The investigation of WT-10545 led to some results that need to be investigated further.

Context:

When a bulk cursor closes, the dhandle associated with the file is closed and a checkpoint on the file is performed. This is not a system-wide checkpoint but a single-file one.

If this is followed by a crash, the data inserted by the bulk operations is not present after a restart despite the single-file checkpoint.

Reproducer:

# Make sure to have the reproducer attached to the JIRA ticket test_bulk03.py in test/suite/
python3 ../test/suite/run.py -p bulk03

# Dump the content of the clean shutdown
./wt -h WT_TEST/test_bulk.0/ dump file:bulk03.wt

# The output shows the following lines at the end
# a\00
# aaa\00

# Dump the content of the unclean shutdown
./wt -h WT_TEST/test_bulk.0/RESTART dump file:bulk03.wt

# The output shows no data.

Goals:

Understand if this is normal behaviour
Understand where the data has gone
Check if the underlying blocks used by the bulk data are freed after restart, i.e confirm that we are freeing the space used by the bulk operations after crash and restart

• Create a test that verifies the expected behaviour
• Update the documentation if required