Collector configuration¶
Common options¶
Options common to all collectors are specified in the [collect]
section of
the configuration file. The following options are available:
collector
: Defaults tognocchi
. The name of the collector to load. Must be one of [gnocchi
,monasca
,prometheus
].period
: Default to 3600. Duration (in seconds) of the collect period.wait_periods
: Defaults to 2. Periods to wait before the current timestamp. This is done to avoid missing some data that hasn’t been retrieved by the data source yet.metrics_conf
: Defaults to/etc/cloudkitty/metrics.yml
. Path of the metric collection configuration file. See “Metric collection” section below for details.scope_key
: Defaults toproject_id
. Key at which the scope can be found. The scope defines how data collection is split between the processors.
Collector-specific options¶
Collector-specific options must be specified in the
collector_{collector_name}
section of cloudkitty.conf
.
Gnocchi¶
Section: collector_gnocchi
.
gnocchi_auth_type
: Defaults tokeystone
. Defines what authentication method should be used by the gnocchi collector. Must be one ofbasic
(for gnocchi basic authentication) orkeystone
(for classic keystone authentication). Ifkeystone
is chosen, credentials can be specified in a section pointed at by theauth_section
parameter.gnocchi_user
: For gnocchi basic authentication only. The gnocchi user.gnocchi_endpoint
: For gnocchi basic authentication only. The gnocchi endpoint.interface
: Defaults tointernalURL
. For keystone authentication only. The interface to use for keystone URL discovery.region_name
: Defaults toRegionOne
. For keystone authentication only. Region name.
Monasca¶
Section: collector_monasca
.
interface
: Defaults tointernal
. The interface to use for keystone URL discovery.monasca_service_name
: Defaults tomonasca
. Name of the Monasca service in Keystone.
Note
By default, cloudkitty retrieves all metrics from Monasca in the
project it is identified in. However, some metrics may need to be
fetched from another tenant (for example if ceilometer is publishing
metrics to monasca in the service
tenant but monasca-agent is
publishing metrics to the admin
tenant). See the monasca-specific
section in “Metric collection” below for details on how to configure
this.
Prometheus¶
Section collector_prometheus
.
prometheus_url
: Prometheus HTTP API URL.prometheus_user
: For HTTP basic authentication. The username.prometheus_password
: For HTTP basic authentication. The password.cafile
: Option to allow custom certificate authority file.insecure
: Option to explicitly allow untrusted HTTPS connections.
Metric collection¶
Metric collection is highly configurable in cloudkitty. In order to keep the
main configuration file as clean as possible, metric collection is configured
in a yaml file. The path to this file defaults to
/etc/cloudkitty/metrics.yml
, but can be configured:
[collect]
metrics_conf = /my/custom/path.yml
Minimal Configuration¶
This config file has the following format:
metrics: # top-level key
metric_one: # metric name
unit: squirrel
groupby: # attributes by which metrics should be grouped
- id
metadata: # additional attributes to retrieve
- color
At the top level of the file, a metrics
key is required. It contains a dict
of metrics to collect, each key of the dict being the name of a metric as it is
called in the datasource (volume.size
or image.size
for example).
For each metric, the following attributes are required:
unit
: the unit in which the metric will be stored after conversion. This is just an indication for humans and has absolutely no impact on metric collection, conversion or rating.groupby
: A list of attributes by which metrics should be grouped on collection. These will allow to re-group data when it is retrieved through the v2 API. A typical usecase would be to group data by ID, project ID, domain ID and user ID on collection, but only by user ID on retrieval.metadata
: A list of additional attributes that should be collected for the given metric. These can be used for rating rules and will appear in monthly reports. However, it is not possible to group on these attributes. If you need to group on ametadata
attribute, move it to thegroupby
list.
Note
The scope_key
is automatically added to groupby
.
Optional parameters¶
Unit conversion¶
If you need to convert the collected qty (from MiB to GiB for example), it can
be done with the factor
and offset
options. factor
defaults to 1
and offset
to 0. These options are used to calculate the final result with
the following formula: qty = collected_qty * factor + offset
.
Note
factor
and offset
can be floats, integers or fractions.
Example from the default configuration file, conversion from B to MiB for the
image.size
metric:
metrics:
image.size:
groupby:
- id
metadata:
- disk_format
unit: MiB # Final unit
factor: 1/1048576 # Dividing by 1024 * 1024
Note
Here we don’t add anything, so there is no need to specify offset
.
Quantity mutation¶
It is also possible to mutate the collected qty with the mutate
option.
Four values are accepted for this parameter:
NONE
: This is the default. The collected data is not modifed.CEIL
: The qty is rounded up to the closest integer.FLOOR
: The qty is rounded down to the closest integer.NUMBOOL
: If the collected qty equals 0, leave it at 0. Else, set it to 1.NOTNUMBOOL
: If the collected qty equals 0, set it to 1. Else, set it to 0.
