Postgres UPDATE..FROM查询在同一行上有多个更新

Question

I am attempting to optimise a bulk UPDATE statement in Postgres using the UPDATE..FROM syntax to update from a list of values. It works except when the same row might be updated more than once in the same query.

For example say I have a table "test" with columns "key" and "value".

update test as t set value = v.value from (values 
    ('key1', 'update1'), 
    ('key1', 'update2') ) 
    as v (key, value) 
where t.key = v.key;

My desired behavior is for the row with key 'key1' to be updated twice, finishing with value set to 'update2'. In practice sometimes the value is updated to update1 and sometimes to update2. Also an update trigger function on the table is only invoked once.

The documentation ( http://www.postgresql.org/docs/9.1/static/sql-update.html ) explains why:

When a FROM clause is present, what essentially happens is that the target table is joined to the tables mentioned in the from_list, and each output row of the join represents an update operation for the target table. When using FROM you should ensure that the join produces at most one output row for each row to be modified. In other words, a target row shouldn't join to more than one row from the other table(s). If it does, then only one of the join rows will be used to update the target row, but which one will be used is not readily predictable.

Because of this indeterminacy, referencing other tables only within sub-selects is safer, though often harder to read and slower than using a join.

Is there any way to reformulate this query to achieve the behavior I'm looking for? Does the reference to sub-selects in the documentation give a hint?

Answer 1

Example (assuming id is a PK in the target table, and {id, date_modified} is a PK in the source table)

UPDATE target dst
Set a = src.a , b = src.b
FROM source src
WHERE src.id = dst.id
AND NOT EXISTS (
        SELECT *
        FROM source nx
        WHERE nx.id = src.id
        -- use an extra key field AS tie-breaker
        AND nx.date_modified > src.date_modified
        );

---

(in fact, this is deduplication of the source table -> forcing the source table to the same PK as the target table)