Minor updates, and docs, for schema delta files

changelog.d/11823.misc

@@ -0,0 +1 @@
+Minor updates and documentation for database schema delta files.

docs/development/database_schema.md

@@ -96,6 +96,60 @@ Ensure postgres is installed, then run:
 NB at the time of writing, this script predates the split into separate `state`/`main`
 databases so will require updates to handle that correctly.
 
+## Delta files
+
+Delta files define the steps required to upgrade the database from an earlier version.
+They can be written as either a file containing a series of SQL statements, or a Python
+module.
+
+Synapse remembers which delta files it has applied to a database (they are stored in the
+`applied_schema_deltas` table) and will not re-apply them (even if a given file is
+subsequently updated).
+
+Delta files should be placed in a directory named `synapse/storage/schema/<database>/delta/<version>/`.
+They are applied in alphanumeric order, so  by convention the first two characters
+of the filename should be an integer such as `01`, to put the file in the right order.
+
+### SQL delta files
+
+These should be named `*.sql`, or —  for changes which should only be applied for a
+given database engine — `*.sql.posgres` or `*.sql.sqlite`. For example, a delta which
+adds a new column to the `foo` table might be called `01add_bar_to_foo.sql`.
+
+Note that our SQL parser is a bit simple - it understands comments (`--` and `/*...*/`),
+but complex statements which require a `;` in the middle of them (such as `CREATE
+TRIGGER`) are beyond it and you'll have to use a Python delta file.
+
+### Python delta files
+
+For more flexibility, a delta file can take the form of a python module. These should
+be named `*.py`. Note that database-engine-specific modules are not supported here –
+instead you can write `if isinstance(database_engine, PostgresEngine)` or similar.
+
+A Python delta module should define either or both of the following functions:
+
+```python
+import synapse.config.homeserver
+import synapse.storage.engines
+import synapse.storage.types
+
+
+def run_create(
+    cur: synapse.storage.types.Cursor,
+    database_engine: synapse.storage.engines.BaseDatabaseEngine,
+) -> None:
+    """Called whenever an existing or new database is to be upgraded"""
+    ...
+
+def run_upgrade(
+    cur: synapse.storage.types.Cursor,
+    database_engine: synapse.storage.engines.BaseDatabaseEngine,
+    config: synapse.config.homeserver.HomeServerConfig,
+) -> None:
+    """Called whenever an existing database is to be upgraded."""
+    ...
+```
+
 ## Boolean columns
 
 Boolean columns require special treatment, since SQLite treats booleans the

synapse/storage/prepare_database.py

@@ -499,9 +499,12 @@ def _upgrade_existing_database(
                 module = importlib.util.module_from_spec(spec)
                 spec.loader.exec_module(module)  # type: ignore
 
-                logger.info("Running script %s", relative_path)
-                module.run_create(cur, database_engine)  # type: ignore
-                if not is_empty:
+                if hasattr(module, "run_create"):
+                    logger.info("Running %s:run_create", relative_path)
+                    module.run_create(cur, database_engine)  # type: ignore
+
+                if not is_empty and hasattr(module, "run_upgrade"):
+                    logger.info("Running %s:run_upgrade", relative_path)
                     module.run_upgrade(cur, database_engine, config=config)  # type: ignore
             elif ext == ".pyc" or file_name == "__pycache__":
                 # Sometimes .pyc files turn up anyway even though we've