Hi,
This regression can happen when:
- Upgrading from any 2.1 version (till 2.1.15) to latest 2.1.20
- Upgrading from any 2.2 version (till 2.2.7) to latest 2.2.12
- And having a descending clustering key
- And using CAS
Note that once SSTables are upgraded, the problem disappears.
A ticket
Hi,
I'm the reporter for CASSANDRA-11349.
Just a bit of history:
This was a big pain point for us because the over-streaming was really
important.
Each week, repair increased data size by a few percents.
To give an example, one cluster with 12 nodes had a CF with around 250GB of
data per node and