[CASSANDRA-13489] Cassandra Repair in 2.2.8

• Type: Task
• Status: Resolved
• Priority: Urgent
• Resolution: Feedback Received

Environment

Linux Redhat Enterprise, 32 GB RAM,

• Assignee: Unassigned
• Reporter: Shoban
• Created: 03/May/17 14:13
• Updated: 16/Apr/19 09:30
• Resolved: 04/May/17 10:31

Description

I am using 4 node, 2 data center Cassandra cluster. While running nodetool repair -dcpar it takes arund 2 hours 30 minutes for database size of 20 MB. I checked how to tune streaming of data between data centers from below url:

https://support.datastax.com/hc/en-us/articles/205409646-How-to-performance-tune-data-streaming-activities-like-repair-and-bootstrap

But still the repair takes 2 hours and 30 mins. I drilled down the repair logs and identified while repair Cassandra repairs 256 ranges per node which is 4*256=1024. In a single token range merkle tree for each column families is compared which takes around 110 ms. We have 80 column families thus it takes 110*80*1024 which results in 2 hours 30 mins.

Can we reduce number of traffic by generating merkle tree for more than one column family at a time?

Or is there any other way to reduce the repair procedure in Cassandra 2.2.