Introduction to dynamic pollster subsystem¶
The dynamic pollster feature allows system administrators to
create/update REST API pollsters on the fly (without changing code).
The system reads YAML configures that are found in
pollsters_definitions_dirs parameter, which has the default at
/etc/ceilometer/pollsters.d. Operators can use a single file per
dynamic pollster or multiple dynamic pollsters per file.
Current limitations of the dynamic pollster system¶
Currently, the following types of APIs are not supported by the dynamic pollster system:
Tenant APIs: Tenant APIs are the ones that need to be polled in a tenant fashion. This feature is “a nice” to have, but is currently not implemented.
The dynamic pollsters system configuration (for OpenStack APIs)¶
Each YAML file in the dynamic pollster feature can use the following attributes to define a dynamic pollster:
Warning
Caution: Ceilometer does not accept complex value data structure for
value and metadata configurations. Therefore, if you are extracting
a complex data structure (Object, list, map, or others), you can take
advantage of the Operations on extracted attributes feature to transform
the object into a simple value (string or number)
name: mandatory field. It specifies the name/key of the dynamic pollster. For instance, a pollster for magnum can use the namedynamic.magnum.cluster;sample_type: mandatory field; it defines the sample type. It must be one of the values:gauge,delta,cumulative;unit: mandatory field; defines the unit of the metric that is being collected. For magnum, for instance, one can useclusteras the unit or some other meaningful String value;value_attribute: mandatory attribute; defines the attribute in the JSON response from the URL of the component being polled. We also accept nested values dictionaries. To use a nested value one can simply useattribute1.attribute2.<asMuchAsNeeded>.lastattribute. It is also possible to reference the sample itself using"." (dot); the self reference of the sample is interesting in cases when the attribute might not exist. Therefore, together with the operations options, one can first check if it exist before retrieving it (example:". | value['some_field'] if 'some_field' in value else ''"). In our magnum example, we can usestatusas the value attribute;endpoint_type: mandatory field; defines the endpoint type that is used to discover the base URL of the component to be monitored; for magnum, one can usecontainer-infra. Other values are accepted such asvolumefor cinder endpoints,object-storefor swift, and so on;url_path: mandatory attribute. It defines the path of the request that we execute on the endpoint to gather data. For example, to gather data from magnum, one can usev1/clusters/detail;metadata_fields: optional field. It is a list of all fields that the response of the request executed withurl_paththat we want to retrieve. To use a nested value one can simply useattribute1.attribute2.<asMuchAsNeeded>.lastattribute. As an example, for magnum, one can use the following values:
metadata_fields: - "labels" - "updated_at" - "keypair" - "master_flavor_id" - "api_address" - "master_addresses" - "node_count" - "docker_volume_size" - "master_count" - "node_addresses" - "status_reason" - "coe_version" - "cluster_template_id" - "name" - "stack_id" - "created_at" - "discovery_url" - "container_version"
skip_sample_values: optional field. It defines the values that might come in thevalue_attributethat we want to ignore. For magnun, one could for instance, ignore some of the status it has for clusters. Therefore, data is not gathered for clusters in the defined status.
