mirror of https://github.com/apache/cassandra
a5ce963117
In ttl-heavy use cases (especially tables with default time to live set), the streaming histograms calculated during compaction and streaming are very inefficient. This patch addresses that in two ways: 1) It creates a system property -Dcassandra.streaminghistogram.roundseconds=60, and rounds all histograms to the next highest multiple of that value, and 2) Rather than maintaining a histogram of 100 bins that have to be merged on every new value, we keep a temporary spool of 100000 bins, and merge down to the 100 bin final histogram only when the temporary spool overflows. Patch by Jeff Jirsa; Reviewed by Nate McCall for CASSANDRA-13038 |
||
|---|---|---|
| .. | ||
| burn/org/apache/cassandra | ||
| conf | ||
| data | ||
| long/org/apache/cassandra | ||
| microbench/org/apache/cassandra/test/microbench | ||
| resources | ||
| unit/org/apache/cassandra | ||