The assertion that fires is the same as in WT-10934:

ASSERT(_icontext->_dyn_table_runtime[tint]._in_use > 0);

But this time it seems to come from a different root cause. When this occurs, the metadata file looks fine, there is no missing mirror. The reason might be the original thought I had when I created WT-10934:

The field _in_use, when greater than 0, indicates that a thread is currently using that table. This prevents other threads from dropping the table for instance.

Where the assertion is, since a thread is done working on the table, we want to make sure the field is greater than 0 before decrementing it.

There are two things we should do to investigate:

• Why is this only done for operations on random tables?
• We should investigate where the field _in_use is incremented/decremented to make sure we have not forgotten anything.