Secondary index is used for numeric filtering. It is composed of two parts:

1. The data of each numeric column is stored in an inverted sorted form, with a compressed format.
2. Another in-memory part of index data which is based on pgm, it could provide very fast approximate range query with bounded error, which has already been added into the repository.

The mechanism of range filtering of secondary index is as follows:

1. Query the pgm index to get the bounded range.
2. Scan the raw index data according to bounded range to get the RowIDs of the query filter.