redistributed table中的数据放在临时文件么？

In Greenplum, when data is redistributed during a Redistribute Motion, the data is not stored in temporary files on disk in the traditional sense. Instead, the data is sent directly to the target segments over the network and processed in memory. Here’s how it works:

1. Hash Calculation:

   • Each segment calculates the hash value of the specified key (e.g., cust_id) for each row in the table being redistributed (e.g., sales).

2. Data Transfer:

   • Rows are sent from the originating segment to the appropriate target segment based on the calculated hash value. This transfer happens over the network and is managed by the interconnect layer.

3. Receiving Segments:

   • When a segment receives a row, it does not recompute the hash value. Instead, it directly processes the incoming row in memory. The receiving segment then uses this data for further operations, such as joins or aggregations.

Temporary Files and Tablespaces

While the redistributed data itself is not stored in temporary files, Greenplum does use temporary tablespaces for other purposes, such as:

• Temporary Tables: When creating temporary tables or indexes, Greenplum can use the temp_tablespaces configuration parameter to specify where these temporary objects are stored.

• Spill Files: For operations like hash aggregates or hash joins that require sorting large datasets, Greenplum may create temporary spill files. These spill files are also stored in the tablespaces specified by temp_tablespaces.

Example Configuration

You can configure the temp_tablespaces parameter to control the location of temporary objects:

SET temp_tablespaces = 'fastspace';

This command sets the temporary tablespace to fastspace, which can be a tablespace created for faster access and processing of temporary data.

Conclusion

During a Redistribute Motion, the data is sent directly to the target segments and processed in memory without being stored in temporary files on disk. However, Greenplum does use temporary tablespaces for other temporary objects and spill files, which can be configured using the temp_tablespaces parameter.