Metrics for monitoring applications and alerting.
Applications/services can have a lot of metrics and one of the greatest
challenges is organizing them. This is what
metrix tries to help with.
Metrix does not aim for providing exact numbers and aims for applications monitoring only.
This crate is in a very early stage and the API might still change. There may be backends provided for monitoring solutions in the future but currently only a snapshot that can be serialized to JSON is provided.
Metrix is based on observations collected while running your application. These observations will then be sent to a backend where the actual metrics(counters etc.) are updated. For the metrics configured a snapshot can be queried.
The primary focus of metrix is to organize these metrics. There are several building blocks available. Most of them can have a name that will then be part of a path within a snapshot.
Labels link observations to panels. Labels can be of any type that
Clone + Eq + Send + 'static. An
enum is a good choice for a
An observation is made somewhere within your application. When an observation is sent to the backend it must have a label attached. This label is then matched against the label of a panel to determine whether an observation is handled for updating or not.
Instruments are gauges, meters, etc. An instrument gets updated by an
observation where an update is meaningful. Instruments are grouped by
You can find instruments in the module
Panel groups instruments under same same label. So each instrument
within a panel will be updated by observations that have the same label as
Lets say you defined a label
OutgoingRequests. If you are interested
in the request rate and the latencies. You would then create a panel with a
OutgoingRequests and add a histogram and a meter.
A cockpit aggregates multiple
Panels. A cockpit can be used to monitor
different tasks/parts of a component or workflow. A cockpit
is bound to a label type.
An example can be that you have service component that calls an external HTTP client. You could be interested in successful calls and failed calls individually. So for both cases you would create a value for your label and then add two panels to the cockpit.
Cockpits are in the module
The most important processor is the
TelemetryProcessor. It has
a label type as a type parameter and consist of a
that sends observations to the backend(used within your app)
and the actual
TelemetryProcessor that forms the backend and
processes observations. The
can own several cockpits for a label type.
There is also a
ProcessorMount that is label agnostic and can group
several processors. It can also have a name that will be included in the
The processors can be found the module
The driver owns processors and asks the owned processors to process their messages. You need to add your processors to a driver to start the machinery. A driver is also a processor which means it can have a name and it can also be part of another hierarchy.
Each driver has its own thread for polling its processors so even when attached to another hierarchy all processors registered with the driver will only be driven by that driver.
Contributing is welcome. Criticism is also welcome!
Metrix is primarily distributed under the terms of both the MIT license and the Apache License (Version 2.0).
Copyright (c) 2018 Christian Douven
Cockpits are used to monitor different aspects of a component
The thing that makes it happen… You need it!
Instruments that track values and/or derive values from observations.
Transmitting observations and grouping metrics.
Pulling data from the backend for monitoring
Changes a value by the given amount (e.g. in a
Decrements a value by one (e.g. in a
Decrements a value by the given amount (e.g. in a
Increments a value by one (e.g. in a
Increments a value by the given amount (e.g. in a
An observation that has been made.
Something that has a title and a description
Something that can react on
Implementors are able to write their current data into given
Transmits telemetry data to the backend.