Foreign Data Wrappers in PostgreSQL and a closer look at postgres_fdw
source link: https://www.tuicool.com/articles/hit/iQb2qir
Go to the source link to view the article. You can view the picture content, updated content and better typesetting reading experience. If the link is broken, please click the button below to view the snapshot at that time.
There are a few features in PostgreSQL that are very compelling, and that I rarely see in other RDBMSs. Some of these features are the driving force behind the growing popularity of PostgreSQL. This blog post is about one of my favourite features: FDW (Foreign Data Wrapper). As the name indicates, this feature allows a PostgreSQL database to treat tables in a remote PostgreSQL database as locally available tables.
The history of FDW began when SQL/MED came out as part of the ANSI SQL standard specification in 2003. MED stands for “Management of External Data”. By definition, “external data” is the data that the DBMS is able to access but does not manage. There are two parts for this specification:
- Foreign Table : this is about how to access external data sources and present them as relational tables.
- Datalink : this extends the functionality of database systems to include control over external files without the need to store their contents directly in the database, such as LOBs. A column of a table could directly refer a file.
PostgreSQL’s FDW capabilities addresses foreign tables only. It was introduced in PostgreSQL 9.1 and has been receiving improvements ever since.
Today there are a variety of FDWs which allow PostgreSQL to talk to most of the data sources we can think of. However, most FDWs are independent open source projects implemented as Postgres Extensions, and not officially supported by the PostgreSQL Global Development Group.
postgres_fdw
In this blog post we will take a closer look at the postgres_fdw which can be considered as the “reference implementation” for other FDW development efforts, and showcases its capabilities. This is the one FDW which comes with PostgreSQL source as a contrib extension module. The only other FDW which is part of PostgreSQL source tree is file_fdw .
Let’s look into postgres_fdw with a use case. In many organizations, there could be multiple systems catering to different functionalities/departments. For example, while an HR database may be holding the employee information the finance and payroll systems may need to access that same data. A common—but bad—solution for this is to duplicate the data in both systems. Data duplication often leads to problems, starting with data maintenance and accuracy. A smarter option, to avoid duplication while providing access to foreign databases to only the required data, is through FDWs.
Installation postgres_fdw
The Postgres Development Group (PGDG) offers PostgreSQL packages for all major Linux distributions. postgres_fdw itself is provided as a module that is usually included in the contrib package. In the example below we install such package for PostgreSQL 10 running on Red Hat/CentOS:
$ sudo yum install postgresql10-contrib.x86_64
Steps to setup
Let’s consider two PostgreSQL Instances, source instance and a destination instance
- source is the remote postgres server from where the tables are accessed by the destination database server as foreign tables.
- destination is another postgres server where the foreign tables are created which is referring tables in source database server.
We are going to use these definitions of source and destination in the rest of the post. Let’s assume that current application connects to destination database using a user app_user.
Step 1 : Create a user on the source
Create a user in the source server using the following syntax. This user account will be used by the destination server to access the source tables
postgres=# CREATE USER fdw_user WITH ENCRYPTED PASSWORD 'secret';
Step 2 : Create test tables (optional)
Let’s create a test table in the source server and insert a few records.
postgres=> create table employee (id int, first_name varchar(20), last_name varchar(20)); CREATE TABLE postgres=# insert into employee values (1,'jobin','augustine'),(2,'avinash','vallarapu'),(3,'fernando','camargos'); INSERT 0 3
Step 3 : Grant privileges to user in the source
Give appropriate privileges to the fdw_user on the source table. Always try to limit the scope of privilege to minimum to improve security.
An example syntax is as following :
postgres=# GRANT SELECT, INSERT, UPDATE, DELETE ON TABLE employee TO fdw_user; GRANT
Step 4 : Modify ACL in pg_hba.conf
We need to ensure that the proper authentication is setup for accessing source server from destination server.
Add an entry into pg_hba.conf as shown below, preferably at the beginning of the file.
host all all destination_server_ip/32 md5
Step 5 : Test connectivity and privileges on source
Before proceeding further, It is a good idea to make sure that we are able to connect to the source machine from this destination machine using the newly created database user (fdw_user).
In order to validate, on the destination server, use psql to connect to the source server:
$ psql -h hr -U fdw_user postgres
You could even validate all privileges on the tables which are to be presented as foreign tables using this connection.
