Selecting a Distribution Key¶
Using the following principles to select a distribution key for a hash table:
The values of the distribution key should be discrete so that data can be evenly distributed on each DN. You can select the primary key of the table as the distribution key. For example, for a person information table, choose the ID number column as the distribution key.
Do not select the column that has a constant filter. For example, if a constant constraint (for example, zqdh= '000001') exists on the zqdh column in some queries on the dwcjk table, you are not advised to use zqdh as the distribution key.
With the above principles met, you can select join conditions as distribution keys, so that join tasks can be pushed down to DNs for execution, reducing the amount of data transferred between the DNs.
For a hash table, an inappropriate distribution key may cause data skew or poor I/O performance on certain DNs. Therefore, you need to check the table to ensure that data is evenly distributed on each DN. You can run the following SQL statements to check for data skew:
SELECT xc_node_id, count(1) FROM tablename group by xc_node_id order by xc_node_id desc;
xc_node_id corresponds to a DN. Generally, over 5% difference between the amount of data on different DNs is regarded as data skew. If the difference is over 10%, choose another distribution key.
You are not advised to add a column as a distribution key, especially add a new column and use the SEQUENCE value to fill the column. (Sequences may cause performance bottlenecks and unnecessary maintenance costs.)
Multiple distribution columns can be selected in GaussDB(DWS) to evenly distribute data.