================
 REST API Usage
================

Authentication
==============

By default, no authentication is configured in Gnocchi. You need to provides
these headers in your HTTP requests:

* X-User-Id
* X-Project-Id

The `X-Roles` header can also be provided in order to match role based ACL
specified in `policy.json`, as `X-Domain-Id` to match domain based ACL.

If you enable the OpenStack Keystone middleware, you only need to authenticate
against Keystone and provide `X-Auth-Token` header with a valid token for each
request sent to Gnocchi. The headers mentioned above will be filled
automatically based on your Keystone authorizations.

Metrics
=======

Gnocchi provides an object type that is called *metric*. A metric designates
any thing that can be measured: the CPU usage of a server, the temperature of a
room or the number of bytes sent by a network interface.

A metric only has a few properties: a UUID to identify it, a name, the archive
policy that will be used to store and aggregate the measures.

To create a metric, the following API request should be used:

{{ scenarios['create-metric']['doc'] }}

Once created, you can retrieve the metric information:

{{ scenarios['get-metric']['doc'] }}

To retrieve the list of all the metrics created, use the following request:

{{ scenarios['list-metric']['doc'] }}

.. note::

  Considering the large volume of metrics Gnocchi will store, query results are
  limited to `max_limit` value set in the configuration file. Returned results
  are ordered by metrics' id values. To retrieve the next page of results, the
  id of a metric should be given as `marker` for the beginning of the next page
  of results.

Default ordering and limits as well as page start can be modified
using query parameters:

{{ scenarios['list-metric-pagination']['doc'] }}

It is possible to send measures to the metric:

{{ scenarios['post-measures']['doc'] }}

If there are no errors, Gnocchi does not return a response body, only a simple
status code. It is possible to provide any number of measures.

.. IMPORTANT::

   While it is possible to send any number of (timestamp, value), it is still
   needed to honor constraints defined by the archive policy used by the metric,
   such as the maximum timespan.


Once measures are sent, it is possible to retrieve them using *GET* on the same
endpoint:

{{ scenarios['get-measures']['doc'] }}

Depending on the driver, there may be some lag after POSTing measures before
they are processed and queryable. To ensure your query returns all measures
that have been POSTed, you can force any unprocessed measures to be handled:

{{ scenarios['get-measures-refresh']['doc'] }}

.. note::

   Depending on the amount of data that is unprocessed, `refresh` may add
   some overhead to your query.

The list of points returned is composed of tuples with (timestamp, granularity,
value) sorted by timestamp. The granularity is the timespan covered by
aggregation for this point.

It is possible to filter the measures over a time range by specifying the
*start* and/or *stop* parameters to the query with timestamp. The timestamp
format can be either a floating number (UNIX epoch) or an ISO8601 formated
timestamp:

{{ scenarios['get-measures-from']['doc'] }}

By default, the aggregated values that are returned use the *mean* aggregation
method. It is possible to request for any other method by specifying the
*aggregation* query parameter:

{{ scenarios['get-measures-max']['doc'] }}

The list of aggregation method available is: *mean*, *sum*, *last*, *max*,
*min*, *std*, *median*, *first*, *count* and *Npct* (with 0 < N < 100).

It's possible to provide the `granularity` argument to specify the granularity
to retrieve, rather than all the granularities available:

{{ scenarios['get-measures-granularity']['doc'] }}


Measures batching
=================
It is also possible to batch measures sending, i.e. send several measures for
different metrics in a simple call:

{{ scenarios['post-measures-batch']['doc'] }}

Or using named metrics of resources:

{{ scenarios['post-measures-batch-named']['doc'] }}


Archive Policy
==============

When sending measures for a metric to Gnocchi, the values are dynamically
aggregated. That means that Gnocchi does not store all sent measures, but
aggregates them over a certain period of time. Gnocchi provides several
aggregation methods (mean, min, max, sum…) that are builtin.

An archive policy is defined by a list of items in the `definition` field. Each
item is composed of the timespan and the level of precision that must be kept
when aggregating data, determined using at least 2 of the `points`,
`granularity` and `timespan` fields. For example, an item might be defined
as 12 points over 1 hour (one point every 5 minutes), or 1 point every 1 hour
over 1 day (24 points).

By default, new measures can only be processed if they have timestamps in the
future or part of the last aggregation period. The last aggregation period size
is based on the largest granularity defined in the archive policy definition.
To allow processing measures that are older than the period, the `back_window`
parameter can be used to set the number of coarsest periods to keep. That way
it is possible to process measures that are older than the last timestamp
period boundary.

