[CASSANDRA-1332] Scan results out of order

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 0.7 beta 1

Environment

CentOS 5, Java 1.6.0, Cassandra trunk as of 28 July 2010

• Severity: Normal
• Assignee: Jonathan Ellis
• Reporter: Dave Revell

Authors

Jonathan Ellis

• Created: 28/Jul/10 22:43
• Updated: 16/Apr/19 09:33
• Resolved: 06/Aug/10 00:50

Description

After inserting 10 keys ('0', '1', ... '9') and running scan() with start_key='' and count=7, scan() returns the keys ['7', '3', '6', '5', '0', '8', '2']. When I scan() again with start_key='2' and count=7, I get the keys ['2', '1', '9', '4', '7']. Notice that key "7" appears in both result sets, and the relative order of keys "7" and "2" is inconsistent between the two scan results.

I see the problem when running on a 4-node cluster. When I run on a 1-node cluster, the problem does not occur. So the attached system test always passes, since system tests use a 1-node cluster, so the test doesn't actually demonstrate the problem.

A standalone Python program that reproduces the problem is at: http://pastebin.com/FwitG4wf