skip_sample_values: - "CREATE_FAILED" - "DELETE_FAILED"
value_mapping: optional attribute. It defines a mapping for the values that the dynamic pollster is handling. This is the actual value that is sent to Gnocchi or other backends. If there is no mapping specified, we will use the raw value that is obtained with the use ofvalue_attribute. An example for magnum, one can use:
value_mapping: CREATE_IN_PROGRESS: "0" CREATE_FAILED: "1" CREATE_COMPLETE: "2" UPDATE_IN_PROGRESS: "3" UPDATE_FAILED: "4" UPDATE_COMPLETE: "5" DELETE_IN_PROGRESS: "6" DELETE_FAILED: "7" DELETE_COMPLETE: "8" RESUME_COMPLETE: "9" RESUME_FAILED: "10" RESTORE_COMPLETE: "11" ROLLBACK_IN_PROGRESS: "12" ROLLBACK_FAILED: "13" ROLLBACK_COMPLETE: "14" SNAPSHOT_COMPLETE: "15" CHECK_COMPLETE: "16" ADOPT_COMPLETE: "17"
default_value: optional parameter. The default value for the value mapping in case the variable value receives data that is not mapped to something in thevalue_mappingconfiguration. This attribute is only used whenvalue_mappingis defined. Moreover, it has a default of-1.metadata_mapping: optional parameter. The map used to create new metadata fields. The key is a metadata name that exists in the response of the request we make, and the value of this map is the new desired metadata field that will be created with the content of the metadata that we are mapping. Themetadata_mappingcan be created as follows:
metadata_mapping: name: "display_name" some_attribute: "new_attribute_name"
preserve_mapped_metadata: optional parameter. It indicates if we preserve the old metadata name when it gets mapped to a new one. The default value isTrue.response_entries_key: optional parameter. This value is used to define the “key” of the response that will be used to look-up the entries used in the dynamic pollster processing. If noresponse_entries_keyis informed by the operator, we will use the first we find. Moreover, if the response contains a list, instead of an object where one of its attributes is a list of entries, we use the list directly. Therefore, this option will be ignored when the API is returning the list/array of entries to be processed directly. We also accept nested values dictionaries. To use a nested value one can simply useattribute1.attribute2.<asMuchAsNeeded>.lastattributeuser_id_attribute: optional parameter. The default value isuser_id. The name of the attribute in the entries that are processed fromresponse_entries_keyelements that will be mapped touser_idattribute that is sent to Gnocchi.project_id_attribute: optional parameter. The default value isproject_id. The name of the attribute in the entries that are processed fromresponse_entries_keyelements that will be mapped toproject_idattribute that is sent to Gnocchi.resource_id_attribute: optional parameter. The default value isid. The name of the attribute in the entries that are processed fromresponse_entries_keyelements that will be mapped toidattribute that is sent to Gnocchi.headers: optional parameter. It is a map (similar to the metadata_mapping) of key and value that can be used to customize the header of the request that is executed against the URL. This configuration works for both OpenStack and non-OpenStack dynamic pollster configuration.
headers: "x-openstack-nova-api-version": "2.46"
timeout: optional parameter. Defines the request timeout for the requests executed by the dynamic pollsters to gather data. The default timeout value is 30 seconds. If it is set to None, this means that the request never times out on the client side. Therefore, one might have problems if the server never closes the connection. The pollsters are executed serially, one after the other. Therefore, if the request hangs, all pollsters (including the non-dynamic ones) will stop executing.
The complete YAML configuration to gather data from Magnum (that has been used as an example) is the following:
---
- name: "dynamic.magnum.cluster"
sample_type: "gauge"
unit: "cluster"
value_attribute: "status"
endpoint_type: "container-infra"
url_path: "v1/clusters/detail"
metadata_fields:
- "labels"
- "updated_at"
- "keypair"
- "master_flavor_id"
- "api_address"
- "master_addresses"
- "node_count"
- "docker_volume_size"
- "master_count"
- "node_addresses"
- "status_reason"
- "coe_version"
- "cluster_template_id"
- "name"
- "stack_id"
- "created_at"
- "discovery_url"
- "container_version"
value_mapping:
CREATE_IN_PROGRESS: "0"
CREATE_FAILED: "1"
CREATE_COMPLETE: "2"
UPDATE_IN_PROGRESS: "3"
UPDATE_FAILED: "4"
UPDATE_COMPLETE: "5"
DELETE_IN_PROGRESS: "6"
DELETE_FAILED: "7"
DELETE_COMPLETE: "8"
RESUME_COMPLETE: "9"
RESUME_FAILED: "10"
RESTORE_COMPLETE: "11"
ROLLBACK_IN_PROGRESS: "12"
ROLLBACK_FAILED: "13"
ROLLBACK_COMPLETE: "14"
SNAPSHOT_COMPLETE: "15"
CHECK_COMPLETE: "16"
ADOPT_COMPLETE: "17"
We can also replicate and enhance some hardcoded pollsters. For instance, the pollster to gather VPN connections. Currently, it is always persisting 1 for all of the VPN connections it finds. However, the VPN connection can have multiple statuses, and we should normally only bill for active resources, and not resources on ERROR states. An example to gather VPN connections data is the following (this is just an example, and one can adapt and configure as he/she desires):
---
- name: "dynamic.network.services.vpn.connection"
sample_type: "gauge"
unit: "ipsec_site_connection"
value_attribute: "status"
endpoint_type: "network"
url_path: "v2.0/vpn/ipsec-site-connections"
metadata_fields:
- "name"
- "vpnservice_id"
- "description"
- "status"
- "peer_address"
value_mapping:
ACTIVE: "1"
metadata_mapping:
name: "display_name"
default_value: 0
The dynamic pollsters system configuration (for non-OpenStack APIs)¶
The dynamic pollster system can also be used for non-OpenStack APIs.
