[Yandex Cloud documentation](../../index.md) > [Yandex Managed Service for Trino](../index.md) > [Concepts](index.md) > Connector Greenplum® # Greenplum®/Cloudberry connector The Greenplum®/Cloudberry [connector](index.md#connector) developed by Yandex based on the [PostgreSQL connector](https://trino.io/docs/current/connector/postgresql.html) allows Managed Service for Trino to read and write data to a Greenplum®/Cloudberry cluster. The connector supports [parallel data reading](#parallel-reading) from SEVERAL Greenplum® segments at the same time and [direct segment reading](#gpfdist-reading) over the GPFDIST protocol, which greatly improves query performance for large-scale data reads. You can use both data reading methods at the same time to optimize the use of resources in Trino and Greenplum® clusters. The Greenplum®/Cloudberry connector is available in Trino `476` or higher. ## Parallel data reading {#parallel-reading} During parallel reading from a table, data is parallelized based on the `gp_segment_id` metadata column value. The level of parallelism depends on the number of segments in the Greenplum® cluster. The maximum parallelism level is limited by the `greenplum.max-read-parallelism` [connector setting](#settings) and the relevant `max_read_parallelism` session property. Parallel reading is illustrated on the following diagram: ```mermaid sequenceDiagram participant w1 as Trino Worker 1 participant w2 as Trino Worker 2 participant m as Greenplum Master participant s1 as Greenplum Segment 1 participant s2 as Greenplum Segment 2 par Parallel data reading rect rgba(0, 128, 255, 0.25) w1 ->> m : SELECT WHERE gp_segment_id % 2 = 0 activate m m ->> s1 : read data activate s1 s1 -->> m : data deactivate s1 m -->> w1 : data deactivate m end and rect rgba(0, 128, 255, 0.25) w2 ->> m : SELECT WHERE gp_segment_id % 2 = 1 activate m m ->> s2 : read data activate s2 s2 -->> m : data deactivate s2 m -->> w2 : data deactivate m end end ``` When parallel reading is used, the connector performs only partial row filtering during [`LIMIT` pushdown](https://trino.io/docs/current/optimizer/pushdown.html#limit-pushdown). This does not affect the validity of the query results. ## Reading data over the GPFDIST protocol {#gpfdist-reading} The connector allows reading data directly from Greenplum® segments via [GPFDIST servers](https://techdocs.broadcom.com/us/en/vmware-tanzu/data-solutions/tanzu-greenplum/7/greenplum-database/utility_guide-ref-gpfdist.html) created on Trino workers. GPFDIST server activation is controlled by the `greenplum.gpfdist.server.enabled` [connector setting](#settings). In a Trino cluster, you can create not more than eight catalogs with active GPFDIST servers. Direct reading from Greenplum® segments follow these steps: 1. The connector creates an external table giving the address of the Trino worker to read data: ```sql CREATE WRITABLE EXTERNAL TEMPORARY TABLE ... LOCATION('gpfdist://'); ``` 1. The connector runs the following query: ```sql INSERT INTO SELECT ... FROM ; ``` 1. The Greenplum® segments send data to the Trino worker at the specified address. Data reading from segments is illustrated on the following diagram: ```mermaid sequenceDiagram participant w1 as Trino Worker participant m as Greenplum Master participant s1 as Greenplum Segment 1 participant s2 as Greenplum Segment 2 activate w1 w1 ->> m : CREATE WRITABLE EXTERNAL TABLE ext_table
LOCATION ('gpfdist://worker1/table1') activate m m -->> w1 : OK deactivate m w1 ->> m : INSERT INTO ext_table SELECT ... activate m par Reading data from segments m ->> s1 : send data to LOCATION activate s1 rect rgba(0, 128, 255, 0.25) loop Reading data from segment s1 ->> w1 : POST /ext_table {CSV} activate w1 w1 -->> s1 : OK deactivate w1 end end s1 -->> m : OK deactivate s1 and m ->> s2 : send data to LOCATION activate s2 rect rgba(0, 128, 255, 0.25) loop Reading data from segment s2 ->> w1 : POST /ext_table {CSV} activate w1 w1 -->> s2 : OK deactivate w1 end end s2 -->> m : OK deactivate s2 end m -->> w1 : OK deactivate m deactivate w1 ``` The use of the GPFDIST protocol for data reads introduces the following limitations for the connector: * No support for reading multidimensional arrays. * No support for reading string type arrays. * No support for [AS_JSON](https://trino.io/docs/current/connector/postgresql.html#array-type-handling) array processing mode. * When `LIMIT` and `ORDER BY` are pushed down at the same time ([Top-N pushdown](https://trino.io/docs/current/optimizer/pushdown.html#top-n-pushdown)), the connector sorts data only partially. This does not affect the validity of the query results. ## Connector settings {#settings} The connector's basic settings and their corresponding session properties match those of the [PostgreSQL connector](https://trino.io/docs/current/connector/postgresql.html) of the same version. In addition, the following settings are available: | Configuration | Description | Default
value | |--------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-------------------------------| | `greenplum.gpfdist.server.enabled` | Enables GPFDIST servers on Trino workers | `false` | | `greenplum.gpfdist.max-processing-threads` | Maximum size of thread pool for asynchronous GPFDIST query processing | `32` | | `greenplum.gpfdist.max-query-threads` | Maximum size of thread pool creating external Greenplum® tables and initiating data writes to an external table | `32` | | `greenplum.gpfdist.read.enabled` | Enables reading data directly from Greenplum® segments over the GPFDIST protocol | `false` | | `greenplum.gpfdist.read.buffer-size` |

Buffer size for GPFDIST data reads in [data size](https://trino.io/docs/current/admin/properties.html#data-size) format. If the buffer overflows, the connector suspends data reception from Greenplum® segments.

Matches the `gpfdist_read_buffer_size` session property

| `32MB` | | `greenplum.gpfdist.retry-timeout` |

Maximum time a Greenplum® segment will wait for a response to a GPFDIST query, in [duration](https://trino.io/docs/current/admin/properties.html#duration) format.

If the value is other than `null`, this setting overrides the Greenplum® [gpfdist_retry_timeout](https://techdocs.broadcom.com/us/en/vmware-tanzu/data-solutions/tanzu-greenplum/7/greenplum-database/ref_guide-config_params-guc-list.html#gpfdist_retry_timeout) property (300 seconds by default)

| `null` | | `greenplum.max-read-parallelism` |

Maximum parallelism for data reads from Greenplum®.

Matches the `max_read_parallelism` session property

| `1` (no parallelism) | | `greenplum.segment-fetch-required` |

Decides the connector's behavior if it fails to get informed about the number of Greenplum® segments:

  • If `true`, the Trino query will fail.
  • If `false`, the level of parallelism will be equal to the `max_read_parallelism` session property value.

Matches the `segment_fetch_required` session property

| `true` | #### Useful links {#see-also} * [Creating a Trino catalog](../operations/catalog-create.md)