Upgrade to 26ai may hang due to block checking events

This issue is caused by an internal change, and we hit this now with two very large customer upgrades recently. Therefore, I would like to point to a fix, a workaround and the following RU in case you see your Upgrade to 26ai may hang due to block checking events.

Upgrade to 26ai may hang due to block checking events

Photo by Nareeta Martin on Unsplash

What would you see?

This issue can happen with cloning, PDB refreshable clones or just a copy or restore of your database for the subsequent upgrade to Oracle AI Database 26ai. It is not dependent on AutoUpgrade but since AutoUpgrade is the only supported method to upgrade your databases to Oracle 26ai, you will see the effect obviously when you operate AutoUpgrade.

You will see the upgrade process hanging. This can happen during the upgrade but in another customer case it happened in the POSTFIXUPS phase.

A typical pattern could look like this:

upg> status -job 102
Details
..
Stages
      SETUP            <1 min
      GRP              <1 min
      PREUPGRADE       <1 min
      PRECHECKS        <1 min
      PREFIXUPS        <1 min
      DRAIN            <1 min
      DBUPGRADE        ~1190 min (RUNNING)
      POSTCHECKS
      POSTFIXUPS
      POSTUPGRADE
      SYSUPDATES
.
Stage-Progress Per Container
.
      +--------+---------+
      |Database|DBUPGRADE|
      +--------+---------+
      |CDB$ROOT|    100% |
      |PDB$SEED|    36 % |
      |PDB     |    0  % |
      +--------+---------+

In another customer case, it did happen during gathering index stats:

WAIT-state since 13:34 with event "cell single block physical read: xrmem cache"

In the dbwriter traces you would find the function kdxd4bu on top of the stack trace indicating the long wait.

 

What could be the reason?

The reason for this hang could be a change in the maximum INITRANS for index blocks since Oracle Database 23ai. The maximum value was 255 before but has been changed to 169. Even if you request 255 in a CREATE statement you would only get 169 ITLs (Interested Transaction List) in the index block header.

Then, we introduced an enhancement in 23.26.0 for better block checking – which assumed and checks for 169 ITLs at maximum.

In an upgrade case, this situation could or can lead to a hang unfortunately.

 

What is the solution?

There are three options (or technically four) to solve this:

  • Apply the fix for Bug 38946554 – HANG ON BLOCKCHECK DUE OVERFLOW ON MAX ITL COUNT
  • Move to Oracle AI Database 26ai Release Update 23.26.3 (July 2026) or newer once it is available
  • Use these underscores as a temporary workaround:
    db_block_checking = FALSE 
    _db_always_check_system_ts = FALSE
  • Rebuild the index(es) affected – but this is a theoretical solution only when it happens during a database upgrade

 

Is this documented?

Luckily, Oracle Support has released a MOS Note already describing the whole pattern:

 You find all the details in KB914929. And this blog post is only meant to unblock you in case you run into such an issue.

 

–Mike