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”

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”

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 APIs are supported: ceilomenter and nova. Both should generate equivalent information.

  • 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.

    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.

Other cASO configuration options

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

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).