- Changes to the timestamp_abort test to run with columns.
(Use -c for columns.)
- Add a rollback_to_stable test that fails on today's RLE-counting bug.
- Fix RLE counting bug in column rollback to stable.
- New Python tests exercising rollback-to-stable vs. column-store RLE cells.
- __rollback_ondisk_fixup_key: avoid error-path free twice
- Add and enable column-store support in test_truncated_log.
- Changes to the wt6616_checkpoint_oldest_ts test to run with both columns and rows.
- Changes to the wt6185_modify_ts test to run with columns as well as rows.
- Changes to the schema_abort test to allow running with columns.
- Re-enable test_wt2246_col_append.
(It was disabled in
WT-5790 because it was failing, but it works now.)
- Changes from initial review.
Also, while crosschecking options lists it turned out that I'd
forgotten to actually add the columns option handling to wt6185,
probably because there was already a 'c' case in the switch, so do
that correctly and use -C. "oops"
- Fix a missed sessionconf -> tableconf.
- test_rollback_to_stable25: Clarify the code that tracks the expected values.
- Remove debug bits accidentally left in.
- Remove rollback_to_stable26.
It looks like the reality of integration testing and the available
test machines is that it won't work reliably, and probably can't be
It isn't worthwhile making a C test for this; the point was to have
something quick in the Python suite. If the behavior regresses, at
least timestamp_abort and I think others in the C suite will start
running forever; this won't necessarily cause overt failures, but
because they're part of make check someone will notice.
- Improve comment. | 20 Aug 21 01:49 UTC
Evergreen Subscription: ; Evergreen Event: