[CASSANDRA-2397] Improve or remove replicate-on-write setting

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Won't Fix
• Severity: Normal
• Assignee: Unassigned
• Reporter: Stu Hood
• Created: 28/Mar/11 18:52
• Updated: 16/Apr/19 09:33
• Resolved: 16/Aug/12 23:11

Description

The replicate on write setting breaks assumptions in various places in the codebase dealing with whether data will be replicated in a timely fashion. It's worthwhile to discuss whether we should go all-the-way on replicate-on-write, such that it is a fully supported feature, or whether we should remove it entirely.

On one hand, ROW could be considered to be just another replication tunable like HH, RR and AES. On the other hand, a lazily replicating store is very rarely what you actually wanted.

Open issues related to ROW are linked, but additionally, we'd need to:

• Make the setting have an effect for standard column families
• Change the default for ROW to enabled and properly warn of the effects

Issue Links

• relates to: CASSANDRA-2389 Send writes to stable replicas when replicate-on-write is disabled during bootstrap
• relates to: CASSANDRA-2393 Ring changes should stream from the replicas that are losing responsibility