Skip to content

Posmulten library is an open-source project for the generation of SQL DDL statements that make it easy for implementation of Shared Schema Multi-tenancy strategy via the Row Security Policies in the Postgres database.

License

LGPL-2.1, Unknown licenses found

Licenses found

LGPL-2.1
LICENSE
Unknown
license.txt
Notifications You must be signed in to change notification settings

starnowski/posmulten

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Posmulten

Build Status Run tests for posmulten Maven Central CodeQL CodeQL for release branch

Introduction

Posmulten library is an open-source project for the generation of SQL DDL statements that make it easy to implementation of the shared schema multi-tenancy strategy via the row security policies in the Postgres database. Project is tested for compatibility with the Postgres database in versions 9.6, 10.14, 11.9, 12.4, and 13.0. The library is written in a java programming language. The project required at least java 8.

Multi-tenancy

Based on hibernate documentation the term multi-tenancy in general is applied to software development to indicate an architecture in which a single running instance of an application simultaneously serves multiple clients (tenants). There are three main strategies separate database, separate schema and shared schema Below you can find short description of those strategies. Of course we will focus more on shared schema. For more information about what pros and cons of each approach please check below links:

Separate database

In this strategy each tenant's data is stored in separate database. For this obvious reason this approach gives the highest isolation level.

Separate schema

Strategy assumes that each tenant's data is kept in his own schema but all those schemas exists in single database instance. Obviously this approach offers lower isolation level than separate database but allows to potentially save costs for infrastructure by not having multiple database instances.

Shared schema

In this strategies data all tenants are kept in single database and same schema. Although there is no limitation that there has to be only one schema in database but all tenants should have same access to them. The strategy assumes that all tables in a database (with an exception for tables that stores vocabulary data or data available for all tenants) have a column that stores tenant identifier. Of course, based on this value column, we know which tenant is the owner of the table raw. Obviously, this approach offers a lower isolation level than both previous strategies but, just like shared schema strategy, allows to save costs for infrastructure potentially. In comparison to shared schema strategy, we can say that although Postgres can handle multiple schemas and even we can find comments from persons who use this strategy. We can find statements that with larger number of tenants the maintenance for this approach is harder than when schema is shared between tenants. Executing ddl scripts for hundreds schemas might be easily automated but still it might create complexity and slower deployment. It does not mean that executing ddl scripts for database with shared schema approach is error free. Every change should be considered. There are some other important considerations which should taken before deciding for shared schema strategy.

Implementation requirements

Just like it mention in section above, the shared schema strategy assumes that all tables shared by tenant have column which defines its owner. This means that the value of this column has to be checked in each SQL query and operation. Column has to checked during execution of SELECT statement and all operations that modifies data like UPDATE, INSERT and DELETE. In case when decision is made that application should handle this checks then every sql statement send to database by application need to contains in WHERE statement condition for tenant column for each table. It might be easier if application use some ORM framework. But still if application code contains any custom query then developer has to be aware that he needs to add checks for tenant identifier column. The pros for this approach is that if we would ever consider changing database engine for application then it would be quick easy However there are cons for such situation if for example we would consider of having many application that use same database. All application should have implemented tenant column checks. Not to mention if those application would be writen in different programing languages.

Another approach is that the database engine is going to handle tenant column checks. Of course, not all database engines support such features. One such database is Oracle with a feature called the virtual private database. And also Postgres has a similar feature called the row security policy which posmulten is using. This approach might be better if you planned to have multiple different applications connected to your database. Besides some connection adjustments, there is no additional logic that has to be added to every SQL statement created by the application code. Not to mention the situation when connected applications are written in different programming languages.

One thing that might be considered during implementation is constraints that check if the foreign key columns reference rows that belong to the same tenant. Of course, assuming if the above requirement is fulfilled, then even if SQL injection will succeed, the application that checks tenant column should not display the record for other tenants or modify it. But there might be a situation when a separate application is also connected to a database but operate on data without checking the tenant column. For example, statistical data gathering or other specific goals where joining data from different tables is crucial. Posmulten helps to create such constraints for the foreign key columns.

How posmulten helps to implement shared schema strategy?

Below there is short explanation how posmulten helps to implement shared schema strategy and what ddl statements are generated. The ddl statements examples are generated for tables users and posts.

CREATE TABLE public.users
(
    id bigint NOT NULL,
    name character varying(255),
    tenant_id character varying(255),
    CONSTRAINT users_pkey PRIMARY KEY (id)
)
WITH (
    OIDS = FALSE
)
TABLESPACE pg_default;

CREATE TABLE public.posts
(
    id bigint NOT NULL,
    text text NOT NULL,
    user_id bigint NOT NULL,
    tenant_id character varying(255),
    CONSTRAINT fk_posts_user_id FOREIGN KEY (user_id)
              REFERENCES users (id) MATCH SIMPLE,
    CONSTRAINT posts_pkey PRIMARY KEY (id)
)
WITH (
    OIDS = FALSE
)
TABLESPACE pg_default;

Setting RLS policy

Posmulten use row security policy mechanism to handle tenant data isolation. Creates policy for all operations (syntax "ALL" applies to SQL commands like INSERT, SELECT, UPDATE, DELETE) and specific user.

CREATE POLICY posts_table_rls_policy ON posts
FOR ALL
TO "postgresql-core-owner"
USING (tenant_has_authorities(tenant_id, 'ALL', 'USING', 'posts', 'public'))
WITH CHECK (tenant_has_authorities(tenant_id, 'ALL', 'WITH_CHECK', 'posts', 'public'));

A statement that enables the row security policy mechanism is also created.

ALTER TABLE "posts" ENABLE ROW LEVEL SECURITY;

IMPORTANT!
When the database user for who row security policy is supposed to be created is table owner, it is required to force row security policy mechanism. By default, posmulten does not force this mechanism.

Function that checks tenant access to a table row

Creates a DDL statement for a function that checks if the current tenant for the database session has access to table row based on tenant column (for the case below it is "tenant_id") value. Current function logic is not complex, but this might be changed in the next release. The function name can be customize. The type of function first argument is the same as the default type for tenant identifier value which can be customized.

CREATE OR REPLACE FUNCTION tenant_has_authorities(VARCHAR(255), VARCHAR(255), VARCHAR(255), VARCHAR(255), VARCHAR(255)) RETURNS BOOLEAN AS $$
SELECT is_id_equals_current_tenant_id($1)
$$ LANGUAGE sql
STABLE
PARALLEL SAFE;

Function that checks if the passed identifier is the same as the current tenant identifier

Creates a DDL statement for a function that checks if the current tenant identifier set for the database session is equal to the passed tenant identifier. The function name and function argument type can be changed.

