The Memory connector stores all data and metadata in RAM on workers and both are discarded when Trino restarts.
To configure the Memory connector, create a catalog properties file
etc/catalog/memory.properties with the following contents:
memory.max-data-per-node defines memory limit for pages stored in this
connector per each node (default value is 128MB).
Create a table using the Memory connector:
CREATE TABLE memory.default.nation AS SELECT * from tpch.tiny.nation;
Insert data into a table in the Memory connector:
INSERT INTO memory.default.nation SELECT * FROM tpch.tiny.nation;
Select from the Memory connector:
SELECT * FROM memory.default.nation;
DROP TABLE memory.default.nation;
Trino supports all data types used within the Memory schemas so no mapping is required.
The connector provides read and write access to temporary data and metadata stored in memory. In addition to the globally available and read operation statements, the connector supports the following features:
Upon execution of a
DROP TABLE operation, memory is not released
immediately. It is instead released after the next write operation to the
The Memory connector supports the dynamic filtering optimization. Dynamic filters are pushed into local table scan on worker nodes for broadcast joins.
Delayed execution for dynamic filters#
For the Memory connector, a table scan is delayed until the collection of dynamic filters.
This can be disabled by using the configuration property
in the catalog file.
When one worker fails/restarts, all data that was stored in its memory is lost. To prevent silent data loss the connector throws an error on any read access to such corrupted table.
When a query fails for any reason during writing to memory table, the table enters an undefined state. The table should be dropped and recreated manually. Reading attempts from the table may fail, or may return partial data.
When the coordinator fails/restarts, all metadata about tables is lost. The tables remain on the workers, but become inaccessible.
This connector does not work properly with multiple coordinators, since each coordinator has different metadata.