[code notes] implementation of creating partition table

SQL

CREATE TABLE measurement (
    city_id     int not null,
    logdate     date not null,
    data        text
) PARTITION BY RANGE (logdate);

CREATE TABLE p1 PARTITION OF measurement
    FOR VALUES FROM ('2020-01-01') TO ('2020-02-01');

Semantic Struct

CreateStmt *n = makeNode(CreateStmt);
n->relation = $4; /* `p1` here */
n->inhRelations = list_make1($7); /* `measurement` here */
n->partbound = $9; /* ('2020-01-01') TO ('2020-02-01') */
/* other attributes are not important */

Execution

DDL statements are handled by function ProcessUtilitySlow. For T_CreateStmt:

• run the parse analysis through transformCreateStmt
• After transforming, the create table statment could have three types:
  • CreateStmt
  • CreateForeignTableStmt
  • TableLikeClause
    we are focusing on CreateStmt.
• Create the table itself by function DefineRelation
• record a ddl_command_end event
• parse and validate reloptions for the toast table
• check and create toast table

DefineRelation

• Look up the namespace in which we are supposed to create the relation, check we have permission to create there, lock it against concurrent drop, and mark stmt->relation as RELPERSISTENCE_TEMP if a temporary namespace is selected.
• record the list of OIDs of the parents and lock the parents
• Select tablespace to use
• Check permissions except when using database's default
• In all cases disallow placing user relations in pg_global
• Identify user ID that will own the table
• Parse and validate reloptions, if any.
• Look up inheritance ancestors and generate relation schema, including inherited attributes.
• Create a tuple descriptor from the relation schema.
• Find columns with default values and prepare for insertion of the defaults.
• If the statement hasn't specified an access method, but we're defining a type of relation that needs one, use the default.
• Create the relation by function heap_create_with_catalog.
• make the newly created relation visible by function CommandCounterIncrement
• Now add any newly specified column default and generation expressions to the new relation.
• Make column generation expressions visible for use by partitioning.
• lock default partition table with AccessExclusiveLock if exists. Here's the reason.
• Transform the bound values by function transformPartitionBound
• Check first that the new partition's bound is valid and does not overlap with any of existing partitions of the parent.
• If the default partition exists, checks if there exists a row in the default partition that would properly belong to the new partition being added. If it finds one, it throws an error.
• store the partition bound in the pg_class entry
• Updates the system catalogs with proper inheritance information.
• If we're creating a partition, create now all the indexes, triggers, FKs defined in the parent.
• indexes are created by DefineIndex, triggers are cloned by CloneRowTriggersToPartition, foreign keys ared cloned by CloneForeignKeyConstraints
• Now add any newly specified CHECK constraints to the new relation by AddRelationNewConstraints
• Finally, merge the not-null constraints that are declared directly with those that come from parent relations