In the current read repair video on the Academy, the read repair occurs before data is returned to the client only with CL of ALL, in the 3.0 book it says with CL of Quorum or All. I am learning and I want to verify that there was a change in versions(?) and the read repair occurs before data is returned to the client with a CL of All. I am taking the certification test soon. Thank you!
No, it hasn't changed.
There are two types of read repairs: foreground read repairs (blocking) and background read repairs.
If any replica involved in read requests with consistency higher than ONE or LOCAL_ONE is out-of-date or inconsistent, the replica is repaired in the foreground before the results are returned to the client. This type of read repair is a blocking operation since the coordinator waits for the repair to complete.
A read repair triggered by read_repair_chance or dclocal_read_repair_chance, the coordinator returns the result to the client and issues a read repair in the background to repair inconsistent replicas.
Note that read repair chance was removed in C* 3.0.17, 3.11.3 and 4.0 (CASSANDRA-13910).
For more info, see Read repair in Apache Cassandra. Cheers!
5 People are following this question.