Step 6 : Create postgres_fdw extension on the destination
Connect to destination server, and create the postgres_fdw extension in the destination database from where you wish to access the tables of source server. You must be a superuser to create the extension.
No postgres_fdw extension is needed on the source server.
postgres=# create extension postgres_fdw; CREATE EXTENSION
Validate if the extension is created using \dx. Following is an example validation log.
postgres=# \dx postgres_fdw List of installed extensions Name | Version | Schema | Description --------------+---------+--------+---------------------------------------------------- postgres_fdw | 1.0 | public | foreign-data wrapper for remote PostgreSQL servers (1 row)
Step 7: Grant privileges to user in the destination
Always better to limit the scope of the server definition to an application user. If a regular user needs to define a server, that user needs to have USAGE permission on the foreign data wrapper. Superuser can grant the privilege
postgres=# grant usage on FOREIGN DATA WRAPPER postgres_fdw to app_user; GRANT
Alternatively, superuser (postgres) can create a server definition and then grant USAGE permission on that server definition to the application user like this:
postgres=# GRANT USAGE ON FOREIGN SERVER hr TO app_user; GRANT
Step 8: Create a server definition
Now we can create a server definition. This foreign server is created using the connection details of the source server running on host “hr”. Let’s name the foreign server as itself as “hr”
postgres=> CREATE SERVER hr FOREIGN DATA WRAPPER postgres_fdw OPTIONS (dbname 'postgres', host 'hr', port '5432'); CREATE SERVER
Step 9: Create user mapping from destination user to source user
Create a mapping on the destination side for destination user (app_user) to remote source user (fdw_user)
postgres=> CREATE USER MAPPING for app_user SERVER hr OPTIONS (user 'fdw_user', password 'secret'); CREATE USER MAPPING
Step 10 : Create foreign table definition on the destination
Create a foreign table in the destination server with the same structure as the source table, but with OPTIONS specifying schema_name and table_name
postgres=# CREATE FOREIGN TABLE employee (id int, first_name character varying(20), last_name character varying(20)) SERVER hr OPTIONS (schema_name 'public', table_name 'employee');
Step 11 : Test foreign table
Validate whether we can query the foreign table we just created in the destination server.
postgres=> select * from employee; id | first_name | last_name ----+------------+----------- 1 | jobin | augustine 2 | avinash | vallarapu 3 | fernando | camargos (3 rows)
As we can see from the above example, data is been accessed from the source database.
Now you might be thinking: “ creating foreign tables one by one like this on the destination server is painful. Is it possible to do it automatically? “. The answer is yes – there is an option to import a full schema.
On the destination server, you can use the following syntax to import a schema.
postgres=# IMPORT FOREIGN SCHEMA "public" FROM SERVER hr INTO public;
If you wish to choose a certain list of tables for import, you can use the following syntax.
postgres=# IMPORT FOREIGN SCHEMA "public" limit to (employee) FROM SERVER hr INTO public;
In the above example, it will import the definition of only one table (employee).
Advantages of foreign tables
The main use case of the foreign tables is to make the data available to systems without actually duplicating/replicating it. There are even simple implementations of sharding using FDW, because data in the other shards can be made available for queries though FDWs.
A person coming from an Oracle-like background might think: “ I can get data from a remote database table using simple DBLinks so what is the difference? “. The main difference is that FDW will maintain the meta-data/table definition about the foreign table locally. This results in better decisions compared to sending a simple SELECT * FROM <TABLE> to pull all results. We are going to see some of these advantages.
