require 'forwardable' module Partitioned class PartitionedBase # # SqlAdapter # manages requests of partitioned tables. # class RedshiftSqlAdapter attr_reader :parent_table_class def initialize(parent_table_class) @parent_table_class = parent_table_class end # # Ensure our function for warning about improper partition usage is in place. # # Name: always_fail_on_insert(text); Type: FUNCTION; Schema: public # # Used to raise an exception explaining why a specific insert (into a parent # table which should never have records) should never be attempted. # def ensure_always_fail_on_insert_exists # XXX nothing can be done here end # # Child tables whose parent table is 'foos', typically exist in a schema named foos_partitions. # # *partition_key_values are needed here to support the use of multiple schemas to keep tables in. # def create_partition_schema(*partition_key_values) create_schema(configurator.schema_name, :unless_exists => true) end # # Does a specific child partition exist. # def partition_exists?(*partition_key_values) return find(:first, :from => "pg_tables", :select => "count(*) as count", :conditions => ["schemaname = ? and tablename = ?", configurator.schema_name, configurator.part_name(*partition_key_values) ]).count.to_i == 1 end # # Returns an array of partition table names from last to first limited to # the number of entries requested by its first parameter. # # The magic here is in the overridden method "last_n_partitions_order_by_clause" # which is designed to order a list of partition table names (table names without # their schema name) from last to first. # # If the child table names are the format "pYYYYMMDD" where YYYY is a four digit year, MM is # a month number and DD is a day number, you would use the following to order from last to # first: # tablename desc # # For child table names of the format "pXXXX" where XXXX is a number, you may want something like: # substring(tablename, 2)::integer desc # # For clarity, the sql executed is: # select tablename from pg_tables where schemaname = $1 order by $2 limit $3 # where: # $1 = the name of schema (foos_partitions) # $2 = the order by clause that would make the greatest table name listed first # $3 = the parameter 'how_many' # def last_n_partition_names(how_many = 1) return find(:all, :from => "pg_tables", :select => :tablename, :conditions => ["schemaname = ?", configurator.schema_name], :order => last_n_partitions_order_by_clause, :limit => how_many).map(&:tablename) end # # Override this or order the tables from last (greatest value? greatest date?) to first. # def last_n_partitions_order_by_clause return configurator.last_partitions_order_by_clause end # # Used to create the parent table rule to ensure. # # This will cause an error on attempt to insert into the parent table. # # We want all records to exist in one of the child tables so the # query planner can optimize access to the records. # def add_parent_table_rules(*partition_key_values) # XXX nothing can be done here end # # The name of the table (schemaname.childtablename) given the check constraint values. # def partition_table_name(*partition_key_values) return configurator.table_name(*partition_key_values) end # # A reasonable alias for the partition table # def partition_table_alias_name(*partition_key_values) return configurator.table_alias_name(*partition_key_values) end # # Create a single child table. # def create_partition_table(*partition_key_values) options = { :partitioned_model => @parent_table_class, :temporary => false, :table_name => configurator.part_name(*partition_key_values) } @redshift_table_creator = ::ActiverecordRedshift::TableManager.new(@parent_table_class.connection, options) @redshift_table_creator.duplicate_table end # # Remove a specific single child table. # def drop_partition_table(*partition_key_values) drop_table(configurator.table_name(*partition_key_values)) end # # Add indexes that must exist on child tables. Only leaf child tables # need indexes as parent table indexes are not used in postgres. # def add_partition_table_index(*partition_key_values) configurator.indexes(*partition_key_values).each do |field,options| used_options = options.clone unless used_options.has_key?(:name) name = [*field].join('_') used_options[:name] = used_options[:unique] ? unique_index_name(name, *partition_key_values) : index_name(name, *partition_key_values) end add_index(partition_table_name(*partition_key_values), field, used_options) end end # # Used when creating the name of a SQL rule. # def parent_table_rule_name(name, suffix = "rule", *partition_key_values) return "#{configurator.table_name(*partition_key_values).gsub(/[.]/, '_')}_#{name}_#{suffix}" end # # Used to create index names. # def index_name(name, *partition_key_values) return "#{configurator.part_name(*partition_key_values)}_#{name}_idx" end # # Used to create index names. # def unique_index_name(name, *partition_key_values) return "#{configurator.part_name(*partition_key_values)}_#{name}_udx" end # # This is here for derived classes to set up references to added columns # (or columns in the parent that need foreign key constraints). # # Foreign keys are not inherited in postgres. So, a parent table # of the form: # # -- this is the referenced table # create table companies # ( # id serial not null primary key, # created_at timestamp not null default now(), # updated_at timestamp, # name text not null # ); # # -- this is the parent table # create table employees # ( # id serial not null primary key, # created_at timestamp not null default now(), # updated_at timestamp, # name text not null, # company_id integer not null references companies, # supervisor_id integer not null references employees # ); # # -- some children # create table employees_of_company_1 ( CHECK ( company_id = 1 ) ) INHERITS (employees); # create table employees_of_company_2 ( CHECK ( company_id = 2 ) ) INHERITS (employees); # create table employees_of_company_3 ( CHECK ( company_id = 3 ) ) INHERITS (employees); # # Since postgres does not inherit referential integrity from parent tables, the following # insert will work: # insert into employees_of_company_1 (name, company_id, supervisor_id) values ('joe', 1, 10); # even if there is no record in companies with id = 1 and there is no record in employees with id = 10 # # For proper referential integrity handling you must do the following: # ALTER TABLE employees_of_company_1 add foreign key (company_id) references companies(id) # ALTER TABLE employees_of_company_2 add foreign key (company_id) references companies(id) # ALTER TABLE employees_of_company_3 add foreign key (company_id) references companies(id) # # ALTER TABLE employees_of_company_1 add foreign key (supervisor_id) references employees_of_company_1(id) # ALTER TABLE employees_of_company_2 add foreign key (supervisor_id) references employees_of_company_2(id) # ALTER TABLE employees_of_company_3 add foreign key (supervisor_id) references employees_of_company_3(id) # # The second set of alter tables brings up a good another consideration about postgres references and partitions. # postgres will not follow references to a child table. So, a foreign key reference to "employees" in this # set of alter statements would not work because postgres would expect the table "employees" to have # the specific referenced record, but the record really exists in a child of employees. So, the alter statement # forces the reference check on the specific child table we know must contain this employees supervisor (since # such a supervisor would have to work for the same company in our model). # def add_references_to_partition_table(*partition_key_values) configurator.foreign_keys(*partition_key_values).each do |foreign_key| add_foreign_key(partition_table_name(*partition_key_values), foreign_key.referencing_field, foreign_key.referenced_table, foreign_key.referenced_field) end end ## # :method: connection # delegated to the connection of the parent table class ## # :method: execute # delegated to the connection of the parent table class ## # :method: create_schema # delegated to the connection of the parent table class ## # :method: drop_schema # delegated to the connection of the parent table class ## # :method: add_index # delegated to the connection of the parent table class ## # :method: remove_index # delegated to the connection of the parent table class ## # :method: transaction # delegated to the connection of the parent table class ## # :method: find_by_sql # delegated to the connection of the parent table class ## # :method: find # delegated to the connection of the parent table class extend Forwardable def_delegators :parent_table_class, :connection, :find_by_sql, :transaction, :find, :configurator def_delegators :connection, :execute, :add_index, :remove_index, :create_schema, :drop_schema, :add_foreign_key, :create_table, :drop_table end end end