[CASSANDRA-14635] Support table level configuration of monotonic reads

• Type: Improvement
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 4.0-alpha1, 4.0
• Component/s: Legacy/Coordination
• Assignee: Blake Eggleston
• Reporter: Ariel Weisberg

Authors

Blake Eggleston

Reviewers

Aleksey Yeschenko

• Created: 09/Aug/18 20:42
• Updated: 15/May/20 08:01
• Resolved: 24/Aug/18 16:53

Description

In CASSANDRA-10726 it was discussed that allowing expert users to forgo monotonic reads might be desirable. It is practical to control monotonicity of reads at a fine grained level because it involves changing the behavior of read repair on a per read basis.

Per CASSANDRA-14593 we already don't preserve update atomicity down to the column level. You could read the key out of a row and read repair the key, pass the key to another process which attempts to read the value, but finds the value is null because read repair only repairs the data (including columns) that is part of the read. IMO it's a stretch to say that reads are monotonic. It is technically correct, the best kind of correct, but far from as useful as it should be.

An initial implementation could make read repair asynchronous or forgo read repair entirely. This would improve the throughput and latency of reads.

Issue Links

• is related to: CASSANDRA-15442 Read repair implicitly increases read timeout value
• relates to: CASSANDRA-14593 Read-Repair breaks partition-level update atomicity
• relates to: CASSANDRA-10726 Read repair inserts should not be blocking