[CASSANDRA-3750] Migrations and Schema CFs use disk space proportional to the square of the number of CFs

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Duplicate

Environment

Linux (CentOS 5.7)

• Severity: Normal
• Assignee: Unassigned
• Reporter: John Chakerian
• Created: 18/Jan/12 05:08
• Updated: 16/Apr/19 09:32
• Resolved: 18/Jan/12 08:47

Description

The system keyspace grows proportional to the square of the number of CFs (more likely, it grows quadratically with # of schema changes in general). The major offenders in the keyspace are the Migrations table & the Schema table. On clusters with very large #s of CFs (in the low thousands), we think that these large system tables may be contributing to various performance issues.

The approximate expression is: s = 0.0003253*n^2 + 2.58, where n is # of keyspaces + # of schemas and s is the size of the system keyspace in megabytes. See attached plot of the regression curve showing fit.

Sampled data:

NUM_CFS SYSTEM_SIZE_IN_MB
100 4.4
200 15
300 32
400 55
500 85
600 120
700 162
800 211
900 266
1000 327

This was hit in 1.0.1, but is almost certainly not version specific.

Issue Links

• is superceded by: CASSANDRA-1391 Allow Concurrent Schema Migrations