to configure non-OpenStack APIs, one can use all but one attribute of
the Dynamic pollster system. The attribute that is not supported is
the endpoint_type. The dynamic pollster system for non-OpenStack APIs
is activated automatically when one uses the configurations module.
The extra parameters (in addition to the original ones) that are available when using the Non-OpenStack dynamic pollster sub-subsystem are the following:
module: required parameter. It is the python module name that Ceilometer has to load to use the authentication object when executing requests against the API. For instance, if one wants to create a pollster to gather data from RadosGW, he/she can use theawsauthpython module.authentication_object: mandatory parameter. The name of the class that we can find in themodulethat Ceilometer will use as the authentication object in the request. For instance, when using theawsauthpython module to gather data from RadosGW, one can use the authentication object asS3Auth.authentication_parameters: optional parameter. It is a comma separated value that will be used to instantiate theauthentication_object. For instance, if we gather data from RadosGW, and we use theS3Authclass, theauthentication_parameterscan be configured as<rados_gw_access_key>, rados_gw_secret_key, rados_gw_host_name.barbican_secret_id: optional parameter. The Barbican secret ID, from which, Ceilometer can retrieve the comma separated values of theauthentication_parameters.
As follows we present an example on how to convert the hard-coded pollster for radosgw.api.request metric to the dynamic pollster model:
---
- name: "dynamic.radosgw.api.request"
sample_type: "gauge"
unit: "request"
value_attribute: "total.ops"
url_path: "http://rgw.service.stage.i.ewcs.ch/admin/usage"
module: "awsauth"
authentication_object: "S3Auth"
authentication_parameters: "<access_key>,<secret_key>,<rados_gateway_server>"
user_id_attribute: "user"
project_id_attribute: "user"
resource_id_attribute: "user"
response_entries_key: "summary"
We can take that example a bit further, and instead of gathering the total .ops variable, which counts for all the requests (even the unsuccessful ones), we can use the successful_ops.
---
- name: "dynamic.radosgw.api.request.successful_ops"
sample_type: "gauge"
unit: "request"
value_attribute: "total.successful_ops"
url_path: "http://rgw.service.stage.i.ewcs.ch/admin/usage"
module: "awsauth"
authentication_object: "S3Auth"
authentication_parameters: "<access_key>, <secret_key>,<rados_gateway_server>"
user_id_attribute: "user"
project_id_attribute: "user"
resource_id_attribute: "user"
response_entries_key: "summary"
Operations on extracted attributes¶
The dynamic pollster system can execute Python operations to transform the attributes that are extracted from the JSON response that the system handles.
One example of use case is the RadosGW that uses <project_id$project_id> as the username (which is normally mapped to the Gnocchi resource_id). With this feature (operations on extracted attributes), one can create configurations in the dynamic pollster to clean/normalize that variable. It is as simple as defining resource_id_attribute: “user | value.split(‘$’)[0].strip()”
The operations are separated by | symbol. The first element of the expression is the key to be retrieved from the JSON object. The other elements are operations that can be applied to the value variable. The value variable is the variable we use to hold the data being extracted. The previous example can be rewritten as: resource_id_attribute: “user | value.split (‘$’) | value[0] | value.strip()”
As follows we present a complete configuration for a RadosGW dynamic pollster that is removing the $ symbol, and getting the first part of the String.