For example, if an archive policy is defined with coarsest aggregation of 1
hour, and the last point processed has a timestamp of 14:34, it's possible to
process measures back to 14:00 with a `back_window` of 0. If the `back_window`
is set to 2, it will be possible to send measures with timestamp back to 12:00
(14:00 minus 2 times 1 hour).

The REST API allows to create archive policies in this way:

{{ scenarios['create-archive-policy']['doc'] }}

By default, the aggregation methods computed and stored are the ones defined
with `default_aggregation_methods` in the configuration file. It is possible to
change the aggregation methods used in an archive policy by specifying the list
of aggregation method to use in the `aggregation_methods` attribute of an
archive policy.

{{ scenarios['create-archive-policy-without-max']['doc'] }}

The list of aggregation methods can either be:

- a list of aggregation methods to use, e.g. `["mean", "max"]`

- a list of methods to remove (prefixed by `-`) and/or to add (prefixed by `+`)
  to the default list (e.g. `["+mean", "-last"]`)

If `*` is included in the list, it's substituted by the list of all supported
aggregation methods.

Once the archive policy is created, the complete set of properties is computed
and returned, with the URL of the archive policy. This URL can be used to
retrieve the details of the archive policy later:

{{ scenarios['get-archive-policy']['doc'] }}

It is also possible to list archive policies:

{{ scenarios['list-archive-policy']['doc'] }}

Existing archive policies can be modified to retain more or less data depending
on requirements. If the policy coverage is expanded, measures are not
retroactively calculated as backfill to accommodate the new timespan:

{{ scenarios['update-archive-policy']['doc'] }}

.. note::

   Granularities cannot be changed to a different rate. Also, granularities
   cannot be added or dropped from a policy.

It is possible to delete an archive policy if it is not used by any metric:

{{ scenarios['delete-archive-policy']['doc'] }}

.. note::

   An archive policy cannot be deleted until all metrics associated with it
   are removed by a metricd daemon.


Archive Policy Rule
===================

Gnocchi provides the ability to define a mapping called `archive_policy_rule`.
An archive policy rule defines a mapping between a metric and an archive policy.
This gives users the ability to pre-define rules so an archive policy is assigned to
metrics based on a matched pattern.

An archive policy rule has a few properties: a name to identify it, an archive
policy name that will be used to store the policy name and metric pattern to
match metric names.

An archive policy rule for example could be a mapping to default a medium archive
policy for any volume metric with a pattern matching `volume.*`. When a sample metric
is posted with a name of `volume.size`, that would match the pattern and the
rule applies and sets the archive policy to medium. If multiple rules match,
the longest matching rule is taken. For example, if two rules exists which
match `*` and `disk.*`, a `disk.io.rate` metric would match the `disk.*` rule
rather than `*` rule.

To create a rule, the following API request should be used:

{{ scenarios['create-archive-policy-rule']['doc'] }}

The `metric_pattern` is used to pattern match so as some examples,

- `*` matches anything
- `disk.*` matches disk.io
- `disk.io.*` matches disk.io.rate

Once created, you can retrieve the rule information:

{{ scenarios['get-archive-policy-rule']['doc'] }}

It is also possible to list archive policy rules. The result set is ordered by
the `metric_pattern`, in reverse alphabetical order:

{{ scenarios['list-archive-policy-rule']['doc'] }}

It is possible to delete an archive policy rule:

{{ scenarios['delete-archive-policy-rule']['doc'] }}

Resources
=========

Gnocchi provides the ability to store and index resources. Each resource has a
type. The basic type of resources is *generic*, but more specialized subtypes
also exist, especially to describe OpenStack resources.

The REST API allows to manipulate resources. To create a generic resource:

{{ scenarios['create-resource-generic']['doc'] }}

The *id*, *user_id* and *project_id* attributes must be UUID. The timestamp
describing the lifespan of the resource are optional, and *started_at* is by
default set to the current timestamp.

It's possible to retrieve the resource by the URL provided in the `Location`
header.

More specialized resources can be created. For example, the *instance* is used
to describe an OpenStack instance as managed by Nova_.

{{ scenarios['create-resource-instance']['doc'] }}

All specialized types have their own optional and mandatory attributes,
but they all include attributes from the generic type as well.

