-
Type:
Bug
-
Resolution: Done
-
Priority:
Major - P3
-
Affects Version/s: None
-
Component/s: None
-
None
-
5
-
Storage - Ra 2020-06-29, Storage - Ra 2020-07-13, Storage - Ra 2020-07-27, Storage - Ra 2020-08-10
In rec_append_orig_value, we don't check aborted updates and we can return on aborted updates. This may have something to do with WT-6479 but unlikely.
/*
* Prepared updates should already be in the update list, add the original update to the
* list only when the prepared update is a tombstone.
*/
if (unpack->tw.prepare && upd->type != WT_UPDATE_TOMBSTONE)
return (0);
Also there are some assumptions no longer true after all the code changes we have, i.e, like
if (unpack->tw.start_ts == upd->start_ts && unpack->tw.start_txn == upd->txnid &&
upd->type != WT_UPDATE_TOMBSTONE)
The onpage time window may not have the same transaction ids and timestamp as on the update chain as we may write 0 to the time window.
So the following assert is also wrong:
WT_ASSERT(session, unpack->tw.prepare || (unpack->tw.stop_ts == oldest_upd->start_ts &&
unpack->tw.stop_txn == oldest_upd->txnid));