问题
I've seen solutions for updating a sequence when it goes out of sync with the primary key it's generating, but I don't understand how this problem occurs in the first place.
Does anyone have insight into how a primary key field, with its default defined as the nextval of a sequence, whose primary keys aren't set explicitly anywhere, can go out of sync with the sequence? I'm using Postgres, and we see this occur now and then. It results eventually in a duplicate key constraint when the sequence produces an id for an existing row.
回答1:
Your application is probably occasionally setting the value for the primary key for a new row. Then postgresql has no need to get a new sequence and the sequence doesn't get updated.
回答2:
When a sequence number is allocated, it remains allocated, even if the TX that requested it is rolled back. So a number can be allocated that does not appear in the stable database. Of course, rows can also be deleted after they are created, so the maximum number found in the table need not be the maximum number ever allocated. This applies to any auto-incrementing type.
Also, depending on the technology used, separate sequences can be used with multiple tables, so a value might be missing from TableA but present in TableB. That could be because of a mistake in the use of sequence names, or it might be intentional.
来源：https://stackoverflow.com/questions/9314382/why-do-sql-id-sequences-go-out-of-sync-specifically-using-postgres