It is possible to create metrics at the same time you create a resource to save
some requests:

{{ scenarios['create-resource-with-new-metrics']['doc'] }}

To retrieve a resource by its URL provided by the `Location` header at creation
time:

{{ scenarios['get-resource-generic']['doc'] }}

It's possible to modify a resource by re-uploading it partially with the
modified fields:

{{ scenarios['patch-resource']['doc'] }}

And to retrieve its modification history:

{{ scenarios['get-patched-instance-history']['doc'] }}

It is possible to delete a resource altogether:

{{ scenarios['delete-resource-generic']['doc'] }}

It is also possible to delete a batch of resources based on attribute values, and
returns a number of deleted resources.

To delete resources based on ids:

{{ scenarios['delete-resources-by-ids']['doc'] }}

or delete resources based on time:

{{ scenarios['delete-resources-by-time']['doc']}}

.. IMPORTANT::

  When a resource is deleted, all its associated metrics are deleted at the
  same time.

  When a batch of resources are deleted, an attribute filter is required to
  avoid deletion of the entire database.


All resources can be listed, either by using the `generic` type that will list
all types of resources, or by filtering on their resource type:

{{ scenarios['list-resource-generic']['doc'] }}

No attributes specific to the resource type are retrieved when using the
`generic` endpoint. To retrieve the details, either list using the specific
resource type endpoint:

{{ scenarios['list-resource-instance']['doc'] }}

or using `details=true` in the query parameter:

{{ scenarios['list-resource-generic-details']['doc'] }}

.. note::

  Similar to metric list, query results are limited to `max_limit` value set in
  the configuration file.

Returned results represent a single page of data and are ordered by resouces'
revision_start time and started_at values:

{{ scenarios['list-resource-generic-pagination']['doc'] }}

Each resource can be linked to any number of metrics. The `metrics` attributes
is a key/value field where the key is the name of the relationship and
the value is a metric:

{{ scenarios['create-resource-instance-with-metrics']['doc'] }}

It's also possible to create metrics dynamically while creating a resource:

{{ scenarios['create-resource-instance-with-dynamic-metrics']['doc'] }}

The metric associated with a resource can be accessed and manipulated using the
usual `/v1/metric` endpoint or using the named relationship with the resource:

{{ scenarios['get-resource-named-metrics-measures']['doc'] }}

The same endpoint can be used to append metrics to a resource:

{{ scenarios['append-metrics-to-resource']['doc'] }}

.. _Nova: http://launchpad.net/nova

Resource Types
==============

Gnocchi is able to manage resource types with custom attributes.

To create a new resource type:

{{ scenarios['create-resource-type']['doc'] }}

Then to retrieve its description:

{{ scenarios['get-resource-type']['doc'] }}

All resource types can be listed like this:

{{ scenarios['list-resource-type']['doc'] }}

It can also be deleted if no more resources are associated to it:

{{ scenarios['delete-resource-type']['doc'] }}

Attributes can be added or removed:

{{ scenarios['patch-resource-type']['doc'] }}

Creating resource type means creation of new tables on the indexer backend.
This is heavy operation that will lock some tables for a short amount of times.
When the resource type is created, its initial `state` is `creating`. When the
new tables have been created, the state switches to `active` and the new
resource type is ready to be used. If something unexpected occurs during this
step, the state switches to `creation_error`.

The same behavior occurs when the resource type is deleted. The state starts to
switch to `deleting`, the resource type is no more usable. Then the tables are
removed and the finally the resource_type is really deleted from the database.
If some unexpected occurs the state switches to `deletion_error`.

Searching for resources
=======================

It's possible to search for resources using a query mechanism, using the
`POST` method and uploading a JSON formatted query.

When listing resources, it is possible to filter resources based on attributes
values:

{{ scenarios['search-resource-for-user']['doc'] }}

Or even:

{{ scenarios['search-resource-for-host-like']['doc'] }}

Complex operators such as `and` and `or` are also available:

{{ scenarios['search-resource-for-user-after-timestamp']['doc'] }}

Details about the resource can also be retrieved at the same time:

{{ scenarios['search-resource-for-user-details']['doc'] }}

It's possible to search for old revisions of resources in the same ways:

{{ scenarios['search-resource-history']['doc'] }}

It is also possible to send the *history* parameter in the *Accept* header:

{{ scenarios['search-resource-history-in-accept']['doc'] }}

The timerange of the history can be set, too:

{{ scenarios['search-resource-history-partial']['doc'] }}

The supported operators are: equal to (`=`, `==` or `eq`), less than (`<` or
`lt`), greater than (`>` or `gt`), less than or equal to (`<=`, `le` or `≤`)
greater than or equal to (`>=`, `ge` or `≥`) not equal to (`!=`, `ne` or `≠`),
value is in (`in`), value is like (`like`), or (`or` or `∨`), and (`and` or
`∧`) and negation (`not`).