CREATE OR REPLACE FUNCTION is_id_equals_current_tenant_id(VARCHAR(255)) RETURNS BOOLEAN AS $$
SELECT $1 = get_current_tenant_id()
$$ LANGUAGE sql
STABLE
PARALLEL SAFE;

Function that returns the current tenant identifier

Next function that is created by Posmulten project is function that returns value of identifier for current tenant. Function reads property value that is save for database session. The property name and function name can customized.

CREATE OR REPLACE FUNCTION get_current_tenant_id() RETURNS VARCHAR(255) AS $$
SELECT current_setting('c.c_ten')
$$ LANGUAGE sql
STABLE
PARALLEL SAFE;

Function that set the current tenant identifier

Another crucial function is that one which sets current tenant identifier in database connection. The property name and function name can customized. The type of function argument is the same as the default type for tenant identifier value which can be customized.

CREATE OR REPLACE FUNCTION set_current_tenant_id(VARCHAR(255)) RETURNS VOID AS $$
BEGIN
PERFORM set_config('c.c_ten', $1, false);
END
$$ LANGUAGE plpgsql
VOLATILE;

Connecting to Database

After correct setup of RLS policies the way how we connect to database and execute sql script has to be changed a little bit. For example let's assume that the name of session property that stores current tenant identifier is "c.c_ten". The function that set value for this property is set_current_tenant_id(VARCHAR(255)) and function that return its value is called get_current_tenant_id(). If opening connection as user for which the RLS was configured it is required to set tenant identifier. Below there is example that show what happens in situation when tenant identifier is not set. Important, all examples were generated in the Postgres database in version 9.6.

SELECT * FROM users;

For displaying rows from the users table without setting tenant identifier there is going to be thrown exception.

ERROR:  unrecognized configuration parameter "c.c_ten"
SQL state: 42704

Let's pretend that our newly created tenant should have the name "SOME_TENANT_1".

SELECT set_current_tenant_id('SOME_TENANT_1');
SELECT COUNT(*) FROM users;

In the beginning, there are no records. Below there is an example how to insert into table for tenant.

SELECT set_current_tenant_id('SOME_TENANT_1');
INSERT INTO users (id, name) VALUES (1, 'Szymon Tarnowski');
INSERT INTO users (id, name, tenant_id) VALUES (2, 'John Doe', 'SOME_TENANT_1');

IMPORTANT!The first insert statement without tenant_id is possible because it was used a statement that adds default value. After inserting above records, the previous select statements should return result 2. In case if we want to change current tenant to 'TENANT_X_2' and display all rows we should get zero results.

SELECT set_current_tenant_id('TENANT_X_2');
SELECT COUNT(*) FROM users;

After adding a single record we should get the one as a select query result.

SELECT set_current_tenant_id('TENANT_X_2');
INSERT INTO users (id, name) VALUES (3, 'Jimmy Doe');
SELECT COUNT(*) FROM users;

After deleting all results from the users table, the select query should return zero results.

SELECT set_current_tenant_id('TENANT_X_2');
DELETE FROM users;
SELECT COUNT(*) FROM users;

But if we change current tenant to the 'SOME_TENANT_1' then for select query we should get two records.

SELECT set_current_tenant_id('SOME_TENANT_1');
SELECT COUNT(*) FROM users;

Adding constraints for foreign key columns

The library has the possibility to add a foreign key constraint that checks if foreign key value references to the row which belongs to the current tenant.

ALTER TABLE "posts" ADD CONSTRAINT posts_users_fk_cu CHECK ((user_id IS NULL) OR (is_user_belongs_to_current_tenant(user_id)));

The library also creates the function that checks if the passed identifier exists in a specific table. The statement is created only when the is a request for the creation of a foreign key constraint. The types of arguments for this function are based on column types that a part of the primary key declared for the table which foreign key references to.

CREATE OR REPLACE FUNCTION is_user_belongs_to_current_tenant(bigint) RETURNS BOOLEAN AS $$
SELECT EXISTS (
	SELECT 1 FROM users rt WHERE rt.id = $1 AND rt.tenant_id = get_current_tenant_id()
)
$$ LANGUAGE sql
STABLE
PARALLEL SAFE;

Other columns modifications

There are also operations generated by the library that alters table columns. For example, the library can add a statement that adds default value for the tenant identifier column or add a tenant column to the table.

How to start using posmulten

Posmulten is a java project, so besides other projects written in java, you can also use the project as a dependency on projects written in languages executed on java virtual machine. The project required at least java 8.

IMPORTANT!
There is also posmulten-ddl wrapper script that gives the possibility of using the Posmulten library outside the java project.

Setting maven dependency

The project is available in the central maven repository. You can use it just by adding it as a dependency in the project descriptor file (pom.xml).

        <dependency>
            <groupId>com.github.starnowski.posmulten</groupId>
            <artifactId>postgresql-core</artifactId>
            <version>0.9.0</version>
        </dependency>

Building project locally

If someone would like to build the project locally from the source please see the CONTRIBUTING.md file to check how to set up the project locally.

How to start using builder

The library's main public component is DefaultSharedSchemaContextBuilder, which produces all required DDL statements based on passed criteria. For example:

import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
//...
    Map<String, String> usersTablePrimaryKeyNameToType = new HashMap();
    usersTablePrimaryKeyNameToType.put("id", "bigint");
    Map<String, String> postsTablePrimaryKeyNameToType = new HashMap();
    postsTablePrimaryKeyNameToType.put("id", "bigint");
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null);
    defaultSharedSchemaContextBuilder.setGrantee("db_user");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", usersTablePrimaryKeyNameToType, "tenant_id", "users_table_rls_policy");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("posts", postsTablePrimaryKeyNameToType, "tenant_id", "posts_table_rls_policy");
    //... other criteria
    ISharedSchemaContext sharedSchemaContext = defaultSharedSchemaContextBuilder.build();

The builder component, as a result of method build(), returns an object of type ISharedSchemaContext. The type contains all properties required to create a shared schema strategy and components that help using it correctly in your java code.

Applying builder changes

One of the crucial methods of the ISharedSchemaContext interface is "getSqlDefinitions()". It returns list of object of type "SQLDefinition" that contains method getCreateScript(). The method returns DDL statement that should be applied to implement shared schema strategy.
IMPORTANT!
The list's order is crucial because the DDL statements returned by the getCreateScript() method that represents each object of the list should be applied based on the list's order. Just like in code example below:

import com.github.starnowski.posmulten.postgresql.core.common.SQLDefinition;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.jdbc.core.JdbcTemplate;
//...
        @Autowired
        JdbcTemplate jdbcTemplate;
//...
        List<SQLDefinition> sqlDefinitions = sharedSchemaContext.getSqlDefinitions();
        sqlDefinitions.forEach(sqlDefinition ->
        {
            jdbcTemplate.execute(sqlDefinition.getCreateScript());
        });

Dropping builder changes

The second important method of the SQLDefinition type is getDropScript(). It returns a statement that drops changes applied by the statement returned by the getCreateScript() method.
IMPORTANT!
By default, there is no assumption that statement has to contains the compensation operation for operation returned by the getCreateScript() method. This means that the operation can not be by default treated as a rollback operation, but an operation that removes changes applied by statement returned by the getCreateScript() method.


