postgres_fdw establishes a connection to a foreign server during the first query that uses a foreign table associated with the foreign server. This connection is kept and re-used for subsequent queries in the same session. However, if multiple user identities (user mappings) are used to access the foreign server, a connection is established for each user mapping.

During a query that references any remote tables on a foreign server, postgres_fdw opens a transaction on the remote server if one is not already open corresponding to the current local transaction. The remote transaction is committed or aborted when the local transaction commits or aborts. Savepoints are similarly managed by creating corresponding remote savepoints.

The remote transaction uses SERIALIZABLE isolation level when the local transaction has SERIALIZABLE isolation level; otherwise it uses REPEATABLE READ isolation level. This choice ensures that if a query performs multiple table scans on the remote server, it will get snapshot-consistent results for all the scans. A consequence is that successive queries within a single transaction will see the same data from the remote server, even if concurrent updates are occurring on the remote server due to other activities. That behavior would be expected anyway if the local transaction uses SERIALIZABLE or REPEATABLE READ isolation level, but it might be surprising for a READ COMMITTED local transaction. A future LightDB release might modify these rules.

Note that it is currently not supported by postgres_fdw to prepare the remote transaction for two-phase commit.

postgres_fdw attempts to optimize remote queries to reduce the amount of data transferred from foreign servers. This is done by sending query WHERE clauses to the remote server for execution, and by not retrieving table columns that are not needed for the current query. To reduce the risk of misexecution of queries, WHERE clauses are not sent to the remote server unless they use only data types, operators, and functions that are built-in or belong to an extension that's listed in the foreign server's extensions option. Operators and functions in such clauses must be IMMUTABLE as well. For an UPDATE or DELETE query, postgres_fdw attempts to optimize the query execution by sending the whole query to the remote server if there are no query WHERE clauses that cannot be sent to the remote server, no local joins for the query, no row-level local BEFORE or AFTER triggers or stored generated columns on the target table, and no CHECK OPTION constraints from parent views. In UPDATE, expressions to assign to target columns must use only built-in data types, IMMUTABLE operators, or IMMUTABLE functions, to reduce the risk of misexecution of the query.

When postgres_fdw encounters a join between foreign tables on the same foreign server, it sends the entire join to the foreign server, unless for some reason it believes that it will be more efficient to fetch rows from each table individually, or unless the table references involved are subject to different user mappings. While sending the JOIN clauses, it takes the same precautions as mentioned above for the WHERE clauses.

The query that is actually sent to the remote server for execution can be examined using EXPLAIN VERBOSE.

In the remote sessions opened by postgres_fdw, the search_path parameter is set to just pg_catalog, so that only built-in objects are visible without schema qualification. This is not an issue for queries generated by postgres_fdw itself, because it always supplies such qualification. However, this can pose a hazard for functions that are executed on the remote server via triggers or rules on remote tables. For example, if a remote table is actually a view, any functions used in that view will be executed with the restricted search path. It is recommended to schema-qualify all names in such functions, or else attach SET search_path options (see CREATE FUNCTION) to such functions to establish their expected search path environment.

postgres_fdw likewise establishes remote session settings for various parameters:

These are less likely to be problematic than search_path, but can be handled with function SET options if the need arises.

It is not recommended that you override this behavior by changing the session-level settings of these parameters; that is likely to cause postgres_fdw to malfunction.

postgres_fdw can be used with remote servers dating back to LightDB 8.3. Read-only capability is available back to 8.1. A limitation however is that postgres_fdw generally assumes that immutable built-in functions and operators are safe to send to the remote server for execution, if they appear in a WHERE clause for a foreign table. Thus, a built-in function that was added since the remote server's release might be sent to it for execution, resulting in “function does not exist” or a similar error. This type of failure can be worked around by rewriting the query, for example by embedding the foreign table reference in a sub-SELECT with OFFSET 0 as an optimization fence, and placing the problematic function or operator outside the sub-SELECT.

Here is an example of creating a foreign table with postgres_fdw. First install the extension:

CREATE EXTENSION postgres_fdw;

Then create a foreign server using CREATE SERVER. In this example we wish to connect to a LightDB server on host 192.83.123.89 listening on port 5432. The database to which the connection is made is named foreign_db on the remote server:

CREATE SERVER foreign_server
        FOREIGN DATA WRAPPER postgres_fdw
        OPTIONS (host '192.83.123.89', port '5432', dbname 'foreign_db');

A user mapping, defined with CREATE USER MAPPING, is needed as well to identify the role that will be used on the remote server:

CREATE USER MAPPING FOR local_user
        SERVER foreign_server
        OPTIONS (user 'foreign_user', password 'password');

Now it is possible to create a foreign table with CREATE FOREIGN TABLE. In this example we wish to access the table named some_schema.some_table on the remote server. The local name for it will be foreign_table:

CREATE FOREIGN TABLE foreign_table (
        id integer NOT NULL,
        data text
)
        SERVER foreign_server
        OPTIONS (schema_name 'some_schema', table_name 'some_table');

It's essential that the data types and other properties of the columns declared in CREATE FOREIGN TABLE match the actual remote table. Column names must match as well, unless you attach column_name options to the individual columns to show how they are named in the remote table. In many cases, use of IMPORT FOREIGN SCHEMA is preferable to constructing foreign table definitions manually.

Shigeru Hanada <shigeru.hanada@gmail.com>