On 4/28/16, Richard Hipp <drh at sqlite.org> wrote: > On 4/28/16, Dominique Devienne <ddevienne at gmail.com> wrote: >> >> Richard, is STAT3 or STAT4 required for >> https://www.sqlite.org/optoverview.html#skipscan ? >> Or essential for skip-scan to bring the most benefit? --DD > > No. But sqlite_stat1 data is. Skip-scan is only attempted if there > are an average of 18 or more repeats of the skipped column(s) - a > number we came up with by doing actual queries to see where the > pay-off point started. Without sqlite_stat1 data, SQLite will assume > that there are only 10 repeats of each entry, at most, so without > sqlite_stat1 data, skip-scan will never be used.
I should add: The above is for the current implementation and is subject to change if we discover a better way of working things. -- D. Richard Hipp drh at sqlite.org