IMPORTANT!
Just like was mentioned in the previous section that the statements returned by the getCreateScript() method for objects from the list returned from the getSqlDefinitions() method, should be applied based on the list's order. Based on that fact, the statements returned by the getDropScript() method for objects, should be executed in the reverse list's order.
import com.github.starnowski.posmulten.postgresql.core.common.SQLDefinition;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.jdbc.core.JdbcTemplate;
//...
        @Autowired
        JdbcTemplate jdbcTemplate;
//...
        List<SQLDefinition> sqlDefinitions = sharedSchemaContext.getSqlDefinitions();
        //Run sql statements in reverse order
        LinkedList<SQLDefinition> stack = new LinkedList<>();
        sqlDefinitions.forEach(stack::push);
        stack.forEach(sqlDefinition ->
        {
            jdbcTemplate.execute(sqlDefinition.getDropScript());
        });

Using posmulten components with database connection

Other useful components that type ISharedSchemaContext contains is object of type "ISetCurrentTenantIdFunctionPreparedStatementInvocationFactory" returned by method getISetCurrentTenantIdFunctionPreparedStatementInvocationFactory(). Component of type ISetCurrentTenantIdFunctionPreparedStatementInvocationFactory returns statement that sets current tenant identifier and can be used by PreparedStatement object.

import com.github.starnowski.posmulten.postgresql.core.rls.function.ISetCurrentTenantIdFunctionPreparedStatementInvocationFactory;
import java.sql.PreparedStatement;
import java.sql.ResultSet;
//...
        @Autowired
        JdbcTemplate jdbcTemplate;
//...
        ISetCurrentTenantIdFunctionPreparedStatementInvocationFactory setCurrentTenantIdFunctionPreparedStatementInvocationFactory = sharedSchemaContext.getISetCurrentTenantIdFunctionPreparedStatementInvocationFactory();
        jdbcTemplate.execute(setCurrentTenantIdFunctionPreparedStatementInvocationFactory.returnPreparedStatementThatSetCurrentTenant(), (PreparedStatementCallback<Integer>) preparedStatement -> {
            preparedStatement.setString(1, "some-tenant-id-SDFAFD-DZXCV");
            preparedStatement.executeQuery();
            ResultSet rs = preparedStatement.getConnection().createStatement().executeQuery("SELECT COUNT(*) FROM users");
            rs.next();
            return rs.getInt(1);
        });

Assuming that function that sets current tenant has name 'set_tenant' the ISetCurrentTenantIdFunctionPreparedStatementInvocationFactory#returnPreparedStatementThatSetCurrentTenant() method is going to return string just like below:

SELECT set_tenant(?);

Type ISharedSchemaContext contains also object of type ISetCurrentTenantIdFunctionInvocationFactory returned by method getISetCurrentTenantIdFunctionInvocationFactory(). The ISetCurrentTenantIdFunctionInvocationFactory type has method that returns similar result as ISetCurrentTenantIdFunctionPreparedStatementInvocationFactory but without "?" mark. Instead it sets specific value passed as argument, for example:

    String tenant = "TXDS-tenant-id";
    ISetCurrentTenantIdFunctionInvocationFactory setCurrentTenantIdFunctionDefinition = sharedSchemaContext.getISetCurrentTenantIdFunctionInvocationFactory();
    jdbcTemplate.execute(String.format("%1$s UPDATE users %2$s SET name = '%2$s' WHERE id = %3$d ;", setCurrentTenantIdFunctionDefinition.generateStatementThatSetTenant(tenant), updatedName, user.getId()));

IMPORTANT!

Please in mind that if it is required to use custom sql statement that might use values passed out side application, for instance from application user then is more secure to use PreparedStatement type. It is also worth mentioning that some frameworks for optimization purposes use database connection pools. Just like in the examples above where the JdbcTemplate is used. In case when the component is used in the context of the already existed transaction, then the Spring framework might use a connection that is kept in context for the local thread.

Setting default database schema

Builder component has one constructor with one String parameter (there is a task to add non-argument constructor 135). The constructor is name of default schema for which statement are going to be created. If null is going to passed then by default the schema name is not going to be added to created DDL statements. For example the code below:

import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
//...
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null); // Constructor with passed null parameter
    Map<String, String> usersTablePrimaryKeyNameToType = new HashMap();
    usersTablePrimaryKeyNameToType.put("id", "bigint");
    Map<String, String> postsTablePrimaryKeyNameToType = new HashMap();
    postsTablePrimaryKeyNameToType.put("id", "bigint");
    Map<String, String> postsTableForeignKeyToUserPrimaryKey = new HashMap();
    postsTableForeignKeyToUserPrimaryKey.put("user_id", "id");
    defaultSharedSchemaContextBuilder.setGrantee("postgresql-core-owner");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", usersTablePrimaryKeyNameToType, "tenant_id", "users_table_rls_policy");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("posts", postsTablePrimaryKeyNameToType, "tenant_id", "posts_table_rls_policy");
    //...
    defaultSharedSchemaContextBuilder.createSameTenantConstraintForForeignKey("posts", "users, postsTableForeignKeyToUserPrimaryKey, "posts_users_fk_cu");

is going to produce below statements (displayed only few results)

CREATE POLICY users_table_rls_policy ON users
FOR ALL
TO "postgresql-core-owner"
USING (tenant_has_authorities(tenant_id, 'ALL', 'USING', 'users', 'public'))
WITH CHECK (tenant_has_authorities(tenant_id, 'ALL', 'WITH_CHECK', 'users', 'public'));

---..
ALTER TABLE "posts" ADD CONSTRAINT posts_users_fk_cu CHECK ((user_id IS NULL) OR (is_user_belongs_to_current_tenant(user_id)));

In case if the "non_public_schema" value would be passed to constructor the result would look different.

//...
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder("non_public_schema"); // Constructor with passed "non_public_schema" parameter
//...

It would contain the "non_public_schema" schema name.

