Name an extension used through an array or a domain in the data copy preflight (S-007-7)

changelog.d/173-preflight-extension-arrays.md

@@ -0,0 +1,3 @@
+### Fixed
+
+- The data copy preflight now names an extension that a selected table needs through an array or a domain of the extension's type, such as a `citext[]` column or a column of a domain over `citext` (chained domains included), under `extensions` and passes it to `pg_dump --extension`; before, only a direct use of the extension's type was seen, so the copy skipped the extension and the restore failed (#173)

engine/ansible/roles/cloudfall_data_migration/defaults/main.yml

@@ -27,15 +27,18 @@ cloudfall_data_migration_target_tables_query: >-
 #   extensions: an extension the selected schemas need: one installed in a
 #     selected schema, one whose member object an object of a selected schema
 #     depends on in pg_depend (a column type, a default, an index operator
-#     class, a view, a trigger, a policy), or, with no schema named, every
-#     extension but plpgsql, since the whole dump creates them all
+#     class, a view, a trigger, a policy), following an array type to its
+#     element type and a domain to its base type, chained, since a citext[]
+#     column depends on _citext and only citext is the extension's member,
+#     or, with no schema named, every extension but plpgsql, since the whole
+#     dump creates them all
 #   policyRoles: a role a row-level security policy on a table of a selected
 #     schema names (pg_policy.polroles; 0 is PUBLIC)
 #   authReferences: when auth isn't selected, a function of a selected schema
 #     whose source mentions auth. or that depends on an object in auth, and a
 #     policy on a table of a selected schema whose USING or WITH CHECK does
 cloudfall_data_migration_source_preflight_query: >-
-  WITH selected AS (
+  WITH RECURSIVE selected AS (
     SELECT jsonb_array_elements_text(:'cloudfall_data_schemas'::jsonb)
       AS nspname
   ),
@@ -83,15 +86,28 @@ cloudfall_data_migration_source_preflight_query: >-
     WHERE d.refclassid = 'pg_catalog.pg_extension'::regclass
       AND d.deptype = 'e'
   ),
+  referenced(classid, objid) AS (
+    SELECT d.refclassid, d.refobjid
+    FROM pg_catalog.pg_depend d
+    JOIN selected_objects o ON o.classid = d.classid AND o.objid = d.objid
+    UNION
+    SELECT 'pg_catalog.pg_type'::regclass::oid, under.oid
+    FROM referenced r JOIN pg_catalog.pg_type t
+      ON r.classid = 'pg_catalog.pg_type'::regclass AND t.oid = r.objid
+    CROSS JOIN LATERAL (VALUES
+      (CASE WHEN t.typcategory = 'A' THEN t.typelem END),
+      (CASE WHEN t.typtype = 'd' THEN t.typbasetype END)
+    ) AS under(oid)
+    WHERE under.oid <> 0
+  ),
   needed AS (
     SELECT e.extname FROM pg_catalog.pg_extension e
     WHERE (:'cloudfall_data_every_schema'::boolean AND e.extname <> 'plpgsql')
       OR e.extnamespace IN (SELECT oid FROM selected_ns)
       OR e.oid IN (
         SELECT m.extoid
-        FROM pg_catalog.pg_depend d
-        JOIN selected_objects o ON o.classid = d.classid AND o.objid = d.objid
-        JOIN members m ON m.classid = d.refclassid AND m.objid = d.refobjid
+        FROM referenced r JOIN members m
+          ON m.classid = r.classid AND m.objid = r.objid
         UNION
         SELECT d.refobjid
         FROM pg_catalog.pg_depend d

engine/tests/test_data_migration_role.py

@@ -472,6 +472,23 @@ def test_s007_7_an_extension_the_target_lacks_is_refused() -> None:
     assert "e.extname <> 'plpgsql'" in sql
 
 
+def test_s007_7_an_array_or_domain_of_an_extension_type_names_it() -> None:
+    sql = _source_preflight_sql(["public"], every_schema=False)
+    needed = sql[sql.index("needed AS (") : sql.index("missing AS (")]
+
+    # A citext[] column depends on _citext, which depends on citext with
+    # deptype 'i', not on the extension: the walk follows an array to its
+    # element type and a domain to its base type, chained, before it
+    # matches extension members, in the one query that feeds both the
+    # extensions finding and pg_dump --extension.
+    assert sql.startswith("WITH RECURSIVE selected AS (")
+    assert "referenced(classid, objid) AS (" in sql
+    assert "FROM referenced r JOIN pg_catalog.pg_type t" in sql
+    assert "WHEN t.typcategory = 'A' THEN t.typelem" in sql
+    assert "WHEN t.typtype = 'd' THEN t.typbasetype" in sql
+    assert "FROM referenced r JOIN members m" in needed
+
+
 def test_s007_7_named_schemas_dump_their_needed_extensions() -> None:
     shell = _module(
         "Dump the source database on the target host", "ansible.builtin.shell"