---
- name: "dynamic.radosgw.api.request.successful_ops"
sample_type: "gauge"
unit: "request"
value_attribute: "total.successful_ops"
url_path: "http://rgw.service.stage.i.ewcs.ch/admin/usage"
module: "awsauth"
authentication_object: "S3Auth"
authentication_parameters: "<access_key>,<secret_key>,<rados_gateway_server>"
user_id_attribute: "user | value.split ('$') | value[0]"
project_id_attribute: "user | value.split ('$') | value[0]"
resource_id_attribute: "user | value.split ('$') | value[0]"
response_entries_key: "summary"
The Dynamic pollster configuration options that support this feature are the following:
value_attribute
response_entries_key
user_id_attribute
project_id_attribute
resource_id_attribute
Multi metric dynamic pollsters (handling attribute values with list of objects)¶
The initial idea for this feature comes from the categories fields that we can find in the summary object of the RadosGW API. Each user has a categories attribute in the response; in the categories list, we can find the object that presents in a granular fashion the consumption of different RadosGW API operations such as GET, PUT, POST, and may others.
As follows we present an example of such a JSON response.
{
"entries": [
{
"buckets": [
{
"bucket": "",
"categories": [
{
"bytes_received": 0,
"bytes_sent": 40,
"category": "list_buckets",
"ops": 2,
"successful_ops": 2
}
],
"epoch": 1572969600,
"owner": "user",
"time": "2019-11-21 00:00:00.000000Z"
},
{
"bucket": "-",
"categories": [
{
"bytes_received": 0,
"bytes_sent": 0,
"category": "get_obj",
"ops": 1,
"successful_ops": 0
}
],
"epoch": 1572969600,
"owner": "someOtherUser",
"time": "2019-11-21 00:00:00.000000Z"
}
]
}
]
"summary": [
{
"categories": [
{
"bytes_received": 0,
"bytes_sent": 0,
"category": "create_bucket",
"ops": 2,
"successful_ops": 2
},
{
"bytes_received": 0,
"bytes_sent": 2120428,
"category": "get_obj",
"ops": 46,
"successful_ops": 46
},
{
"bytes_received": 0,
"bytes_sent": 21484,
"category": "list_bucket",
"ops": 8,
"successful_ops": 8
},
{
"bytes_received": 6889056,
"bytes_sent": 0,
"category": "put_obj",
"ops": 46,
"successful_ops": 46
}
],
"total": {
"bytes_received": 6889056,
"bytes_sent": 2141912,
"ops": 102,
"successful_ops": 102
},
"user": "user"
},
{
"categories": [
{
"bytes_received": 0,
"bytes_sent": 0,
"category": "create_bucket",
"ops": 1,
"successful_ops": 1
},
{
"bytes_received": 0,
"bytes_sent": 0,
"category": "delete_obj",
"ops": 23,
"successful_ops": 23
},
{
"bytes_received": 0,
"bytes_sent": 5371,
"category": "list_bucket",
"ops": 2,
"successful_ops": 2
},
{
"bytes_received": 3444350,
"bytes_sent": 0,
"category": "put_obj",
"ops": 23,
"successful_ops": 23
}
],
"total": {
"bytes_received": 3444350,
"bytes_sent": 5371,
"ops": 49,
"successful_ops": 49
},
"user": "someOtherUser"
}
]
}
In that context, and having in mind that we have APIs with similar data structures, we developed an extension for the dynamic pollster that enables multi-metric processing for a single pollster. It works as follows.
The pollster name will contain a placeholder for the variable that identifies the “submetric”. E.g. dynamic.radosgw.api.request.{category}. The placeholder {category} indicates the object’s attribute that is in the list of objects that we use to load the sub metric name. Then, we must use a special notation in the value_attribute configuration to indicate that we are dealing with a list of objects. This is achieved via [] (brackets); for instance, in the dynamic.radosgw.api.request.{category}, we can use [categories].ops as the value_attribute. This indicates that the value we retrieve is a list of objects, and when the dynamic pollster processes it, we want it (the pollster) to load the ops value for the sub metrics being generated.