CREATE POLICY users_table_rls_policy ON non_public_schema.users
FOR ALL
TO "postgresql-core-owner"
USING (non_public_schema.tenant_has_authorities(tenant_id, 'ALL', 'USING', 'users', 'non_public_schema'))
WITH CHECK (non_public_schema.tenant_has_authorities(tenant_id, 'ALL', 'WITH_CHECK', 'users', 'non_public_schema'));

---..
ALTER TABLE "non_public_schema"."posts" ADD CONSTRAINT posts_users_fk_cu CHECK ((user_id IS NULL) OR (non_public_schema.is_user_belongs_to_current_tenant(user_id)));

Setting default database user for RLS policy

Builder required to specify default database user for which the row security policies are going to be created. For the below criteria:

import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
//...
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null); // Constructor with passed null parameter
    Map<String, String> usersTablePrimaryKeyNameToType = new HashMap();
    usersTablePrimaryKeyNameToType.put("id", "bigint");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", usersTablePrimaryKeyNameToType, "tenant_id", "users_table_rls_policy");
    //...

and database user with name "postgresql-core-owner"

    //...
    defaultSharedSchemaContextBuilder.setGrantee("postgresql-core-owner");
    //...

the builder will produce:

CREATE POLICY users_table_rls_policy ON users
FOR ALL
TO "postgresql-core-owner"
USING (tenant_has_authorities(tenant_id, 'ALL', 'USING', 'users', 'public'))
WITH CHECK (tenant_has_authorities(tenant_id, 'ALL', 'WITH_CHECK', 'users', 'public'));

For user "app-first-user"

    //...
    defaultSharedSchemaContextBuilder.setGrantee("app-first-user");
    //...

the builder will produce:

CREATE POLICY users_table_rls_policy ON users
FOR ALL
TO "app-first-user"
USING (tenant_has_authorities(tenant_id, 'ALL', 'USING', 'users', 'public'))
WITH CHECK (tenant_has_authorities(tenant_id, 'ALL', 'WITH_CHECK', 'users', 'public'));

Setting RLS Policy for table

The most crucial thing from builder perspective is to define which tables need have created row security policy. The RLS policy is added via method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#createRLSPolicyForTable(String table, Map<String, String> primaryKeyColumnsList, String tenantColumnName, String rlsPolicyName)

table - (Required) table name.
primaryKeyColumnsList - (Required) map of primary key columns and their types in the table. The column name is the map key and the column type is its value.
tenantColumnName - (Optional) name of the column that stores tenant identifier in table. In null value will be passed then default tenant column name will be used. Set by builder or custom value.
rlsPolicyName - (Required) name of row level security policy (see task 48).

For the below criteria:

import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
//...
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder("some_schema");
    defaultSharedSchemaContextBuilder.setGrantee("db-us");
    //...

and database policy declaration for table users_tab:

    //...
    usersTablePrimaryKeyNameToType.put("user_uuid", "UUID");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users_tab", usersTablePrimaryKeyNameToType, "ten_col", "users_policy");
    //...

the builder will produce:

CREATE POLICY users_policy ON some_schema.users_tab
FOR ALL
TO "db-us"
USING (some_schema.tenant_has_authorities(ten_col, 'ALL', 'USING', 'users_tab', 'some_schema'))
WITH CHECK (some_schema.tenant_has_authorities(ten_col, 'ALL', 'WITH_CHECK', 'users_tab', 'some_schema'));

Setting RLS Policy for a table with a multi-column primary key

In case when the table has a more complex primary key than a single column. All columns that are part of the primary key have to be passed to the method. For example, for below comments table with two-column primary:

CREATE TABLE public.comments
(
id int NOT NULL,
user_id bigint NOT NULL,
text text NOT NULL,
tenant character varying(255),

parent_comment_id int,
parent_comment_user_id bigint,

CONSTRAINT fk_comments_users_id FOREIGN KEY (user_id)
REFERENCES public.users (id) MATCH SIMPLE,

CONSTRAINT fk_comments_parent_id FOREIGN KEY (parent_comment_id, parent_comment_user_id)
REFERENCES public.comments (id, user_id) MATCH SIMPLE,

CONSTRAINT comments_pkey PRIMARY KEY (id, user_id)

the declaration should like below:

    //...
    Map<String, String> commentsTablePrimaryKeyNameToType = new HashMap();
    commentsTablePrimaryKeyNameToType.put("id", "int");
    commentsTablePrimaryKeyNameToType.put("user_id", "bigint");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("comments", commentsTablePrimaryKeyNameToType, "tenant", "comments_table_rls_policy");
    //...

IMPORTANT!
The primary key columns map is not required to create RLS policy, but it is required when there is any relation between tables (or a single table has a relation to itself) where there is a foreign key constraint. In such a case, posmulten needs to know the primary key table definition to create the correct function that checks if a table row exists for a specific tenant. Such a function is created only if there is any foreign key constraint declaration.

Setting RLS Policy for a table without primary key

In case when RLS policy has to be created for a table without a primary key, for example, for join table in many to many relations. The empty map has to be passed as a method argument.

For example, for the below SQL table:

CREATE TABLE public.users_groups
(
    user_id bigint NOT NULL,
    group_id uuid NOT NULL,
    tenant_id character varying(255),
    CONSTRAINT fk_users_groups_user_id FOREIGN KEY (user_id)
        REFERENCES users (id) MATCH SIMPLE,
    CONSTRAINT fk_users_groups_group_id FOREIGN KEY (group_id)
            REFERENCES groups (uuid) MATCH SIMPLE
)
WITH (
    OIDS = FALSE
)
TABLESPACE pg_default;

the RLS policy declaration should look like this:

    //...
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users_groups", new HashMap<>(), "tenant_id", "users_groups_table_rls_policy");
    //...

INFORMATION! There is a task whose goal is to add the ability to pass a null value as a map to give the same result.

Force RLS Policy for table owner

In situation when RLS policy has to be created for database user that is a tables owner in schema then there has to additional DDL instruction created for each table. It is because just like is mentioned in Postgres documentation Table owners normally bypass row security as well, though a table owner can choose to be subject to row security with ALTER TABLE ... FORCE ROW LEVEL SECURITY. To specify this option builder has method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setForceRowLevelSecurityForTableOwner(boolean forceRowLevelSecurityForTableOwner)

For example, for below requirements:

import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
//...
    Map<String, String> usersTablePrimaryKeyNameToType = new HashMap();
    usersTablePrimaryKeyNameToType.put("id", "bigint");
    Map<String, String> postsTablePrimaryKeyNameToType = new HashMap();
    postsTablePrimaryKeyNameToType.put("id", "bigint");
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null);
    defaultSharedSchemaContextBuilder.setForceRowLevelSecurityForTableOwner(true);
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", usersTablePrimaryKeyNameToType, "tenant_id", "users_table_rls_policy");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("posts", postsTablePrimaryKeyNameToType, "tenant_id", "posts_table_rls_policy");
    //... other criteria

builder will produce

ALTER TABLE "users" FORCE ROW LEVEL SECURITY;
ALTER TABLE "posts" FORCE ROW LEVEL SECURITY;

Adding a foreign key constraint

The builder can create an additional constraint that checks if foreign key value references to the table row that belongs to the current tenant.

createSameTenantConstraintForForeignKey(String mainTable, String foreignKeyTable, Map<String, String> foreignKeyPrimaryKeyColumnsMappings, String constraintName)

mainTable - (Required) table name that has foreign key columns.
foreignKeyTable - (Required) table name that is in relation.
foreignKeyPrimaryKeyColumnsMappings - (Required) map of foreign key columns and primary key columns from table which is in relation. The foreign key column name is the map key and the column name from relation table is its value.
constraintName - (Required) constraint name.

