gecko-dev/services/docs/metrics.rst
Gregory Szorc b931cfa945 NO BUG - Fix reStructuredText warnings
Sphinx has been complaining about a number of reStructuredText warnings
for a while. Fix all the ones in .rst files.

Not asking for review because this is docs only and changing .rst files
can't break anything important.

DONTBUILD (NPOTB)
2015-03-01 22:51:32 -08:00

131 lines
4.5 KiB
ReStructuredText

.. _services_metrics:
============================
Metrics Collection Framework
============================
The ``services/metrics`` directory contains a generic data metrics
collecting and persisting framework for Gecko applications.
Overview
========
The Metrics framework by itself doesn't do much: it simply provides a
generic mechanism for collecting and persisting data. It is up to users
of this framework to drive collection and do something with the obtained
data. A consumer of this framework is :ref:`healthreport`.
Relationship to Telemetry
-------------------------
Telemetry provides similar features to code in this directory. The two
may be unified in the future.
Usage
=====
To use the code in this directory, import Metrics.jsm. e.g.
Components.utils.import("resource://gre/modules/Metrics.jsm");
This exports a *Metrics* object which holds references to the main JS
types and functions provided by this feature. Read below for what those
types are.
Metrics Types
=============
``Metrics.jsm`` exports a number of types. They are documented in the
sections below.
Metrics.Provider
----------------
``Metrics.Provider`` is an entity that collects and manages data. Providers
are typically domain-specific: if you need to collect a new type of data,
you create a ``Metrics.Provider`` type that does this.
Metrics.Measurement
-------------------
A ``Metrics.Measurement`` represents a collection of related pieces/fields
of data.
All data recorded by the metrics framework is modeled as
``Metrics.Measurement`` instances. Instances of ``Metrics.Measurement``
are essentially data structure descriptors.
Each ``Metrics.Measurement`` consists of a name and version to identify
itself (and its data) as well as a list of *fields* that this measurement
holds. A *field* is effectively an entry in a data structure. It consists
of a name and strongly enumerated type.
Metrics.Storage
---------------
This entity is responsible for persisting collected data and state.
It currently uses SQLite to store data, but this detail is abstracted away
in order to facilitate swapping of storage backends.
Metrics.ProviderManager
-----------------------
High-level entity coordinating activity among several ``Metrics.Provider``
instances.
Providers and Measurements
==========================
The most important types in this framework are ``Metrics.Provider`` and
``Metrics.Measurement``, henceforth known as ``Provider`` and
``Measurement``, respectively. As you will see, these two types go
hand in hand.
A ``Provider`` is an entity that *provides* data about a specific subsystem
or feature. They do this by recording data to specific ``Measurement``
types. Both ``Provider`` and ``Measurement`` are abstract base types.
A ``Measurement`` implementation defines a name and version. More
importantly, it also defines its storage requirements and how
previously-stored values are serialized.
Storage allocation is performed by communicating with the SQLite
backend. There is a startup function that tells SQLite what fields the
measurement is recording. The storage backend then registers these in
the database. Internally, this is creating a new primary key for
individual fields so later storage operations can directly reference
these primary keys in order to retrieve data without having to perform
complicated joins.
A ``Provider`` can be thought of as a collection of ``Measurement``
implementations. e.g. an Addons provider may consist of a measurement
for all *current* add-ons as well as a separate measurement for
historical counts of add-ons. A provider's primary role is to take
metrics data and write it to various measurements. This effectively
persists the data to SQLite.
Data is emitted from providers in either a push or pull based mechanism.
In push-based scenarios, the provider likely subscribes to external
events (e.g. observer notifications). An event of interest can occur at
any time. When it does, the provider immediately writes the event of
interest to storage or buffers it for eventual writing. In pull-based
scenarios, the provider is periodically queried and asked to populate
data.
SQLite Storage
==============
``Metrics.Storage`` provides an interface for persisting metrics data to a
SQLite database.
The storage API organizes values by fields. A field is a named member of
a ``Measurement`` that has specific type and retention characteristics.
Some example field types include:
* Last text value
* Last numeric value for a given day
* Discrete text values for a given day
See ``storage.jsm`` for more.