[CASSANDRA-5947] Sampling bug in metrics-core-2.0.3.jar used by Cassandra

• Type: Bug
• Status: Resolved
• Priority: Low
• Resolution: Fixed
• Fix Version/s: 1.2.10, 2.0.1
• Component/s: Legacy/Tools
• Severity: Low
• Assignee: Yuki Morishita
• Reporter: J.B. Langston

Authors

Yuki Morishita

Reviewers

Jonathan Ellis

• Created: 27/Aug/13 22:59
• Updated: 16/Apr/19 09:32
• Resolved: 19/Sep/13 05:59

Description

There is a sampling bug in the version of the metrics library we're using in Cassandra. See https://github.com/codahale/metrics/issues/421. ExponentiallyDecayingSample is used by the Timer's histogram that is used in stress tool, and according to brandon.williams it is also in a few other places like the dynamic snitch. The statistical theory involved in this bug goes over my head so i'm not sure if this would bug would meaningfully affect its usage by Cassandra. One of the comments on the bug mentions that it affects slow sampling rates (10 samples/min was the example given). We're currently distributing metrics-core-2.0.3.jar and according to the release nodes, this bug is fixed in 2.1.3: http://metrics.codahale.com/about/release-notes/#v2-1-3-aug-06-2012