For example, assuming that we have two tables, "users" and "posts":

CREATE TABLE public.users
(
    id bigint NOT NULL,
    name character varying(255),
    tenant_id character varying(255),
    CONSTRAINT users_pkey PRIMARY KEY (id)
)
WITH (
    OIDS = FALSE
)
TABLESPACE pg_default;

CREATE TABLE public.posts
(
    id bigint NOT NULL,
    text text NOT NULL,
    user_id bigint NOT NULL,
    tenant_id character varying(255),
    CONSTRAINT fk_posts_user_id FOREIGN KEY (user_id)
              REFERENCES users (id) MATCH SIMPLE,
    CONSTRAINT posts_pkey PRIMARY KEY (id)
)
WITH (
    OIDS = FALSE
)
TABLESPACE pg_default;

for below criteria:

import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
//...
    Map<String, String> usersTablePrimaryKeyNameToType = new HashMap();
    usersTablePrimaryKeyNameToType.put("id", "bigint");
    Map<String, String> postsTablePrimaryKeyNameToType = new HashMap();
    postsTablePrimaryKeyNameToType.put("id", "bigint");
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null);
    defaultSharedSchemaContextBuilder.setGrantee("db_user");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", usersTablePrimaryKeyNameToType, "tenant_id", "users_table_rls_policy");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("posts", postsTablePrimaryKeyNameToType, "tenant_id", "posts_table_rls_policy");
    //... foreign key constraint declaration
    Map<String, String> foreignKeyColumnToPrimaryKeyColumn = new HashMap();
    foreignKeyColumnToPrimaryKeyColumn.put("user_id", "id");
    defaultSharedSchemaContextBuilder.createSameTenantConstraintForForeignKey("posts", "users", foreignKeyColumnToPrimaryKeyColumn, "posts_users_fk_cu");

    // setting name for function
    defaultSharedSchemaContextBuilder.setNameForFunctionThatChecksIfRecordExistsInTable("users", "is_user_belongs_to_current_tenant");

posmulten is going to produce the below statements related to foreign key constraint:

CREATE OR REPLACE FUNCTION is_user_belongs_to_current_tenant(bigint) RETURNS BOOLEAN AS $$
SELECT EXISTS (
	SELECT 1 FROM users rt WHERE rt.id = $1 AND rt.tenant_id = get_current_tenant_id()
)
$$ LANGUAGE sql
STABLE
PARALLEL SAFE;

ALTER TABLE "posts" ADD CONSTRAINT posts_users_fk_cu CHECK ((user_id IS NULL) OR (is_user_belongs_to_current_tenant(user_id)));

As was mentioned in previous sections, posmulten creates a function that checks if a row with a specified primary key exists for the current tenant. For this example, the is_user_belongs_to_current_tenant function was created because table posts have a foreign key column that references table users. At this moment, the name for such function has to be specified; otherwise, the builder can throw an exception.

Adding a foreign key constraint with a multi-column primary key

Below there is an example of how to specify a foreign key constraint when the key has many columns. The comments table has a primary key with two columns.

CREATE TABLE public.comments
(
id int NOT NULL,
user_id bigint NOT NULL,
text text NOT NULL,
tenant character varying(255),

parent_comment_id int,
parent_comment_user_id bigint,

CONSTRAINT fk_comments_users_id FOREIGN KEY (user_id)
REFERENCES public.users (id) MATCH SIMPLE,

CONSTRAINT fk_comments_parent_id FOREIGN KEY (parent_comment_id, parent_comment_user_id)
REFERENCES public.comments (id, user_id) MATCH SIMPLE,

CONSTRAINT comments_pkey PRIMARY KEY (id, user_id)
)
WITH (
OIDS = FALSE
)
TABLESPACE pg_default;

The table has a relation to itself (the comment has its parent). Below is an example of how to define foreign key constraints with the builder component.

import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
//...

    Map<String, String> commentsTablePrimaryKeyNameToType = new HashMap();
    commentsTablePrimaryKeyNameToType.put("id", "int");
    commentsTablePrimaryKeyNameToType.put("user_id", "bigint");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("comments", commentsTablePrimaryKeyNameToType, "tenant", "comments_table_rls_policy");
    Map<String, String> foreignKeyColumnToPrimaryKeyColumn = new HashMap();
    foreignKeyColumnToPrimaryKeyColumn.put("parent_comment_id", "id");
    foreignKeyColumnToPrimaryKeyColumn.put("parent_comment_user_id", "user_id");
    defaultSharedSchemaContextBuilder.createSameTenantConstraintForForeignKey("comments", "comments", foreignKeyColumnToPrimaryKeyColumn, "comments_parent_comments_fk_cu");

the builder will produce the below statements:

CREATE OR REPLACE FUNCTION is_comment_belongs_to_current_tenant(bigint, int) RETURNS BOOLEAN AS $$
SELECT EXISTS (
	SELECT 1 FROM comments rt WHERE rt.user_id = $1 AND rt.id = $2 AND rt.tenant = get_current_tenant_id()
)
$$ LANGUAGE sql
STABLE
PARALLEL SAFE;
--
ALTER TABLE "comments" ADD CONSTRAINT comments_parent_comments_fk_cu CHECK ((parent_comment_id IS NULL OR parent_comment_user_id IS NULL) OR (is_comment_belongs_to_current_tenant(parent_comment_user_id, parent_comment_id)));

Setting of type for tenant identifier value

By default, the builder assumes that the tenant column type is going to be VARCHAR(255). This also the type for parameters of a few function:

It is also return type for function function that returns the current tenant identifier. The method that set this type is:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setCurrentTenantIdPropertyType(String currentTenantIdPropertyType

For example, if the value is going to be set to the "UUID" then the builder is going to produce a statement like this:

CREATE OR REPLACE FUNCTION get_current_tenant_id() RETURNS UUID AS $$
SELECT current_setting('c.c_ten')
$$ LANGUAGE sql
STABLE
PARALLEL SAFE;

CREATE OR REPLACE FUNCTION set_current_tenant_id(UUID) RETURNS VOID AS $$
BEGIN
PERFORM set_config('c.c_ten', $1, false);
END
$$ LANGUAGE plpgsql
VOLATILE;

Setting the property name that stores tenant identifier value

By default builder use property name "posmulten.tenant_id" to set current tenant identifier. This property is used in function that set current tenant identifier or gets its value.

