Hi,

On Wed, May 8, 2019 at 12:41 PM Massimo Sgaravatto <massimo.sgaravatto@gmail.com> wrote:

Hi

Fron time to time I use to move entries related to deleted instances to shadow tables, using the command:

nova-manage db archive_deleted_rows

This is now failing [*] for the instance_metadata table because of a 'duplicate entry for the primary key' problem:

DBDuplicateEntry: (pymysql.err.IntegrityError) (1062, u"Duplicate entry '6' for key 'PRIMARY'") [SQL: u'INSERT INTO shadow_instance_metadata (created_at, updated_at, deleted_at, deleted, id, `key`, value, instance_uuid) SELECT instance_metadata.created_at, instance_metadata.updated_at, instance_metadata.deleted_at, instance_metadata.deleted, instance_metadata.id, instance_metadata.`key`, instance_metadata.value, instance_metadata.instance_uuid \nFROM instance_metadata \nWHERE instance_metadata.deleted != %(deleted_1)s ORDER BY instance_metadata.id \n LIMIT %(param_1)s'] [parameters: {u'param_1': 1, u'deleted_1': 0}]

Indeed:

mysql> SELECT instance_metadata.created_at, instance_metadata.updated_at, instance_metadata.deleted_at, instance_metadata.deleted, instance_metadata.id, instance_metadata.`key`, instance_metadata.value, instance_metadata.instance_uuid FROM instance_metadata WHERE instance_metadata.deleted != 0 ORDER BY instance_metadata.id limit 1;

+---------------------+------------+---------------------+---------+----+-------+-------+--------------------------------------+

| created_at          | updated_at | deleted_at          | deleted | id | key   | value | instance_uuid                        |

+---------------------+------------+---------------------+---------+----+-------+-------+--------------------------------------+

| 2018-09-20 07:40:56 | NULL       | 2018-09-20 07:54:26 |       6 |  6 | group | node  | a9000ff7-2298-454c-bf71-9e3c62ec0f3c |

+---------------------+------------+---------------------+---------+----+-------+-------+--------------------------------------+

1 row in set (0.00 sec)

But there is a 5-years old entry (if I am not wrong we were running Havana at that time) in the shadow table with that id:

mysql> select * from shadow_instance_metadata where id='6';

+---------------------+------------+---------------------+----+----------+---------------------+--------------------------------------+---------+

| created_at          | updated_at | deleted_at          | id | key      | value               | instance_uuid                        | deleted |

+---------------------+------------+---------------------+----+----------+---------------------+--------------------------------------+---------+

| 2014-11-04 12:57:10 | NULL       | 2014-11-04 13:06:45 |  6 | director | microbosh-openstack | 5db5b17b-69f2-4f0a-bdd2-efe710268021 |       6 |

+---------------------+------------+---------------------+----+----------+---------------------+--------------------------------------+---------+

1 row in set (0.00 sec)

mysql> 

I wonder how could that happen.

Can I simply remove that entry from the shadow table (I am not really interested to keep it) or are there better (cleaner) way to fix the problem ?

This Cloud is now running Ocata

Thanks, Massimo

From what I can understand, it looks like a record with id 6 was archived long back (havana-ish) and then there was a new record with id 6 again ready to be archived ? (not sure how there could have been two records with same id since ids are incremental even over releases, I am not sure of the history though since I wasn't involved with OS then). I think the only way out is to manually delete that entry from the shadow table if you don't want it. There should be no harm in removing it.

We have a "nova-manage db purge [--all] [--before <date>] [--verbose] [--all-cells]" command that removes records from shadow_tables (https://docs.openstack.org/nova/rocky/cli/nova-manage.html) but it was introduced in rocky. So it won't be available in Ocata unfortunately.

 

Cheers,

Surya.