Note : In the following section always pay special attention on those lines starting with “Remote SQL:”
Query optimization
Since the definition of the foreign table is held locally, all query optimizations are made for remote executions too. Let’s consider a slightly more complex example where we have EMP (employee) and DEPT (department) tables in the HR database and SALGRADE (salary grade) table in the finance database. Suppose we want to know how many employees there are with a particular salary grade:
SELECT COUNT(*) FROM EMP JOIN SALGRADE ON EMP.SAL > SALGRADE.LOSAL AND EMP.SAL < SALGRADE.HISAL WHERE SALGRADE.GRADE = 4;
Let’s see how PostgreSQL handles this:
QUERY PLAN -------------------------------------------------------------------------------------------------------------------------- Aggregate (cost=805.44..805.45 rows=1 width=8) Output: count(*) -> Nested Loop (cost=100.00..798.33 rows=2844 width=0) Join Filter: ((emp.sal > (salgrade.losal)::double precision) AND (emp.sal < (salgrade.hisal)::double precision)) -> Foreign Scan on public.emp (cost=100.00..186.80 rows=2560 width=8) Output: emp.empno, emp.ename, emp.job, emp.mgr, emp.hiredate, emp.sal, emp.comm, emp.deptno <b>Remote SQL: SELECT sal FROM public.emp</b> -> Materialize (cost=0.00..35.55 rows=10 width=8) Output: salgrade.losal, salgrade.hisal -> Seq Scan on public.salgrade (cost=0.00..35.50 rows=10 width=8) Output: salgrade.losal, salgrade.hisal Filter: (salgrade.grade = 4)
Please pay special attention for the line reading :
Remote SQL: SELECT sal FROM public.emp
It knows that only the sal column need to be fetched from the remote database.
If we change the count(*) to ename (Employee Name) column, the remote SQL changes like:
Remote SQL: SELECT ename, sal FROM public.emp
PostgreSQL tries to pull only the absolutely necessary data from the remote server.
Writable foreign tables
At the beginning, foreign tables were just readable. But, with time, the community introduced writable foreign tables functionality in PostgreSQL. Let us consider the following situation where management wants to give a salary increase of 10% to grade 3 employees:
UPDATE emp SET sal = sal * 1.1 FROM salgrade WHERE emp.sal > salgrade.losal AND emp.sal < salgrade.hisal AND salgrade.grade = 3;
In this case, we are updating data on a remote table using a join condition with a local table. As we can see in the explain plan, an UPDATE statement is more complex because it involves 2 steps. First it needs to fetch the data from the remote table to complete the join operation. Then, it updates the rows in the foreign table.
QUERY PLAN ---------------------------------------------------------------------------------------------------------------------------------------------------------- Update on public.emp (cost=100.00..300.71 rows=669 width=118) Remote SQL: UPDATE public.emp SET sal = $2 WHERE ctid = $1 -> Nested Loop (cost=100.00..300.71 rows=669 width=118) Output: emp.empno, emp.ename, emp.job, emp.mgr, emp.hiredate, (emp.sal * '1.1'::double precision), emp.comm, emp.deptno, emp.ctid, salgrade.ctid Join Filter: ((emp.sal > (salgrade.losal)::double precision) AND (emp.sal < (salgrade.hisal)::double precision)) -> Foreign Scan on public.emp (cost=100.00..128.06 rows=602 width=112) Output: emp.empno, emp.ename, emp.job, emp.mgr, emp.hiredate, emp.sal, emp.comm, emp.deptno, emp.ctid Remote SQL: SELECT empno, ename, job, mgr, hiredate, sal, comm, deptno, ctid FROM public.emp FOR UPDATE -> Materialize (cost=0.00..35.55 rows=10 width=14) Output: salgrade.ctid, salgrade.losal, salgrade.hisal -> Seq Scan on public.salgrade (cost=0.00..35.50 rows=10 width=14) Output: salgrade.ctid, salgrade.losal, salgrade.hisal Filter: (salgrade.grade = 3)
Operator and function pushdown
PostgreSQL 9.5 release included the capability to assess and decide on the safety of pushing a function execution to remote server. Built-in functions are good candidates for this:
SELECT avg(sal) FROM EMP WHERE EMP.SAL > (SELECT LOSAL FROM SALGRADE WHERE GRADE = 4);
This statement results in the following query plan
QUERY PLAN --------------------------------------------------------------------------- Foreign Scan (cost=137.63..186.06 rows=1 width=8) Output: (avg(emp.sal)) Relations: Aggregate on (public.emp) Remote SQL: SELECT avg(sal) FROM public.emp WHERE ((sal > $1::integer)) InitPlan 1 (returns $0) -> Seq Scan on public.salgrade (cost=0.00..35.50 rows=10 width=4) Output: salgrade.losal Filter: (salgrade.grade = 4)
If the planner finds that the majority of records needs to be fetched from a remote server, it may not push the function execution to the remote server. For example:
SELECT avg(sal) FROM EMP JOIN SALGRADE ON EMP.SAL > SALGRADE.LOSAL AND EMP.SAL < SALGRADE.HISAL WHERE SALGRADE.GRADE = 4;
In this case, the planner decides to do the function execution on the local server:
QUERY PLAN -------------------------------------------------------------------------------------------------------------------------- Aggregate (cost=805.44..805.45 rows=1 width=8) Output: avg(emp.sal) -> Nested Loop (cost=100.00..798.33 rows=2844 width=8) Output: emp.sal Join Filter: ((emp.sal > (salgrade.losal)::double precision) AND (emp.sal < (salgrade.hisal)::double precision)) -> Foreign Scan on public.emp (cost=100.00..186.80 rows=2560 width=8) Output: emp.empno, emp.ename, emp.job, emp.mgr, emp.hiredate, emp.sal, emp.comm, emp.deptno Remote SQL: SELECT sal FROM public.emp -> Materialize (cost=0.00..35.55 rows=10 width=8) Output: salgrade.losal, salgrade.hisal -> Seq Scan on public.salgrade (cost=0.00..35.50 rows=10 width=8) Output: salgrade.losal, salgrade.hisal Filter: (salgrade.grade = 4) (13 rows)
A great improvement in PostgreSQL 9.6 is that the function does’t need to be even a built-in function. If a user defined function or operator is immutable it becomes a good candidate for being executed in the remote server.
