migrate event table primary keys from integer to bigint

see: #6010

awx/main/migrations/0109_v370_event_bigint.py

@@ -0,0 +1,62 @@
+# 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'
+    ):
+        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);"
+            )
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('main', '0108_v370_unifiedjob_dependencies_processed'),
+    ]
+
+    operations = [
+        migrations.RunPython(migrate_event_data)
+    ]

awx/main/models/__init__.py

@@ -3,6 +3,7 @@
 
 # Django
 from django.conf import settings # noqa
+from django.db import connection, ProgrammingError
 from django.db.models.signals import pre_delete  # noqa
 
 # AWX
@@ -80,6 +81,26 @@
 User.add_to_class('accessible_objects', user_accessible_objects)
 
 
+def enforce_bigint_pk_migration():
+    # see: https://github.com/ansible/awx/issues/6010
+    # look at all the event tables and verify that they have been fully migrated
+    # from the *old* int primary key table to the replacement bigint table
+    # if not, attempt to migrate them in the background
+    for tblname in (
+        'main_jobevent', 'main_inventoryupdateevent',
+        'main_projectupdateevent', 'main_adhoccommandevent'
+    ):
+        with connection.cursor() as cursor:
+            try:
+                cursor.execute(f'SELECT MAX(id) FROM _old_{tblname}')
+                if cursor.fetchone():
+                    from awx.main.tasks import migrate_legacy_event_data
+                    migrate_legacy_event_data.apply_async([tblname])
+            except ProgrammingError:
+                # the table is gone (migration is unnecessary)
+                pass
+
+
 def cleanup_created_modified_by(sender, **kwargs):
     # work around a bug in django-polymorphic that doesn't properly
     # handle cascades for reverse foreign keys on the polymorphic base model

awx/main/models/events.py

@@ -430,6 +430,7 @@ class Meta:
             ('job', 'parent_uuid'),
         ]
 
+    id = models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')
     job = models.ForeignKey(
         'Job',
         related_name='job_events',
@@ -518,6 +519,7 @@ class Meta:
             ('project_update', 'end_line'),
         ]
 
+    id = models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')
     project_update = models.ForeignKey(
         'ProjectUpdate',
         related_name='project_update_events',
@@ -669,6 +671,7 @@ class Meta:
     FAILED_EVENTS = [x[0] for x in EVENT_TYPES if x[2]]
     EVENT_CHOICES = [(x[0], x[1]) for x in EVENT_TYPES]
 
+    id = models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')
     event = models.CharField(
         max_length=100,
         choices=EVENT_CHOICES,
@@ -731,6 +734,7 @@ class Meta:
             ('inventory_update', 'end_line'),
         ]
 
+    id = models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')
     inventory_update = models.ForeignKey(
         'InventoryUpdate',
         related_name='inventory_update_events',

awx/main/tasks.py

@@ -26,7 +26,7 @@
 
 # Django
 from django.conf import settings
-from django.db import transaction, DatabaseError, IntegrityError
+from django.db import transaction, DatabaseError, IntegrityError, connection
 from django.db.models.fields.related import ForeignKey
 from django.utils.timezone import now, timedelta
 from django.utils.encoding import smart_str
@@ -59,7 +59,7 @@
     Inventory, InventorySource, SmartInventoryMembership,
     Job, AdHocCommand, ProjectUpdate, InventoryUpdate, SystemJob,
     JobEvent, ProjectUpdateEvent, InventoryUpdateEvent, AdHocCommandEvent, SystemJobEvent,
-    build_safe_env
+    build_safe_env, enforce_bigint_pk_migration
 )
 from awx.main.constants import ACTIVE_STATES
 from awx.main.exceptions import AwxTaskError
@@ -135,6 +135,12 @@ def dispatch_startup():
     if Instance.objects.me().is_controller():
         awx_isolated_heartbeat()
 
+    # at process startup, detect the need to migrate old event records from int
+    # to bigint; at *some point* in the future, once certain versions of AWX
+    # and Tower fall out of use/support, we can probably just _assume_ that
+    # everybody has moved to bigint, and remove this code entirely
+    enforce_bigint_pk_migration()
+
 
 def inform_cluster_of_shutdown():
     try:
@@ -660,6 +666,38 @@ def update_host_smart_inventory_memberships():
         smart_inventory.update_computed_fields()
 
 
+@task()
+def migrate_legacy_event_data(tblname):
+    if 'event' not in tblname:
+        return
+    with advisory_lock(f'bigint_migration_{tblname}', wait=False) as acquired:
+        if acquired is False:
+            return
+        chunk = 10000
+
+        def _remaining():
+            cursor.execute(f'SELECT MAX(id) FROM _old_{tblname};')
+            return cursor.fetchone()[0]
+
+        with connection.cursor() as cursor:
+            total_rows = _remaining()
+            if total_rows is None:
+                return
+            while total_rows is not None:
+                sql = f'INSERT INTO {tblname} SELECT * FROM _old_{tblname} ORDER BY id DESC LIMIT {chunk};'
+                logger.warn(sql)
+                cursor.execute(sql)
+                cursor.execute(
+                    f'DELETE FROM _old_{tblname} WHERE id IN (SELECT id FROM _old_{tblname} ORDER BY id DESC LIMIT {chunk});'
+                )
+                connection.commit()
+                total_rows = _remaining()
+
+            if total_rows is None:
+                cursor.execute(f'DROP TABLE _old_{tblname}')
+                logger.warn(f'{tblname} primary key migration to BIGINT has finished')
+
+
 @task()
 def delete_inventory(inventory_id, user_id, retries=5):
     # Delete inventory as user