Agent for delivering metrics
Yandex Unified Agent is the agent for delivering additional metrics of virtual machines and user applications to Yandex Monitoring.
The agent supports:
- Collecting Linux system metrics (CPU, RAM, disk).
- Collecting metrics in Prometheus format.
- Delivering metrics to Yandex Monitoring.
- File storage for reliable data delivery.
You can find the examples of using the agent in Working with metrics. Learn more about Unified Agent installation methods in Installing and running Yandex Unified Agent. The syntax of the agent's configuration file is described in Configuration.
Concepts
Yandex Unified Agent transmits streaming data represented as messages, e.g., metrics or logs. A data stream in Unified Agent is called a session. Multiple sessions can be running at the same time.
The agent receives messages via the input and delivers them to one or more outputs. Intermediate processing of messages is done by filters; storages save messages to the disk. Inputs, outputs, filters, and storages are implemented using plugins.
Message routing is implemented with the help of delivery routes which consist of an input and a channel. A channel consist of a pipe and a node of one of the following types: output, channel, or splitter. A pipe may contain filters and links to storages. There can be no pipe in the configuration.
You can create named channels and pipes. This will help you avoid duplicate configurations and route messages from multiple inputs to the same output.
How Unified Agent works
Messages
A message is the minimum atomic unit of user information that a system or application sends to Unified Agent.
A message consist of a body, a timestamp, user metadata in key:value format, and a serial number.
Note
For correct data aggregation in monitoring, Integral COUNTER metrics are converted to DGAUGE when uploading to Monitoring.
Sessions
A session is an ordered stream of messages. A session has an ID that must be unique among all the ongoing sessions as well as user metadata in key:value format.
All messages sent during a session contain both message metadata and session metadata.
Types of information to collect (inputs), pipes (filters), interim storage spaces (storages), and outputs are specified in the Unified Agent configuration file.
Inputs
An input is used by the agent to receive the messages transmitted during sessions. An input can contain the session infrastructure settings used to configure various limits.
See the list of available inputs.
Outputs
Outputs are used by the agent to send messages to third-party systems. The currently supported outputs are the yc_metrics output, which writes a metric to the Yandex Monitoring API, and several debug outputs.
See also the list of available outputs.
Filters
Filters are designed to discard, convert, and aggregate messages.
Filter types:
- regular: Handle each message separately.
- cumulative: Transform a set of input messages into one output message.
See also the list of available filters.
Storages
Storages are intended to store messages from their receipt as an input until their transmission to the output.
Using a storage can help you avoid data loss if the agent fails to write the data to the specified output. This may happen due to network issues or destination API unavailability.
See also the storage configuration guide.
Routing
Pipes
Pipes contain:
- Sequence of filters that messages pass through.
- Link to the storage where the messages will be stored before their transmission to the output.
Pipes can be named.
See also the pipe configuration guide.
Channels
Channels group a pipe with a node, one of an output, a named channel, or a splitter.
Splitters let you specify a set of channels, copying incoming messages to each of them. You can use splitters to duplicate streams to different outputs. By combining splitters and filters, you can send different messages to different channels based on particular attributes such as metadata.
Channels can be named.
See the named channel configuration guide.
Routes
Routes combine an input and a channel.
Routes, channels, and pipes enable you to set up any message processing tree.
See the diagram for the relations between the items of the message processing tree.