Warning
Quantity mutation is done after conversion. Example:
factor: 10
mutate: CEIL
In consequence, the configuration above will convert 9.9 to 99 (9.9 -> 99 -> 99) and not to 100 (9.9 -> 10 -> 100)
A typical usecase for the NUMBOOL
conversion would be instance uptime
collection with the gnocchi collector: In order to know if an instance is
running or paused, you can use the cpu
metric. This metric is at
0 when the instance is paused. Thus, the qty is mutated to a NUMBOOL
because the cpu
metric always represents one instance. Rating rules are
then defined based on the instance metadata. Example:
metrics:
cpu:
unit: instance
mutate: NUMBOOL
groupby:
- id
metadata:
- flavor_id
The NOTNUMBOOL
mutator is useful for status-like metrics where 0 denotes
the billable state. For example the following Prometheus metric has value of 0
when the instance is in ACTIVE state but 4 if the instance is in ERROR state:
metrics:
openstack_nova_server_status:
unit: instance
mutate: NOTNUMBOOL
groupby:
- id
metadata:
- flavor_id
Display name¶
Sometimes, you’ll want to use another name for a metric, either to shorten it a
bit or to make it more explicit. For example, the cpu
metric from the
previous section could be called instance
. That’s what the alt_name
option does:
metrics:
cpu:
unit: instance
alt_name: instance
mutate: NUMBOOL
groupby:
- id
metadata:
- flavor_id
Collector-specific configuration¶
Some collectors require extra options. These must be specified through the
extra_args
option. Some options have defaults, other must be systematically
specified. The extra args for each collector are detailed below.
Gnocchi¶
Note
In order to retrieve metrics from Gnocchi, Cloudkitty uses the
dynamic aggregates endpoint. It builds an operation of the following
format: (aggregate RE_AGGREGATION_METHOD (metric METRIC_NAME
AGGREGATION_METHOD))
. This means “retrieve all aggregates of type
AGGREGATION_METHOD
for the metric named METRIC_NAME
and
re-aggregate them using RE_AGGREGATION_METHOD
”.
By default, the re-aggregation method defaults to the aggregation method.
Setting the re-aggregation method to a different value than the
aggregation method is useful when the granularity of the aggregates
does not match CloudKitty’s collect period, or when using
rate:
aggregation, as you’re probably don’t want a rate of rates,
but rather a sum or max of rates.
resource_type
: No default value. The resource type the current metric is bound to.resource_key
: Defaults toid
. The attribute containing the unique resource identifier. This is an advanced option, do not modify it unless you know what you’re doing.aggregation_method
: Defaults tomax
. The aggregation method to use when retrieving measures from gnocchi. Must be one ofmin
,max
,mean
,rate:min
,rate:max
,rate:mean
.re_aggregation_method
: Defaults toaggregation_method
. The re_aggregation method to use when retrieving measures from gnocchi.force_granularity
: Defaults to0
. If > 0, this granularity will be used for metric aggregations. Else, the lowest available granularity will be used (meaning the granularity covering the longest period).use_all_resource_revisions
: Defaults toTrue
. This option is useful when using Gnocchi with the patch introduced via https://github .com/gnocchixyz/gnocchi/pull/1059. That patch can cause queries to return more than one entry per granularity (timespan), according to the revisions a resource has. This can be problematic when using the ‘mutate’ option of Cloudkitty. This option to allow operators to discard all datapoints returned from Gnocchi, but the last one in the granularity queried by CloudKitty for a resource id. The default behavior is maintained, which means, CloudKitty always use all of the data points returned.custom_query
: Provide means for operators to customize the aggregation query executed against Gnocchi. By default we use the following(aggregate RE_AGGREGATION_METHOD (metric METRIC_NAME AGGREGATION_METHOD))
. Therefore, this option enables operators to take full advantage of operations available in Gnocchi such as any arithmetic operations, logical operations and many others. When using a custom aggregation query, you can keep the placeholdersRE_AGGREGATION_METHOD
,AGGREGATION_METHOD
, andMETRIC_NAME
: they will be replaced at runtime by values from the metric configuration.One example use case is metrics that are supposed to be always growing values, such as RadosGW usage data. The usage data is affected by usage data trimming on RadosGW, which can lead to swaps (meaning, that the right side value of the series is smaller than the left side value) in the data series in Gnocchi. Therefore, to handle this situation one could, for instance, use the following custom query:
(div (+ (aggregate RE_AGGREGATION_METHOD (metric METRIC_NAME AGGREGATION_METHOD)) (abs (aggregate RE_AGGREGATION_METHOD (metric METRIC_NAME AGGREGATION_METHOD)))) 2)
: this custom query would return0
when the value of the series swap.
Monasca¶
resource_key
: Defaults toresource_id
. The attribute containing the unique resource identifier. This is an advanced option, do not modify it unless you know what you’re doing.aggregation_method
: Defaults tomax
. The aggregation method to use when retrieving measures from monasca. Must be one ofmin
,max
,mean
.forced_project_id
: Defaults to None. Force the given metric to be fetched from a specific tenant instead of the one cloudkitty is identified in. For example, if cloudkitty is identified in theservice
project, but needs to fetch a metric from theadmin
project, its ID should be specified through this option. If this option is set toSCOPE_ID
, the metric will be fetched from the current project (this assumes that scopes are configured to be projects/tenants).
Prometheus¶
aggregation_method
: Defaults tomax
. The aggregation method to use when retrieving measures from prometheus. Must be one ofavg
,min
,max
,sum
,count
,stddev
,stdvar
.query_function
: Optional argument. The function to apply to an instant vector after theaggregation_method
orrange_function
has altered the data. Must be one ofabs
,ceil
,exp
,floor
,ln
,log2
,log10
,round
,sqrt
. For more information on these functions, you can check this pagequery_prefix
: Optional argument. An arbitrary prefix to add to the Prometheus query generated by CloudKitty, separated by a space.query_suffix
: Optional argument. An arbitrary suffix to add to the Prometheus query generated by CloudKitty, separated by a space.range_function
: Optional argument. The function to apply instead of the implicit{aggregation_method}_over_time
. Must be one ofchanges
,delta
,deriv
,idelta
,irange
,irate
,rate
. For more information on these functions, you can check this page