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setCurrentTenantIdProperty(String currentTenantIdProperty)

Adding default value for tenant column

The builder can create statements that add default values statements for the tenant column in each table. By default, the builder does not do that. To specify behavior builder use method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setCurrentTenantIdentifierAsDefaultValueForTenantColumnInAllTables(boolean value)

For example, for database tables below:

CREATE TABLE public.users
(
    id bigint NOT NULL,
    name character varying(255),
    tenant_id character varying(255),
    CONSTRAINT users_pkey PRIMARY KEY (id)
)
WITH (
    OIDS = FALSE
)
TABLESPACE pg_default;

CREATE TABLE public.posts
(
    id bigint NOT NULL,
    text text NOT NULL,
    user_id bigint NOT NULL,
    tenant_id character varying(255),
    CONSTRAINT fk_posts_user_id FOREIGN KEY (user_id)
              REFERENCES users (id) MATCH SIMPLE,
    CONSTRAINT posts_pkey PRIMARY KEY (id)
)
WITH (
    OIDS = FALSE
)
TABLESPACE pg_default;

and builder criteria:

import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
    Map<String, String> usersTablePrimaryKeyNameToType = new HashMap();
    usersTablePrimaryKeyNameToType.put("id", "bigint");
    Map<String, String> postsTablePrimaryKeyNameToType = new HashMap();
    postsTablePrimaryKeyNameToType.put("id", "bigint");
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null);
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", usersTablePrimaryKeyNameToType, "tenant_id", "users_table_rls_policy");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("posts", postsTablePrimaryKeyNameToType, "tenant_id", "posts_table_rls_policy");
//... other criteria
    defaultSharedSchemaContextBuilder.setCurrentTenantIdentifierAsDefaultValueForTenantColumnInAllTables(true);

the builder will produce below statements:

ALTER TABLE users ALTER COLUMN tenant_id SET DEFAULT get_current_tenant_id();
ALTER TABLE posts ALTER COLUMN tenant_id SET DEFAULT get_current_tenant_id();

Skipping adding default value for tenant column for a single table

The builder has a method that allows skipping the creation of the default value statements for a specific table when the option of adding default value for all tenant columns is on.

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#skipAddingOfTenantColumnDefaultValueForTable(String value)

For below criteria:

import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
    Map<String, String> usersTablePrimaryKeyNameToType = new HashMap();
    usersTablePrimaryKeyNameToType.put("id", "bigint");
    Map<String, String> postsTablePrimaryKeyNameToType = new HashMap();
    postsTablePrimaryKeyNameToType.put("id", "bigint");
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null);
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", usersTablePrimaryKeyNameToType, "tenant_id", "users_table_rls_policy");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("posts", postsTablePrimaryKeyNameToType, "tenant_id", "posts_table_rls_policy");
//... other criteria
    defaultSharedSchemaContextBuilder.setCurrentTenantIdentifierAsDefaultValueForTenantColumnInAllTables(true);
    defaultSharedSchemaContextBuilder.skipAddingOfTenantColumnDefaultValueForTable("posts");

the builder will produce below statements:

ALTER TABLE users ALTER COLUMN tenant_id SET DEFAULT get_current_tenant_id();

builder will not produce a statement for the posts table.

Adding tenant column to tenant table

The builder can produce a statement that creates a tenant column for a table that does not have it.

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#createTenantColumnForTable(String table)

For below criteria:

import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
    Map<String, String> notificationsTablePrimaryKeyNameToType = new HashMap();
    notificationsTablePrimaryKeyNameToType.put("uuid", "uuid");
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null);
    defaultSharedSchemaContextBuilder.createTenantColumnForTable("notifications")
        .createRLSPolicyForTable("notifications", notificationsTablePrimaryKeyNameToType, "tenant_x", "notifications_table_rls_policy")
//... other criteria

the builder will produce below statements:

ALTER TABLE notifications ADD COLUMN tenant_x VARCHAR(255);
ALTER TABLE notifications ALTER COLUMN tenant_x SET NOT NULL;
ALTER TABLE notifications ALTER COLUMN tenant_x SET DEFAULT get_current_tenant_id();

Setting default tenant column name

By default tenant, the builder assumes that tenant column name is "tenant_id". It is important during adding tenant column. Or specifying RLS policy for a table without specifying a tenant column name. In such a case, the builder assumes that the tenant column for such a table is equal to the default value. The builder allows to change default tenant column name via method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setDefaultTenantIdColumn(String defaultTenantIdColumn)

Setting function name that returns the current tenant identifier

The builder allows to set the name of function that returns the current tenant identifier via method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setGetCurrentTenantIdFunctionName(String getCurrentTenantIdFunctionName)

Setting function name that sets the current tenant identifier

The builder allows to set the name of function that set the current tenant identifier via method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setSetCurrentTenantIdFunctionName(String setCurrentTenantIdFunctionName)

Setting function name that checks if current tenant has authorities to a table row

The builder allows to set the name of function that checks tenant access to a table row via method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setTenantHasAuthoritiesFunctionName(String tenantHasAuthoritiesFunctionName)

Setting function name that checks if passed identifier is the same as current tenant identifier

The builder allows to set the name of function that checks if the passed identifier is the same as the current tenant identifier via method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setEqualsCurrentTenantIdentifierFunctionName(String equalsCurrentTenantIdentifierFunctionName)

Setting function name that checks if passed primary key for a specific table exists for the current tenant

It is required to specify the function that checks if the passed identifier exists in a specific table. The builder has a single method for this purpose.

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setNameForFunctionThatChecksIfRecordExistsInTable(String recordTable, String functionName)

recordTable - (Required) table name.
functionName - (Required) function name.
TODO

Setting a list of invalid tenant identifier values

The builder allows to specify the list of invalid tenant identifier values via method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#createValidTenantValueConstraint(List<String> tenantValuesBlacklist, String isTenantValidFunctionName, String isTenantValidConstraintName)

tenantValuesBlacklist - (Required) list of invalid tenant identifier values.
isTenantValidFunctionName - (Optional) name of the function that checks if passed tenant identifier is valid. If passed value is null then function name is "is_tenant_identifier_valid".
isTenantValidConstraintName - (Optional) name of the constraint that checks if the tenant column has a valid value. If the passed value is null then the constraint name is "tenant_identifier_valid".

For below criteria:

import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
    Map<String, String> usersTablePrimaryKeyNameToType = new HashMap();
    usersTablePrimaryKeyNameToType.put("id", "bigint");
    Map<String, String> postsTablePrimaryKeyNameToType = new HashMap();
    postsTablePrimaryKeyNameToType.put("id", "bigint");
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null);
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", usersTablePrimaryKeyNameToType, "tenant_id", "users_table_rls_policy");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("posts", postsTablePrimaryKeyNameToType, "tenant_id", "posts_table_rls_policy");
//... other criteria
    defaultSharedSchemaContextBuilder.createValidTenantValueConstraint(asList("DUMMMY_TENANT",  "XXX-INVAlid_tenant"), null, null);

the builder will produce below statements:

