migrate event table primary keys from integer to bigint

see: https://github.com/ansible/awx/issues/6010

awx/main/migrations/0113_v370_event_bigint.py

@@ -0,0 +1,118 @@
+# Generated by Django 2.2.8 on 2020-02-21 16:31
+
+from django.db import migrations, models, connection
+
+
+def migrate_event_data(apps, schema_editor):
+    # see: https://github.com/ansible/awx/issues/6010
+    #
+    # the goal of this function is to end with event tables (e.g., main_jobevent)
+    # that have a bigint primary key (because the old usage of an integer
+    # numeric isn't enough, as its range is about 2.1B, see:
+    # https://www.postgresql.org/docs/9.1/datatype-numeric.html)
+
+    # unfortunately, we can't do this with a simple ALTER TABLE, because
+    # for tables with hundreds of millions or billions of rows, the ALTER TABLE
+    # can take *hours* on modest hardware.
+    #
+    # the approach in this migration means that post-migration, event data will
+    # *not* immediately show up, but will be repopulated over time progressively
+    # the trade-off here is not having to wait hours for the full data migration
+    # before you can start and run AWX again (including new playbook runs)
+    for tblname in (
+        'main_jobevent', 'main_inventoryupdateevent',
+        'main_projectupdateevent', 'main_adhoccommandevent',
+        'main_systemjobevent'
+    ):
+        with connection.cursor() as cursor:
+            # rename the current event table
+            cursor.execute(
+                f'ALTER TABLE {tblname} RENAME TO _old_{tblname};'
+            )
+            # create a *new* table with the same schema
+            cursor.execute(
+                f'CREATE TABLE {tblname} (LIKE _old_{tblname} INCLUDING ALL);'
+            )
+            # alter the *new* table so that the primary key is a big int
+            cursor.execute(
+                f'ALTER TABLE {tblname} ALTER COLUMN id TYPE bigint USING id::bigint;'
+            )
+
+            # recreate counter for the new table's primary key to
+            # start where the *old* table left off (we have to do this because the
+            # counter changed from an int to a bigint)
+            cursor.execute(f'DROP SEQUENCE IF EXISTS "{tblname}_id_seq" CASCADE;')
+            cursor.execute(f'CREATE SEQUENCE "{tblname}_id_seq";')
+            cursor.execute(
+                f'ALTER TABLE "{tblname}" ALTER COLUMN "id" '
+                f"SET DEFAULT nextval('{tblname}_id_seq');"
+            )
+            cursor.execute(
+                f"SELECT setval('{tblname}_id_seq', (SELECT MAX(id) FROM _old_{tblname}), true);"
+            )
+
+            # replace the BTREE index on main_jobevent.job_id with
+            # a BRIN index to drastically improve per-UJ lookup performance
+            # see: https://info.crunchydata.com/blog/postgresql-brin-indexes-big-data-performance-with-minimal-storage
+            if tblname == 'main_jobevent':
+                cursor.execute("SELECT indexname FROM pg_indexes WHERE tablename='main_jobevent' AND indexdef LIKE '%USING btree (job_id)';")
+                old_index = cursor.fetchone()[0]
+                cursor.execute(f'DROP INDEX {old_index}')
+                cursor.execute('CREATE INDEX main_jobevent_job_id_brin_idx ON main_jobevent USING brin (job_id);')
+
+            # remove all of the indexes and constraints from the old table
+            # (they just slow down the data migration)
+            cursor.execute(f"SELECT indexname, indexdef FROM pg_indexes WHERE tablename='_old_{tblname}' AND indexname != '{tblname}_pkey';")
+            indexes = cursor.fetchall()
+
+            cursor.execute(f"SELECT conname, contype, pg_catalog.pg_get_constraintdef(r.oid, true) as condef FROM pg_catalog.pg_constraint r WHERE r.conrelid = '_old_{tblname}'::regclass AND conname != '{tblname}_pkey';")
+            constraints = cursor.fetchall()
+
+            for indexname, indexdef in indexes:
+                cursor.execute(f'DROP INDEX IF EXISTS {indexname}')
+            for conname, contype, condef in constraints:
+                cursor.execute(f'ALTER TABLE _old_{tblname} DROP CONSTRAINT IF EXISTS {conname}')
+
+
+class FakeAlterField(migrations.AlterField):
+
+    def database_forwards(self, *args):
+        # this is intentionally left blank, because we're
+        # going to accomplish the migration with some custom raw SQL
+        pass
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('main', '0112_v370_workflow_node_identifier'),
+    ]
+
+    operations = [
+        migrations.RunPython(migrate_event_data),
+        FakeAlterField(
+            model_name='adhoccommandevent',
+            name='id',
+            field=models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID'),
+        ),
+        FakeAlterField(
+            model_name='inventoryupdateevent',
+            name='id',
+            field=models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID'),
+        ),
+        FakeAlterField(
+            model_name='jobevent',
+            name='id',
+            field=models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID'),
+        ),
+        FakeAlterField(
+            model_name='projectupdateevent',
+            name='id',
+            field=models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID'),
+        ),
+        FakeAlterField(
+            model_name='systemjobevent',
+            name='id',
+            field=models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID'),
+        ),
+    ]