via cloning, restore, or PDB refreshable clone operations
- Apply RU 23.26.3 to your 26ai target environment if possible; this is the cleanest path to eliminating the risk entirely
- If you’re mid-upgrade and experiencing an unexplained hang, check for block checking event waits as described in the MOS note before assuming other root causes
- Update your upgrade runbooks to include a pre-check for this condition as a standard step
Oracle upgrades are complex enough without silent bugs lurking in the process. Thanks to Mike Dietrich’s timely disclosure and Oracle Support’s rapid documentation, the community has the tools to navigate this one safely. But only if you take the time to read the advisory before you hit the upgrade button.