squirreldb-cassandra-index-bench2

Documentation

Overview

This benchmark is mostly used to test shard transition issue: * Our postings are shared by time, with each shard being 7 days long. * When a new shard beging, we need to have all postings for active metrics created for this new shard. * Which means that with naive implementation, when new shard start, lots of write are needed

This benchmark will simulate multiple servers (bench.scale) and do LookupIDs (the request done during metric ingestion) for metrics of each similated servers. It does that at a fake clock time which is close to the new shard time. It will then show the time required for LookupIDs (with average, 90th quantile...) each 10 minutes near shard change, to show how much request time increase when shard change.