Note de la release actuelle

25.0.0-27

Nouvelles fonctionnalités

• Ceilometer now supports polling Octavia load balancer metrics using openstacksdk. Two new meters are available:

  • loadbalancer.operating: Reports the operating status of load balancers (ONLINE=1, DRAINING=2, OFFLINE=3, DEGRADED=4, ERROR=5, NO_MONITOR=6, unknown=-1).

  • loadbalancer.provisioning: Reports the provisioning status of load balancers (ACTIVE=1, DELETED=2, ERROR=3, PENDING_CREATE=4, PENDING_UPDATE=5, PENDING_DELETE=6, unknown=-1).

  A new loadbalancer resource type has been added to gnocchi_resources.yaml to enable publishing these metrics to Gnocchi.

• Filtering out metrics from the Gnocchi service project in the Gnocchi publisher can now be disabled using the enable_filter_project publisher option, e.g. gnocchi://?enable_filter_project=false (true by default). This ensures that the Gnocchi service project always has metrics published to Gnocchi itself, which is usually the desired behaviour when the Swift storage driver is not being used (filtering should be enabled when using the Swift storage driver, to eliminate feedback loops during metric generation).

Notes de mises à jours

• The reported units for the following metrics were changed from MB and GB to MiB and GiB respectively, as the metrics are actually in mebibytes/gibibytes:

  • memory/memory.*

  • disk.root.size

  • disk.ephemeral.size

  • volume.size

  • volume.snapshot.size/snapshot.size

  • volume.backup.size/backup.size

  • volume.provider.capacity.*/volume.provider.pool.capacity.*

  • manila.share.size

  Following the upgrade, the storage backends Ceilometer publishes to will go through an intermediary period where metrics using both the old and new units will exist at the same time:

  • In Gnocchi, newly created metrics will set unit to the newer values. Existing metrics on existing resources, however, will not have their unit updated automatically. They will need to be changed manually, if required.

  • In Prometheus, the unit label will change for the above metrics, causing Prometheus to treat them as separate metrics (though with otherwise identical labels) for non-aggregated queries. These separate metrics will co-exist until the old metrics expire, but the overlap between the old and new metrics should be small unless your query window is wide. If you perform any PromQL queries overlapping the changeover period that must have a single metric per resource, you could use aggregations like max without (unit) (...) to take into account this change.

  Regarding the values of the metrics themselves, please note that the actual values have not changed, only the reported unit names. There is no action needed unless you are converting the metrics to other units (or referencing the reported units in some way), in which case we would recommend double checking that the values are being handled correctly.

• The container_format and disk_format attributes have been made optional for image resources in Gnocchi. This fixes an issue where image resources would fail to be created in Gnocchi because container_format and disk_format are set to null (which is possible on images that haven’t had data uploaded to them yet).

• The network.services.firewall pollster now publishes samples for all found firewalls, even if they are known to have an unknown state, when they would previously be dropped. The volume of samples for such firewalls will be set to -1. This improves visibility of firewalls with unknown states, allowing them to be monitored via samples and the Gnocchi/Prometheus metrics, making it easier to discover such resources for troubleshooting. It also moves some of the « business logic » for downstream rating/billing services such as CloudKitty out of Ceilometer itself.

Autres notes

• Support for Neutron FWaaS has been un-deprecated, because FWaaS project was restored and is now maintained.