CREATE OR REPLACE FUNCTION is_tenant_id_valid(VARCHAR(255)) RETURNS BOOLEAN AS $$
SELECT $1 <> CAST ('DUMMMY_TENANT' AS VARCHAR(255)) AND $1 <> CAST ('XXX-INVAlid_tenant' AS VARCHAR(255))
$$ LANGUAGE sql
IMMUTABLE
PARALLEL SAFE;
ALTER TABLE "users" ADD CONSTRAINT tenant_should_be_valid CHECK (tenant_id IS NULL OR is_tenant_id_valid(tenant_id));
ALTER TABLE "posts" ADD CONSTRAINT tenant_should_be_valid CHECK (tenant_id IS NULL OR is_tenant_id_valid(tenant_id));

Setting custom name for table tenant column constraint

Builder allows specifying custom constraint name via the method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#registerCustomValidTenantValueConstraintNameForTable(String table, String constraintName)

table - (Required) name of table.
constraintName - (Required) name of constraint.

for below criteria:

import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
    Map<String, String> usersTablePrimaryKeyNameToType = new HashMap();
    usersTablePrimaryKeyNameToType.put("id", "bigint");
    Map<String, String> postsTablePrimaryKeyNameToType = new HashMap();
    postsTablePrimaryKeyNameToType.put("id", "bigint");
    DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null);
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", usersTablePrimaryKeyNameToType, "tenant_id", "users_table_rls_policy");
    defaultSharedSchemaContextBuilder.createRLSPolicyForTable("posts", postsTablePrimaryKeyNameToType, "tenant_id", "posts_table_rls_policy");
    defaultSharedSchemaContextBuilder.createValidTenantValueConstraint(asList("DUMMMY_TENANT",  "XXX-INVAlid_tenant"), null, null);
//... other criteria
    defaultSharedSchemaContextBuilder.registerCustomValidTenantValueConstraintNameForTable("posts", "posts_tenant_is_valid");

the builder will produce below statements:

ALTER TABLE "users" ADD CONSTRAINT tenant_should_be_valid CHECK (tenant_id IS NULL OR is_tenant_id_valid(tenant_id));
ALTER TABLE "posts" ADD CONSTRAINT posts_tenant_is_valid CHECK (tenant_id IS NULL OR is_tenant_id_valid(tenant_id));

Setting foreign key constraint where tenant column is part of composite key

There might be a situation when your database model assumes that the tenant column is part of the table's primary key and, that being said, also part of the foreign keys. This means that, for example, all your unique constraint has to cover tenant column value. Having such a database model allows easier migration of tenant data between databases. Below there is example how to specify such shared context object: We have two tables, "users" and "notifications".

        DefaultSharedSchemaContextBuilder defaultSharedSchemaContextBuilder = new DefaultSharedSchemaContextBuilder(null);

        // Setting flag with true value
        defaultSharedSchemaContextBuilder.setCreateForeignKeyConstraintWithTenantColumn(true);
        
        //....
        // We have two tables, "users" and "notifications".
        // Setting tables for which RLS should be created
        defaultSharedSchemaContextBuilder.createRLSPolicyForTable("users", mapBuilder().put(("id", "bigint").build() , "tenant_id", "users_table_rls_policy");
        defaultSharedSchemaContextBuilder.createRLSPolicyForTable("notifications", mapBuilder().put(("uuid", "uuid").build(), "notification_tenant", "notifications_table_rls_policy");
        
        // Setting foreign key constraint declaration with method 'createSameTenantConstraintForForeignKey'
        defaultSharedSchemaContextBuilder.createSameTenantConstraintForForeignKey("notification, "users", mapBuilder().put("user_id", "id").build(), "notification_users_fk");

        // Setting a name for a function that checks if user record (reference table for our foreign key example) exists is not needed
        //defaultSharedSchemaContextBuilder.setNameForFunctionThatChecksIfRecordExistsInTable("users", "is_user_belongs_to_current_tenant");

the builder will produce below statements:

ALTER TABLE IF EXISTS "notifications" ADD CONSTRAINT notification_users_fk FOREIGN KEY (notification_tenant, user_id) REFERENCES "users" (tenant_id, id) MATCH SIMPLE;

Naming convention and its constraints

By default function name can have a length from 1 to 63 characters. SQL definitions validation can be disabled by using method:

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setDisableDefaultSqlDefinitionsValidators(boolean disableDefaultSqlDefinitionsValidators)

it can be also disabled by setting null or empty list by method.

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#setSqlDefinitionsValidators(List<ISQLDefinitionsValidator> sqlDefinitionsValidators)

By using this method there is also the possibility of customization for validation.

Adding custom sql definitions

There is an option to pass custom SQL definition to query builder.

com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#addCustomSQLDefinition(com.github.starnowski.posmulten.postgresql.core.context.CustomSQLDefinitionPairPositionProvider positionProvider, String creationScript)
com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#addCustomSQLDefinition(com.github.starnowski.posmulten.postgresql.core.context.CustomSQLDefinitionPairPositionProvider positionProvider, String creationScript, String dropScript)
com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#addCustomSQLDefinition(com.github.starnowski.posmulten.postgresql.core.context.CustomSQLDefinitionPairPositionProvider positionProvider, String creationScript, String dropScript, List<String> checkingStatements)
com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder#addCustomSQLDefinition(com.github.starnowski.posmulten.postgresql.core.context.CustomSQLDefinitionPairPositionProvider positionProvider, com.github.starnowski.posmulten.postgresql.core.common.SQLDefinition sqlDefinition)

Each method pass object of type com.github.starnowski.posmulten.postgresql.core.context.CustomSQLDefinitionPairPositionProvider that returns position for SQL definition. Default implementation of this type is com.github.starnowski.posmulten.postgresql.core.context.CustomSQLDefinitionPairDefaultPosition enum type that has two values:

AT_BEGINNING - custom definition is being added before all definitions created by builder
AT_END - custom definition is being added after all definitions created by builder

Example:

        ISharedSchemaContext result = (new DefaultSharedSchemaContextBuilder(null))
                .setGrantee(CORE_OWNER_USER)
                .addCustomSQLDefinition(CustomSQLDefinitionPairDefaultPosition.AT_BEGINNING, "ALTER TABLE users ADD COLUMN custom_column1 VARCHAR(255);", "ALTER TABLE users DROP COLUMN custom_column1;",
                        singletonList("SELECT COUNT(1) FROM information_schema.columns WHERE table_catalog = 'postgresql_core' AND table_schema = 'public' AND table_name = 'users' AND column_name = 'custom_column1';"))
                .addCustomSQLDefinition(CustomSQLDefinitionPairDefaultPosition.AT_END, "ALTER TABLE users ADD COLUMN custom_column2 VARCHAR(255);", "ALTER TABLE users DROP COLUMN custom_column2;",
                        singletonList("SELECT COUNT(1) FROM information_schema.columns WHERE table_catalog = 'postgresql_core' AND table_schema = 'public' AND table_name = 'users' AND column_name = 'custom_column2';"));

The method that passes only the creation script or creation and drop script for other missing scripts set "SELECT 1" as the default script. Whether the drop script or scripts, check if the creation script was applied correctly.