Join push down
In many cases, it is worth pushing down the entire join operations to the remote server in such a way only the results need to be fetched to the local server. PostgreSQL handles this switching intelligently. Here’s an example:
postgres=# EXPLAIN VERBOSE SELECT COUNT(*) FROM EMP JOIN DEPT ON EMP.deptno = DEPT.deptno AND DEPT.deptno=10;
QUERY PLAN ----------------------------------------------------------------------------------------------------------------------------- Foreign Scan (cost=100.56..194.84 rows=1 width=8) Output: (count(*)) Relations: Aggregate on ((public.emp) INNER JOIN (public.dept)) Remote SQL: SELECT count(*) FROM (public.emp r1 INNER JOIN public.dept r2 ON (((r2.deptno = 10)) AND ((r1.deptno = 10)))) (4 rows)
Predicate push down
There are two options when executing a query against a foreign table:
- Fetch the data locally and apply the predicates like filtering condition locally.
- Send the filtering condition to the remote server and have it applied there.
The latter will can be the best option in many cases.
If you consider the previous example, we can see that the predicate specification like “DEPT.deptno=10;” is pushed down to the remote server through foreign tables and applied there separately like this:
Remote SQL: SELECT count(*) FROM (public.emp r1 INNER JOIN public.dept r2 ON (((r2.deptno = 10)) AND ((r1.deptno = 10))))
PostgreSQL not only pushed the predicate, it also rewrote the query we sent to avoid one extra AND condition.
Aggregate push down
Just like predicate push down, here PostgreSQL also considers 2 options:
- Execute the aggregates on the remote server and pull the result back to the local server
- Do the aggregate calculations on the local database instance after collecting all required data from remote database
We’ve already seen an aggregate pushdown example as part of the function pushdown, since we’ve used an aggregate function for that example. Here’s another simple example:
postgres=# explain verbose select deptno,count(*) from emp group by deptno; QUERY PLAN ------------------------------------------------------------------ Foreign Scan (cost=114.62..159.88 rows=200 width=12) Output: deptno, (count(*)) Relations: Aggregate on (public.emp) Remote SQL: SELECT deptno, count(*) FROM public.emp GROUP BY 1 (4 rows)
In this case, all of the aggregate calculation happens on the remote server.
Triggers and Check constraints on Foreign tables
We have seen that foreign tables can be writable. PostgreSQL provides features to implement check constraints and triggers on the foreign table as well. This allows us to have powerful capabilities in the local database. For example, all validations and auditing can take place on the local server. The remote DMLs can be audited separately, or a different logic can be applied for local and remote triggers and constraint validations.
Conclusion
FDWs in PostgreSQL, postgres_fdw in particular, provides very powerful and useful features by which, in many cases, we can avoid the complex duplicating and replicating of data. It provides a mechanism for ACID compliant transactions between two database systems. postgres_fdw works as a reference implementation for the development of other fdw implementations. In the coming days we will be covering some of these.
More articles you might enjoy:
If you found this article useful, why not take a look at some of our other posts on PostgreSQL?
Recommend
About Joyk
Aggregate valuable and interesting links.
Joyk means Joy of geeK