Effects of sequence caching

In the following example, assume that sequence seq has the default type of 64-bit integer, and each SPU has its own cache of 100,000 values. The select statement retrieves emp table rows, where emp is distributed across several data slices. For each row, the nextval sequence number is obtained from the cache on the host or SPU. The output shows how the nextval numbers can differ depending on which SPU contained the rows of the table, and how there are typically gaps between the sequence numbers.

dev.schema(admin)=> SELECT *, next value for seq FROM emp;
id |     name    |     grp      | nextval
---+-------------+--------------+---------
 4 | John        | mkt          |      49
 8 | Jim K       | sdev         |      52
12 | Jane        | adm          |      54
 1 | Julie       | dev          |  310079
 5 | Jackie      | hdev         |  310080
 9 | Mike        | hdev         |  310088
13 | Jill        | adm          |  310114
 2 | Tom         | adm          |  198235
 6 | Dan         | sdev         |  198312
10 | Craig       | sdev         |  198331
14 | Judy        | sdev         |  199010
 3 | Chuck       | sdev         |  243522
 7 | Dave        | sdev         |  256673
11 | David       | sdev         |  262004