Using template variables in context builder

Very often database schema names can be different depending on the environment, the database schema can have a different name for the dev environment than for the production environment. For example schema can have value "{{template_schema_value}}" and db user can have "{{template_user_grantee}}".

CREATE OR REPLACE FUNCTION {{template_schema_value}}.set_tenant(VARCHAR(255)) RETURNS VOID AS $$
BEGIN
PERFORM set_config('pos.c.ten', $1, false);
END
$$ LANGUAGE plpgsql
VOLATILE;

CREATE POLICY comments_table_rls_policy ON {{template_schema_value}}.comments
FOR ALL
TO "{{template_user_grantee}}"
USING ({{template_schema_value}}._tenant_hast_auth(tenant, 'ALL', 'USING', 'comments', '{{template_schema_value}}'))
WITH CHECK ({{template_schema_value}}._tenant_hast_auth(tenant, 'ALL', 'WITH_CHECK', 'comments', '{{template_schema_value}}'));
...

Such generated statements can be used with Liquibase or Flyway. Both those tools allow specifying configuration variables that be injected into scripts before executing them. Posmulten also has decorator types that can replace template variables with passed values. Such a feature can be useful when there is a need to run generated scripts during application tests. The below code demonstrates it:

import com.github.starnowski.posmulten.configuration.DefaultSharedSchemaContextBuilderFactoryResolver;
import com.github.starnowski.posmulten.configuration.NoDefaultSharedSchemaContextBuilderFactorySupplierException;
import com.github.starnowski.posmulten.configuration.core.context.IDefaultSharedSchemaContextBuilderFactory;
import com.github.starnowski.posmulten.configuration.core.exceptions.InvalidConfigurationException;
import com.github.starnowski.posmulten.postgresql.core.common.SQLDefinition;
import com.github.starnowski.posmulten.postgresql.core.context.DefaultSharedSchemaContextBuilder;
import com.github.starnowski.posmulten.postgresql.core.context.ISharedSchemaContext;
import com.github.starnowski.posmulten.postgresql.core.context.decorator.DefaultDecoratorContext;
import com.github.starnowski.posmulten.postgresql.core.context.decorator.SharedSchemaContextDecoratorFactory;
import com.github.starnowski.posmulten.postgresql.core.context.exceptions.SharedSchemaContextBuilderException;
import com.github.starnowski.posmulten.postgresql.core.db.DatabaseOperationExecutor;
import com.github.starnowski.posmulten.postgresql.core.db.operations.exceptions.ValidationDatabaseOperationsException;
import com.github.starnowski.posmulten.postgresql.core.rls.function.ISetCurrentTenantIdFunctionInvocationFactory;
import com.github.starnowski.posmulten.postgresql.test.utils.MapBuilder;

import javax.sql.DataSource;
import java.net.URISyntaxException;
import java.nio.file.Paths;
/...


    public void createSQLDefinitions() throws SharedSchemaContextBuilderException, NoDefaultSharedSchemaContextBuilderFactorySupplierException, InvalidConfigurationException, URISyntaxException, ValidationDatabaseOperationsException, SQLException {
        DefaultSharedSchemaContextBuilderFactoryResolver defaultSharedSchemaContextBuilderFactoryResolver = new DefaultSharedSchemaContextBuilderFactoryResolver();
        IDefaultSharedSchemaContextBuilderFactory factory = defaultSharedSchemaContextBuilderFactoryResolver.resolve(INTEGRATION_CONFIGURATION_TEST_FILE_PATH);
        DefaultSharedSchemaContextBuilder builder = factory.build(resolveFilePath(INTEGRATION_CONFIGURATION_TEST_FILE_PATH));
        DefaultDecoratorContext decoratorContext = DefaultDecoratorContext.builder()
                .withReplaceCharactersMap(MapBuilder.mapBuilder()
                        .put("{{template_schema_value}}", "non_public_schema")
                        .put("{{template_user_grantee}}", CORE_OWNER_USER)
                        .build())
                .build();
        ISharedSchemaContext defaultContext = builder.build();
        // Display original SQL statements
        this.databaseOperationExecutor.execute(dataSource, defaultContext.getSqlDefinitions(), LOG_ALL);
        SharedSchemaContextDecoratorFactory sharedSchemaContextDecoratorFactory = new SharedSchemaContextDecoratorFactory();
        sharedSchemaContext = sharedSchemaContextDecoratorFactory.build(defaultContext, decoratorContext);
        setCurrentTenantIdFunctionInvocationFactory = sharedSchemaContext.getISetCurrentTenantIdFunctionInvocationFactory();
        sqlDefinitions.addAll(sharedSchemaContext.getSqlDefinitions());
    }


    public void executeSQLDefinitions() {
            try {
            this.databaseOperationExecutor.execute(dataSource, sharedSchemaContext.getSqlDefinitions(), CREATE);
            this.databaseOperationExecutor.execute(dataSource, sharedSchemaContext.getSqlDefinitions(), LOG_ALL);
            this.databaseOperationExecutor.execute(dataSource, sharedSchemaContext.getSqlDefinitions(), VALIDATE);
            } catch (SQLException e) {
            throw new RuntimeException(e);
            } catch (ValidationDatabaseOperationsException e) {
            throw new RuntimeException(e);
            }
            }

Library will generate SQL statements with replaced values:

CREATE OR REPLACE FUNCTION non_public_schema.get_ten_id() RETURNS VARCHAR(255) AS $$
SELECT current_setting('pos.c.ten')
$$ LANGUAGE sql
STABLE
PARALLEL SAFE;

CREATE POLICY comments_table_rls_policy ON non_public_schema.comments
FOR ALL
TO "postgresql-core-owner"
USING (non_public_schema._tenant_hast_auth(tenant, 'ALL', 'USING', 'comments', 'non_public_schema'))
WITH CHECK (non_public_schema._tenant_hast_auth(tenant, 'ALL', 'WITH_CHECK', 'comments', 'non_public_schema'));
...

Please be in mind that above code use configuration and configuration-yaml-interpreter modules. Where configuration for th Posmulten library is store in Yaml file. However, you can also pass template values as parameters to DefaultSharedSchemaContextBuilder methods.

GUI

There is a GUI application implemented in the Swing library. For more details, please check the openwebstart module.

Reporting issues

Project contribution

  • Look for open issues or create your own
  • Fork repository on Github and start applying your changes to master branch or release branch
  • Follow CONTRIBUTING.md document for coding rules
  • Create pull request

About

Posmulten library is an open-source project for the generation of SQL DDL statements that make it easy for implementation of Shared Schema Multi-tenancy strategy via the Row Security Policies in the Postgres database.

Topics

Resources

License

LGPL-2.1, Unknown licenses found

Licenses found

LGPL-2.1
LICENSE
Unknown
license.txt

Stars

Watchers

Forks

Sponsor this project

 

Packages

No packages published

Contributors 3

  •  
  •  
  •