The special attribute `lifespan` which is equivalent to `ended_at - started_at`
is also available in the filtering queries.

{{ scenarios['search-resource-lifespan']['doc'] }}


Searching for values in metrics
===============================

It's possible to search for values in metrics. For example, this will look for
all values that are greater than or equal to 50 if we add 23 to them and that
are not equal to 55. You have to specify the list of metrics to look into by
using the `metric_id` query parameter several times.

{{ scenarios['search-value-in-metric']['doc'] }}

You can specify a time range to look for by specifying the `start` and/or
`stop` query parameter, and the aggregation method to use by specifying the
`aggregation` query parameter.

The supported operators are: equal to (`=`, `==` or `eq`), lesser than (`<` or
`lt`), greater than (`>` or `gt`), less than or equal to (`<=`, `le` or `≤`)
greater than or equal to (`>=`, `ge` or `≥`) not equal to (`!=`, `ne` or `≠`),
addition (`+` or `add`), substraction (`-` or `sub`), multiplication (`*`,
`mul` or `×`), division (`/`, `div` or `÷`). These operations take either one
argument, and in this case the second argument passed is the value, or it.

The operators or (`or` or `∨`), and (`and` or `∧`) and `not` are also
supported, and take a list of arguments as parameters.

Aggregation across metrics
==========================

Gnocchi allows to do on-the-fly aggregation of already aggregated data of
metrics.

It can also be done by providing the list of metrics to aggregate:

{{ scenarios['get-across-metrics-measures-by-metric-ids']['doc'] }}

.. Note::

   This aggregation is done against the aggregates built and updated for
   a metric when new measurements are posted in Gnocchi. Therefore, the aggregate
   of this already aggregated data may not have sense for certain kind of
   aggregation method (e.g. stdev).

By default, the measures are aggregated using the aggregation method provided,
e.g. you'll get a mean of means, or a max of maxs. You can specify what method
to use over the retrieved aggregation by using the `reaggregation` parameter:

{{ scenarios['get-across-metrics-measures-by-metric-ids-reaggregate']['doc'] }}

It's also possible to do that aggregation on metrics linked to resources. In
order to select these resources, the following endpoint accepts a query such as
the one described in `Searching for resources`_.

{{ scenarios['get-across-metrics-measures-by-attributes-lookup']['doc'] }}

It is possible to group the resource search results by any attribute of the
requested resource type, and the compute the aggregation:

{{ scenarios['get-across-metrics-measures-by-attributes-lookup-groupby']['doc'] }}

Similar to retrieving measures for a single metric, the `refresh` parameter
can be provided to force all POSTed measures to be processed across all
metrics before computing the result.

Also aggregation across metrics have different behavior depending
on if boundary are set ('start' and 'stop') and if 'needed_overlap' is set.

If boundaries are not set, Gnocchi makes the aggregation only with points
at timestamp present in all timeseries.

But when boundaries are set, Gnocchi expects that we have certain
percent of timestamps common between timeseries, this percent is controlled
by needed_overlap (defaulted with 100%). If this percent is not reached an
error is returned.

Capabilities
============

The list aggregation methods that can be used in Gnocchi are extendable and
can differ between deployments. It is possible to get the supported list of
aggregation methods from the API server:

{{ scenarios['get-capabilities']['doc'] }}

Status
======
The overall status of the Gnocchi installation can be retrieved via an API call
reporting values such as the number of new measures to process for each metric:

{{ scenarios['get-status']['doc'] }}


Timestamp format
================

Timestamps used in Gnocchi are always returned using the ISO 8601 format.
Gnocchi is able to understand a few formats of timestamp when querying or
creating resources, for example

- "2014-01-01 12:12:34" or "2014-05-20T10:00:45.856219", ISO 8601 timestamps.
- "10 minutes", which means "10 minutes from now".
- "-2 days", which means "2 days ago".
- 1421767030, a Unix epoch based timestamp.