Configuration

OpenStack Configuration

Apart from configuring cASO, several actions need to be performed in your OpenStack installation in order to be able to extract accounting records.

User credentials (required)

In the next section you will configure an OpenStack Keystone credentials in order to extract the records. The cASO user has to be a member of each of the projects (another option is to convert that user in an administrator, but the former option is a safer approach) for which it is extracting the accounting, with the reader role (this is a default OpenStack Keystone role). Otherwise, cASO will not be able to get the usages and will fail:

openstack user create --password <password> accounting
# For each of the projects, add the user with the accounting role
openstack role add --user accounting --project <project> reader

Moreover, you need to grant the user the role reader with a system scope of all in order to get all the project tags, as well as the other user’s information:

openstack role add --system all --user accounting reader

Policy modifications

Important

No policy modifications are needed

The following policy modifications are just shown here for reference, if you wish to use a different role. You do not need to use them.

If you use the role reader as configured above, you do not need to configure anything else in the policy. However, if you wish to use a different role mapping, the accounting user needs access to Keystone so as to extract the users information. Depending on your configuration, you need to modify the JSON policy file (/etc/keystone/policy.json) or the YAML policy file (/etc/keystone/policy-yaml). The modifications in the policy depend on the Keystone version, please ensure that you are applying the correct changes as listed in the following table. In the example show, we are using a dedicated role accounting.

OpenStack Version	Policy contents
From Stein (>= 15.0.0)	Original	`“identity:get_user”: “(role:reader and system_scope:all) or (role:reader and token.domain.id:%(target.user.domain_id)s) or user_id:%(target.user.id)s”`
From Stein (>= 15.0.0)	Modified	`“identity:get_user”: “(role:reader and system_scope:all) or (role:reader and token.domain.id:%(target.user.domain_id)s) or user_id:%(target.user.id)s or role:accounting”`
Up to Rocky (<= 14.0.0)	Original	`“identity:get_user”: “rule:admin_or_owner”`
Up to Rocky (<= 14.0.0)	Modified	`“identity:get_user”: “rule:admin_or_owner or role:accounting”`

Selecting projects to get usages

cASO will extract project usages for those projects that have been explictly marked by the operator by either of the ways explained below. The final project list will result from the merge of both methods, so thay are not mutually exclusive.

Tagging the project with the configured caso_tag in OpenStack Keystone. By default this option is set to caso, so in order to mark a project to get extracted you should use the following command for each of the projects:
```
openstack project set --tag caso <project id>
```
You can check the list of projects to get usages by using:
```
openstack project list --tags caso
```
Using the projects list in the [DEFAULT] section of your configuration file (see below).

Setting VO mapping

In order to publish correct accounting records, cASO needs to know the VO that should be used to publish the records from a given project. In order to do so, you need to specify the correct mapping in each of the projects properties. The name of the property that will be used is defined in the vo_property configuration option, and defaults to accounting:VO, therefore you can configure it as follows:

openstack project set --property accounting:VO=<VO FQAN> <project id>

cASO configuration

cASO uses a config file (default at /etc/caso/caso.conf) with several sections. A sample file is available at etc/caso/caso.conf.sample.

`[DEFAULT]` section

The [DEFAULT] section configures the basic behavior of cASO. The sample config file (/etc/caso/caso.conf.sample) includes a description of every option. You should check at least the following options:

