summary refs log tree commit diff
diff options
context:
space:
mode:
-rw-r--r--changelog.d/13195.misc1
-rw-r--r--docs/postgres.md8
-rwxr-xr-xsynapse/_scripts/synapse_port_db.py34
3 files changed, 43 insertions, 0 deletions
diff --git a/changelog.d/13195.misc b/changelog.d/13195.misc
new file mode 100644
index 0000000000..5506f767b3
--- /dev/null
+++ b/changelog.d/13195.misc
@@ -0,0 +1 @@
+Check that `auto_vacuum` is disabled when porting a SQLite database to Postgres, as `VACUUM`s must not be performed between runs of the script.
\ No newline at end of file
diff --git a/docs/postgres.md b/docs/postgres.md
index cbc32e1836..f2519f6b0a 100644
--- a/docs/postgres.md
+++ b/docs/postgres.md
@@ -143,6 +143,14 @@ to do step 2.
 
 It is safe to at any time kill the port script and restart it.
 
+However, under no circumstances should the SQLite database be `VACUUM`ed between
+multiple runs of the script. Doing so can lead to an inconsistent copy of your database
+into Postgres.
+To avoid accidental error, the script will check that SQLite's `auto_vacuum` mechanism
+is disabled, but the script is not able to protect against a manual `VACUUM` operation
+performed either by the administrator or by any automated task that the administrator
+may have configured.
+
 Note that the database may take up significantly more (25% - 100% more)
 space on disk after porting to Postgres.
 
diff --git a/synapse/_scripts/synapse_port_db.py b/synapse/_scripts/synapse_port_db.py
index d3b4887f69..642fd41629 100755
--- a/synapse/_scripts/synapse_port_db.py
+++ b/synapse/_scripts/synapse_port_db.py
@@ -621,6 +621,25 @@ class Porter:
                 self.postgres_store.db_pool.updates.has_completed_background_updates()
             )
 
+    @staticmethod
+    def _is_sqlite_autovacuum_enabled(txn: LoggingTransaction) -> bool:
+        """
+        Returns true if auto_vacuum is enabled in SQLite.
+        https://www.sqlite.org/pragma.html#pragma_auto_vacuum
+
+        Vacuuming changes the rowids on rows in the database.
+        Auto-vacuuming is therefore dangerous when used in conjunction with this script.
+
+        Note that the auto_vacuum setting can't be changed without performing
+        a VACUUM after trying to change the pragma.
+        """
+        txn.execute("PRAGMA auto_vacuum")
+        row = txn.fetchone()
+        assert row is not None, "`PRAGMA auto_vacuum` did not give a row."
+        (autovacuum_setting,) = row
+        # 0 means off. 1 means full. 2 means incremental.
+        return autovacuum_setting != 0
+
     async def run(self) -> None:
         """Ports the SQLite database to a PostgreSQL database.
 
@@ -637,6 +656,21 @@ class Porter:
                 allow_outdated_version=True,
             )
 
+            # For safety, ensure auto_vacuums are disabled.
+            if await self.sqlite_store.db_pool.runInteraction(
+                "is_sqlite_autovacuum_enabled", self._is_sqlite_autovacuum_enabled
+            ):
+                end_error = (
+                    "auto_vacuum is enabled in the SQLite database."
+                    " (This is not the default configuration.)\n"
+                    " This script relies on rowids being consistent and must not"
+                    " be used if the database could be vacuumed between re-runs.\n"
+                    " To disable auto_vacuum, you need to stop Synapse and run the following SQL:\n"
+                    " PRAGMA auto_vacuum=off;\n"
+                    " VACUUM;"
+                )
+                return
+
             # Check if all background updates are done, abort if not.
             updates_complete = (
                 await self.sqlite_store.db_pool.updates.has_completed_background_updates()