Details
Description
When we replay inProgress, we need to refresh it with the newly prepared ballot, or it will be (correctly) rejected.
Issue Links
- is duplicated by
-
CASSANDRA-6029
Lightweight transactions race render primary key useless
-
- Resolved
-
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
One "nit": I'm not sure sleeping in the case where the propose work (and we commit) is really useful, the commit itself probably play a good enough sleep if we're contending (which, for what its worth, is confirmed by my unscientific test: moving the sleep only if the propose fails perform slightly better without making threads retry more often).
But overall, +1 on the fix (a quick test (that I'll commit to dtests) does confirm the current CAS timeout as soon a 2 inserts contend a bit too much).
Show
Sylvain Lebresne
added a comment - One "nit": I'm not sure sleeping in the case where the propose work (and we commit) is really useful, the commit itself probably play a good enough sleep if we're contending (which, for what its worth, is confirmed by my unscientific test: moving the sleep only if the propose fails perform slightly better without making threads retry more often).
But overall, +1 on the fix (a quick test (that I'll commit to dtests) does confirm the current CAS timeout as soon a 2 inserts contend a bit too much).
committed
Show
Jonathan Ellis
added a comment - committed
Fix pushed to https://github.com/jbellis/cassandra/tree/5985, along with a patch to clean up the trace messages. Also added a sleep to the collision-in-replay branch to match the collision-in-prepare.