extractor (default value: nova), specifies which extractor to use for getting the data. The following extractors are available:
- nova - Extract VM accounting records from OpenStack Nova
- cinder - Extract storage accounting records from OpenStack Cinder
- neutron - Extract network/IP accounting records from OpenStack Neutron
- prometheus - Extract energy consumption metrics from Prometheus
You can configure multiple extractors by providing a list (e.g., nova,cinder,prometheus).
site_name (default value: <None>). Name of the site as defined in GOCDB.
service_name (default value: $site_name). Name of the service within a site. This is used if you have several endpoints within your site.
projects (list value, default empty). List of the projects to extract records from. You can use either the project ID or the project name. We recommend that you use the project ID, especially if you are using domain-based authentication, as otherwise gathering the information might fail. This option, and the usage of caso_tag below will set up the final project list.
caso_tag (default value: caso), specified the tag to be used filter projects to extract their usage. The projects that are listed with this tag, as well as the projects list set above will set up the final project list. If you only use tags, and want to remove a project from being published, you just need to remove the tag from the project.
messengers (list, default: noop). List of the messengers to publish data to. Records will be pushed to all these messengers, in order. Valid messengers shipped with cASO are:
- ssm for publishing APEL records.
- logstash for publishing to Logstash.
- noop do nothing at all.
- greendigit_cim for publishing energy records to GreenDIGIT CIM Service.
Note that there might be other messengers available in the system if they are registered into the caso.messenger entry point namespace. Please also note that versioning of the SSM messenger is deprecated.
vo_property (default: accounting:VO). The project that will be set in the OpenStack Keystone project to map a given project to a specific VO.
DEPRECATED mapping_file (default: /etc/caso/voms.json). File containing the mapping from VOs to local projects as configured in Keystone-VOMS, in the following form:
```
{
    "VO": {
        "projects": ["foo", "bar"],
    }
}
```
Note that you have to use either the project ID or project name for the mapping, as configured in the projects configuration variable.

`[keystone_auth]` section

This section is used to specify the authentication credentials to be used to connect to the OpenStack APIs. cASO leverages the OpenStack keystoneauth library for authentication, so that it is possible to use any authentication plugin that is available there (so starting on version 1.0 of cASO it is possible to use the Keystone V3 API).

Important

You need to specify the auth_type that you want to use (normally v3password is a good choice.

For an exhaustive list of available plugins please refer to the keystoneauth documentation.

`[ssm]` section

Options defined here configure the SSM messenger. There is only one option at the moment:

output_path (default: /var/spool/apel/outgoing/openstack), directory to put the generated SSM records. APEL/SSM should be configured to take records from that directory.

`[logstash]` section

Options defined here configure the logstash messenger. Available options:

host (default: localhost), host of Logstash server.
port (default: 5000), Logstash server port.

Messenger record type filtering

Each messenger can be configured to only receive specific record types. This allows you to control which types of accounting records are published to each messenger.

The configuration is done via [messenger_<name>] sections, where <name> is the messenger name (e.g., ssm, logstash, noop). Each section supports the following option:

record_types (list value). List of record types to publish to this messenger. Valid values are: cloud, ip, accelerator, storage, energy. If empty, all record types will be published.

The following record types are available:

Record types and their extractors
Record Type	Extractor	Description
`cloud`	`nova`	VM/compute accounting records
`ip`	`neutron`	Public IP usage records
`accelerator`	`nova`	Accelerator (GPU) usage records
`storage`	`cinder`	Block storage records
`energy`	`prometheus`	Energy consumption records

Default behavior:

For ssm and ssmv4 messengers, the default is to publish only records from the default extractors (nova, cinder, neutron), which means: cloud, ip, accelerator, storage. Energy records are not published by default to SSM.
For other messengers (logstash, noop), all record types are published by default.

Example configuration:

To publish only cloud and storage records to SSM:

[messenger_ssm]
record_types = cloud,storage

To publish all record types including energy records to SSM:

[messenger_ssm]
record_types = cloud,ip,accelerator,storage,energy

To publish only energy records to logstash:

[messenger_logstash]
record_types = energy

`[prometheus]` section

Options defined here configure the Prometheus extractor for gathering energy consumption metrics. This extractor uses the prometheus-api-client library to query a Prometheus instance and calculate energy consumption from instantaneous power samples. Available options:

prometheus_endpoint (default: http://localhost:9090), Prometheus server endpoint URL.
prometheus_metric_name (default: prometheus_value), Name of the Prometheus metric to query for energy consumption data.
vm_uuid_label_name (default: uuid), Name of the label in the Prometheus metric that matches the VM UUID.
labels (default: ["type_instance:scaph_process_power_microwatts"]), List of label filters expressed as key:value pairs used to filter the Prometheus metric. The VM UUID label will be automatically added to the query.
prometheus_step_seconds (default: 30), Frequency between samples in the time series, in seconds. This is used to calculate energy from power samples.
prometheus_verify_ssl (default: true), Whether to verify SSL certificates when connecting to Prometheus.

