Closed lhofhansl closed 10 years ago
Note that HBASE-10117 will make the call to isFilterDone() in nextRaw() unsynchronized. Together these two changes will yield a 20-30% improvement in COUNT(1) type queries in tall tables.
Wow, awesome!!! Thanks, @lhofhansl.
Hey James,
just worked on https://issues.apache.org/jira/browse/HBASE-10117 and noticed that Phoenix unnecessarily calls RegionScaner.isFilterDone() in some (hot) places. isFilterDone() is synchronized and hence causes a noticeable slow down. nextRaw() internally calls isFilterDone and only returns true if isFilterDone() returns false, so the extra call is unnecessary,