I am wondering if this value an aggregated value in some time interval or if result of all queries are stored?
I am using and modifying rally-eventdata-track and in particular combined-indexing-and-querying and want to measure highest latency over time (95+ percentiles).
as @Christian_Dahlqvist has answered, the samples in rally-metrics-* are raw samples (one entry per request/response). Throughput samples are aggregated once per second.
If you are after specific percentiles, Rally stores the summary report per benchmark in rally-results-* (the one that you see on the command line at then end of a race). I.e. you may have thousands of (raw) metrics documents in rally-metrics-* but only one summary document in rally-results-*. If you want to calculate specific percentiles that are not pre-calculated by Rally (e.g. 95th percentile) I suggest that you use the raw metrics to do that.
My problem was not understanding how rally works. Now I have found target-interval setting so I can properly decrease this interval to match our use-case better. Currently running a race...
@danielmitterdorfer - Have you considered having the option to dumping metrics to Elasticsearch during the race so it is possible to follow the progress? This would be great for us.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.