The extractor calculates energy in Watt-hours (Wh) from microwatt power samples using the formula:

sum_over_time(metric{labels}[range]) * (step_seconds/3600) / 1000000

To use the Prometheus extractor, add prometheus to the extractor option in the main configuration. For more details, see Prometheus Extractor for Energy Consumption Metrics.

Other cASO configuration options

For an exhaustive list of the defined options, please check the following page:

cASO configuration file

Additional (optional) configurations

Publishing benchmark information for OpenStack flavors (optional)

cASO is able to publish benchmark information included in the accounting recors, in order to do CPU normalization at the accounting level.

In order to do so, you need to add this information to the flavor properties and configure caso to retrieve this information. There are two different values that need to be added to the flavor

Default flavor properties used by cASO to publish benchmark information
Property	Value
`accounting:benchmark_name`	Benchmark name (e.g. HEPSPEC06)
`accounting:benchmark_value`	Benchmark value (e.g. 99)

For example, if you are using HEPSPEC06 and the benchmark value is 99 for the flavor m1.foo, the benchmark information is configured as follows:

openstack flavor set --property accounting:benchmark_name="HEPSPEC06" --property accounting:benchmark_value=99 m1.foo

Using different keys to specify bechmark information

If you do not want to use cASO’s default flavor properties accounting:benchmark_name and accounting:benchmark_value (for example because you are using different benchmark types and values) you can specify which properties cASO should look for by using the name_key and value_key in the [benchkmark] section of the configuration file.

Important

Please note that there is an OpenStack scheduler filter that removes hosts based on flavor properties. In order to not interfere with the behaviour of this filter you must prefix the property with a scope: so that cASO’s properties are not taken into account for this filtering. When adding these properties in cASO’s configuration file, please include the complete name (i.e. scope:property).

Important

Option deprecation

Please bear in mind that the old options benchmark_name_key and benchmark_value_key in the [DEFAULT] configuration option are marked as deprecated. Please update your configuration file as soon as possible to avoid warnings.

Publishing accelerator information for OpenStack accelerators (optional)

Starting with cASO >= 3.0.0 it is possible to publish accelerator information using a new accounting record.

In order to do so, you need to add this information to the flavor properties and configure caso to retrieve this information. There are different values that need to be added to the flavor:

Default flavor properties used by cASO to publish accelerator information
Flavor Property	Value
Accelerator:Type	The accelerator type (e.g. GPU))
Accelerator:Vendor	Name of the accelerator vendor (e.g. NVIDIA)
Accelerator:Model	Accelerator model (e.g. V100)
Accelerator:Number	Hoy many accelerators are available for that flavor (e.g. 2)

Using different keys to specify bechmark information

If you do not want to use cASO’s default flavor properties to publish the existing accelerators, you can specify which properties cASO should look for by using the type_key, vendor_key, model_key and number_key in the [acelerator] section of the configuration file.

Important

Please note that there is an OpenStack scheduler filter that removes hosts based on flavor properties. In order to not interfere with the behaviour of this filter you must prefix the property with a scope: so that cASO’s properties are not taken into account for this filtering. When adding these properties in cASO’s configuration file, please include the complete name (i.e. scope:property).

Configuration

OpenStack Configuration

User credentials (required)

Policy modifications

Selecting projects to get usages

Setting VO mapping

cASO configuration

[DEFAULT] section

[keystone_auth] section

[ssm] section

[logstash] section

Messenger record type filtering

[prometheus] section

Other cASO configuration options

Additional (optional) configurations

Publishing benchmark information for OpenStack flavors (optional)

Using different keys to specify bechmark information

Publishing accelerator information for OpenStack accelerators (optional)

Using different keys to specify bechmark information

`[DEFAULT]` section

`[keystone_auth]` section

`[ssm]` section

`[logstash]` section

`[prometheus]` section