The engine inherits from [MergeTree](mergetree.md#table_engines-mergetree). The difference is that when merging data parts for `SummingMergeTree` tables ClickHouse replaces all the rows with the same primary key (or more accurately, with the same [sorting key](mergetree.md)) with one row which contains summarized values for the columns with the numeric data type. If the sorting key is composed in a way that a single key value corresponds to large number of rows, this significantly reduces storage volume and speeds up data selection.
We recommend to use the engine together with `MergeTree`. Store complete data in `MergeTree` table, and use `SummingMergeTree` for aggregated data storing, for example, when preparing reports. Such an approach will prevent you from losing valuable data due to an incorrectly composed primary key.
ClickHouse may sum all the rows not completely ([see below](#data-processing)), so we use an aggregate function `sum` and `GROUP BY` clause in the query.
When data are inserted into a table, they are saved as-is. Clickhouse merges the inserted parts of data periodically and this is when rows with the same primary key are summed and replaced with one for each resulting part of data.
ClickHouse can merge the data parts so that different resulting parts of data cat consist rows with the same primary key, i.e. the summation will be incomplete. Therefore (`SELECT`) an aggregate function [sum()](../../query_language/agg_functions/reference.md#agg_function-sum) and `GROUP BY` clause should be used in a query as described in the example above.
For columns of [AggregateFunction type](../../data_types/nested_data_structures/aggregatefunction.md) ClickHouse behaves as [AggregatingMergeTree](aggregatingmergetree.md) engine aggregating according to the function.
Table can have nested data structures that are processed in a special way.
If the name of a nested table ends with `Map` and it contains at least two columns that meet the following criteria:
- the first column is numeric `(*Int*, Date, DateTime)`, let's call it `key`,
- the other columns are arithmetic `(*Int*, Float32/64)`, let's call it `(values...)`,
then this nested table is interpreted as a mapping of `key => (values...)`, and when merging its rows, the elements of two data sets are merged by `key` with a summation of the corresponding `(values...)`.