mirror of
https://github.com/storytold/artcraft.git
synced 2026-10-25 18:49:38 +00:00
43 lines
1.3 KiB
SQL
43 lines
1.3 KiB
SQL
-- noinspection SqlDialectInspectionForFile
|
|
-- noinspection SqlNoDataSourceInspectionForFile
|
|
-- noinspection SqlResolveForFile
|
|
|
|
-- Batch generations
|
|
CREATE TABLE batch_generations (
|
|
-- Not used for anything except replication.
|
|
id BIGINT(20) NOT NULL AUTO_INCREMENT,
|
|
|
|
-- The batch token
|
|
-- This is **NOT** a unique token.
|
|
token VARCHAR(32) NOT NULL,
|
|
|
|
-- ========== ENTITY LINKAGE ==========
|
|
|
|
-- The type of record this is attached to
|
|
-- Currently supported values:
|
|
-- * 'model_weight'
|
|
-- * 'media_file'
|
|
-- * 'comment'
|
|
entity_type VARCHAR(32) NOT NULL,
|
|
|
|
-- The token for the record.
|
|
entity_token VARCHAR(32) NOT NULL,
|
|
|
|
-- ========== TIMESTAMPS ==========
|
|
|
|
created_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP,
|
|
updated_at TIMESTAMP NOT NULL DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP,
|
|
|
|
-- INDICES --
|
|
PRIMARY KEY (id),
|
|
UNIQUE KEY (token, entity_type, entity_token),
|
|
KEY index_token (token),
|
|
KEY index_entity_token (entity_token)
|
|
|
|
-- NB: Realistically, we probably don't need this index until we have a lot of entity types, and the entity_token
|
|
-- entropy is probably sufficient enough as-is.
|
|
-- KEY index_entity_type_and_token (entity_type, entity_token),
|
|
|
|
|
|
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4 COLLATE=utf8mb4_bin;
|