Examples on how to create multi-metric pollster to handle data from RadosGW API are presented as follows:
---
- name: "dynamic.radosgw.api.request.{category}"
sample_type: "gauge"
unit: "request"
value_attribute: "[categories].ops"
url_path: "http://rgw.service.stage.i.ewcs.ch/admin/usage"
module: "awsauth"
authentication_object: "S3Auth"
authentication_parameters: "<access_key>, <secret_key>,<rados_gateway_server>"
user_id_attribute: "user | value.split('$')[0]"
project_id_attribute: "user | value.split('$') | value[0]"
resource_id_attribute: "user | value.split('$') | value[0]"
response_entries_key: "summary"
- name: "dynamic.radosgw.api.request.successful_ops.{category}"
sample_type: "gauge"
unit: "request"
value_attribute: "[categories].successful_ops"
url_path: "http://rgw.service.stage.i.ewcs.ch/admin/usage"
module: "awsauth"
authentication_object: "S3Auth"
authentication_parameters: "<access_key>, <secret_key>,<rados_gateway_server>"
user_id_attribute: "user | value.split('$')[0]"
project_id_attribute: "user | value.split('$') | value[0]"
resource_id_attribute: "user | value.split('$') | value[0]"
response_entries_key: "summary"
- name: "dynamic.radosgw.api.bytes_sent.{category}"
sample_type: "gauge"
unit: "request"
value_attribute: "[categories].bytes_sent"
url_path: "http://rgw.service.stage.i.ewcs.ch/admin/usage"
module: "awsauth"
authentication_object: "S3Auth"
authentication_parameters: "<access_key>, <secret_key>,<rados_gateway_server>"
user_id_attribute: "user | value.split('$')[0]"
project_id_attribute: "user | value.split('$') | value[0]"
resource_id_attribute: "user | value.split('$') | value[0]"
response_entries_key: "summary"
- name: "dynamic.radosgw.api.bytes_received.{category}"
sample_type: "gauge"
unit: "request"
value_attribute: "[categories].bytes_received"
url_path: "http://rgw.service.stage.i.ewcs.ch/admin/usage"
module: "awsauth"
authentication_object: "S3Auth"
authentication_parameters: "<access_key>, <secret_key>,<rados_gateway_server>"
user_id_attribute: "user | value.split('$')[0]"
project_id_attribute: "user | value.split('$') | value[0]"
resource_id_attribute: "user | value.split('$') | value[0]"
response_entries_key: "summary"
Handling linked API responses¶
If the consumed API returns a linked response which contains a link to the next response set (page), the Dynamic pollsters can be configured to follow these links and join all linked responses into a single one.
To enable this behavior the operator will need to configure the parameter next_sample_url_attribute that must contain a mapper to the response attribute that contains the link to the next response page. This parameter also supports operations like the others *_attribute dynamic pollster’s parameters.
Examples on how to create a pollster to handle linked API responses are presented as follows:
Example of a simple linked response:
API response:
{ "server_link": "http://test.com/v1/test-volumes/marker=c3", "servers": [ { "volume": [ { "name": "a", "tmp": "ra" } ], "id": 1, "name": "a1" }, { "volume": [ { "name": "b", "tmp": "rb" } ], "id": 2, "name": "b2" }, { "volume": [ { "name": "c", "tmp": "rc" } ], "id": 3, "name": "c3" } ] }
Pollster configuration:
--- - name: "dynamic.linked.response" sample_type: "gauge" unit: "request" value_attribute: "[volume].tmp" url_path: "v1/test-volumes" response_entries_key: "servers" next_sample_url_attribute: "server_link"
Example of a complex linked response:
API response:
{ "server_link": [ { "href": "http://test.com/v1/test-volumes/marker=c3", "rel": "next" }, { "href": "http://test.com/v1/test-volumes/marker=b1", "rel": "prev" } ], "servers": [ { "volume": [ { "name": "a", "tmp": "ra" } ], "id": 1, "name": "a1" }, { "volume": [ { "name": "b", "tmp": "rb" } ], "id": 2, "name": "b2" }, { "volume": [ { "name": "c", "tmp": "rc" } ], "id": 3, "name": "c3" } ] }
Pollster configuration:
--- - name: "dynamic.linked.response" sample_type: "gauge" unit: "request" value_attribute: "[volume].tmp" url_path: "v1/test-volumes" response_entries_key: "servers" next_sample_url_attribute: "server_link | filter(lambda v: v.get('rel') == 'next', value) | list(value) | value[0] | value.get('href')"
