instance_type data cannot be extracted from sys_meta after instance deleted

[Chris Behrens]

This commit:

---
commit 17bca438954638d74035d560e826a26a532b3ea3
Author: Joe Gordon <email address hidden>
Date: Wed Mar 13 16:58:13 2013 -0700

    Delete InstanceSystemMetadata on instance deletion

    Thanks to no-db-compute only pre-delete instance data is used in
    notifications. So now we can go back to deleting InstanceSystemMetadata
    when an instance is deleted.

    Fixes bug 1153827

    Change-Id: Ic66b998cde2a15a24f302f08c34753a8b57da73d
---

Changed things so that we delete the instance_system_metadata entries for an instance when instance_destroy() is called. However, the commit message turns out to be wrong, from what I can tell. There's 2 cases that need access to instance_type data that's stored in sys_meta AFTER an instance is deleted:

1) The API. If you specify 'changes-since', the API will return deleted instances. The instance_type data needs to be extracted in order to form the API response
2) _usage_audit_log() task in compute manager. This task potentially queries for deleted instances in the past day to send audit information for them nightly.

I don't have a traceback for #1 handy, but:

Traceback for #2:

Traceback (most recent call last):
  File "/meow/nova/compute/manager.py", line 3813, in _instance_usage_audit
        include_bandwidth=True
  File "/meow/nova/conductor/api.py", line 308, in notify_usage_exists
    system_metadata, extra_usage_info, include_bandwidth)
  File "/meow/nova/utils.py", line 1089, in wrapper
    return func(*args, **kwargs)
  File "/meow/nova/conductor/manager.py", line 405, in notify_usage_exists
        include_bandwidth) File "/meow/nova/compute/utils.py", line 224, in notify_usage_exists
    system_metadata=system_metadata, extra_usage_info=extra_info)
  File "/meow/nova/compute/utils.py", line 250, in notify_about_instance_usage
    network_info, system_metadata, **extra_usage_info)
  File "/meow/nova/notifications.py", line 287, in info_from_instance
        instance_type = flavors.extract_instance_type(instance_ref)
  File "/meow/nova/compute/flavors.py", line 247, in extract_instance_type
    instance_type[key] = type_fn(sys_meta[type_key])
KeyError: 'instance_type_memory_mb'

[Thierry Carrez (ttx)]

Pushing back to H2, please set back to H1 if this is milestone-critical (in which case the fix needs to land in master first and then be backported to milestone-proposed before tomorrow)

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/30894
Committed: http://github.com/openstack/nova/commit/4885aa28706a1858f4fc51a0d2c661eec05139c0
Submitter: Jenkins
Branch: master

commit 4885aa28706a1858f4fc51a0d2c661eec05139c0
Author: Chris Behrens <email address hidden>
Date: Wed May 29 14:21:00 2013 +0000

    Don't delete sys_meta on instance delete

    Unfortunately, we require to access instance_system_metadata to get
    data (specifically at least instance_type data) for instances that have
    been deleted.

    There's 2 cases where this is true:

    1) nova-api supports showing deleting instances when you specify a
    'changes-since' param.
    2) The _usage_audit_log periodic task pulls all instances during the
    audit period, which includes instances that have been deleted during
    that period.

    This reverts commit 17bca438954638d74035d560e826a26a532b3ea3, which was
    attempting to fix a bug where we leave instance_system_metadata entries
    undeleted from the DB. There's not an easy way to query for the deleted
    sys_meta data in an efficient manner, and leaving the entries undeleted is
    the lesser of the 2 evils for now.

    Fixes bug 1185190

    Change-Id: I898f0546c49126dcc56a4237120082f95dc82304

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: milestone-proposed
Review: https://review.openstack.org/30954

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/30954
Committed: http://github.com/openstack/nova/commit/04d9fb406783ba5934d9cb249550a92202555bc7
Submitter: Jenkins
Branch: milestone-proposed

commit 04d9fb406783ba5934d9cb249550a92202555bc7
Author: Chris Behrens <email address hidden>
Date: Wed May 29 14:21:00 2013 +0000

    Don't delete sys_meta on instance delete

    Unfortunately, we require to access instance_system_metadata to get
    data (specifically at least instance_type data) for instances that have
    been deleted.

    There's 2 cases where this is true:

    1) nova-api supports showing deleting instances when you specify a
    'changes-since' param.
    2) The _usage_audit_log periodic task pulls all instances during the
    audit period, which includes instances that have been deleted during
    that period.

    This reverts commit 17bca438954638d74035d560e826a26a532b3ea3, which was
    attempting to fix a bug where we leave instance_system_metadata entries
    undeleted from the DB. There's not an easy way to query for the deleted
    sys_meta data in an efficient manner, and leaving the entries undeleted is
    the lesser of the 2 evils for now.

    Fixes bug 1185190

    Change-Id: I898f0546c49126dcc56a4237120082f95dc82304
    (cherry picked from commit 4885aa28706a1858f4fc51a0d2c661eec05139c0)

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: stable/grizzly
Review: https://review.openstack.org/31244

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/31244
Committed: http://github.com/openstack/nova/commit/8c136a8ee4733d8d3af61406fc9eeccb4e9b7a68
Submitter: Jenkins
Branch: stable/grizzly

commit 8c136a8ee4733d8d3af61406fc9eeccb4e9b7a68
Author: Chris Behrens <email address hidden>
Date: Wed May 29 14:21:00 2013 +0000

    Don't delete sys_meta on instance delete

    Unfortunately, we require to access instance_system_metadata to get
    data (specifically at least instance_type data) for instances that have
    been deleted.

    There's 2 cases where this is true:

    1) nova-api supports showing deleting instances when you specify a
    'changes-since' param.
    2) The _usage_audit_log periodic task pulls all instances during the
    audit period, which includes instances that have been deleted during
    that period.

    This reverts commit 17bca438954638d74035d560e826a26a532b3ea3, which was
    attempting to fix a bug where we leave instance_system_metadata entries
    undeleted from the DB. There's not an easy way to query for the deleted
    sys_meta data in an efficient manner, and leaving the entries undeleted is
    the lesser of the 2 evils for now.

    Fixes bug 1185190

    Change-Id: I898f0546c49126dcc56a4237120082f95dc82304
    (cherry picked from commit 4885aa28706a1858f4fc51a0d2c661eec05139c0)