analyze commands allows us to collect statistics on existing tables/partitions. It works great but might be slow since it scans all files.
There are 2 ways to speed it up:
1. collect stats without file scan. It may not collect all stats but good and fast enough for use case.
HIVE-3917 addresses it
2. collect stats via partial file scan. It doesn't scan all content of files but part of it to get file metadata. some examples are https://cwiki.apache.org/Hive/rcfilecat.html for RCFile, ORC (
HIVE-3874 ) and HFile of Hbase (Edit: That link should be https://cwiki.apache.org/confluence/display/Hive/RCFileCat.)
This jira is targeted to address the #2. More specifically RCFile format.