gNMI Telemetry Exporter for Prometheus

GtExporter is a tool for subscribing and exporting gNMI streaming telemetries to Prometheus in a YANG "data-model-aware" way. The intended usage is to gather and store network devices' operational state metrics, but it should work with any gNMI compliant device that implements any OpenConfig compliant YANG data model.

Overview

The intent of this project is to create an easy-to-scale, vendor-agnostic Prometheus exporter for gNMI streaming telemetries. To satisfy these two requirements, the relevant design choice is to put the YANG data model at the center: Incoming gNMI streams are treated as part of this known data structure and processed accordingly. The "awareness" of the underlying data model enables the export of a stable and predictable metric set for Prometheus and, in turn, allows horizontal scaling of the monitoring system with a robust vendor-neutral approach.

The application design is built around these ideas:

• Several app instances can be spawned on different machines across the network. The produced metrics will seamlessly merge into one or more Prometheus server instances.
• A given instance can run several gNMI clients concurrently, one for each monitored device.
• Each gNMI client can run several schema plugins, one for each set of the YANG path on interest.
• When Prometheus server scrapes, the exporter collects all the received metrics from all the schema plugins and exposes them via its own client library.

Hence, the central component of the app is the schema plugin:

• A schema plugin is responsible for subscribing and rendering a set of schema paths from a selected YANG data model.
• Internally, it is further divided into two components:
  • The Parser decodes and loads the received gNMI notifications into the related GoStruct data structure.
  • The Formatter reads the GoStruct and builds up the metrics to be exported when Prometheus asks for them.
• The GoStruct is a "data container" that represents the structure of the selected YANG data model. It is generated using the Openconfig yGot project and takes as input for code generation the actual .yang files published by the device vendor or the OpenConfig community.

Getting Started

Download from GitHub

Binaries are available for download here
The mandatory argument is a config file: gtexporter-<OS>-<ARCH> -config <path/to/config/file>.

Build from sources

Requires go 1.23.2 or higher and make.

git clone https://github.com/automixer/gtexporter.git
cd gtexporter
make

The binary executable will be created into the build/ project folder. The mandatory argument is a config file: gtexporter -config <path/to/config/file>.

Use the provided Docker image

Requires Docker Engine. The supported platforms are linux/amd64 and linux/arm64.

docker run -p 9456:9456/tcp --mount type=bind,source=.,target=/etc/gte/ automixer/gtexporter -config /etc/gte/config.yaml

The mandatory argument is a config file named config.yaml into the current folder.

The Configuration File

This configuration file template and this addendum describes the supported configuration keys.
The file is subdivided into three sections:

1. The global section contains application-wide settings like the Prometheus client listen address and port.
2. The device_template section contains the settings shared among all devices. This section can help keep the configuration file small and readable. It can be empty, and any key can be overridden by a more specific setting into the devices section.
3. The devices section contains the device-specific settings, like the device name, IP address and, port of the target. It inherits the contents of the device_template section and, if a key is present on both, the more specific wins (i.e.: the one coming from devices).

A Simple Config File Example

# These keys are application-wide.
global:
  instance_name: test_gte
  metric_prefix: gnmi
  listen_address: 0.0.0.0
  listen_port: 9456
  scrape_interval: 1m

# These keys are shared among all devices.
device_template:
  port: 6030
  user: admin
  password: admin

# These keys are device-specific and take precedence over the device_template section.
devices:
  - name: Router1
    address: r1.example.com
    plugins: [oc_interfaces,oc_lldp]
    port: 50051

  - name: Router2
    address: 192.0.2.10
    plugins: [oc_interfaces]

Supported Schema Plugins

These are the currently available schema plugins:

oc_interfaces

This plugin is based on the openconfig-interfaces data model.
Subscribe to these schema paths:

1. /interfaces/interface/state/
2. /interfaces/interface/aggregation/state/
3. /interfaces/interface/subinterfaces/subinterface/state/

Produces two Prometheus metrics:

1. <configured_metric_prefix>_oc_if_total{}.
2. <configured_metric_prefix>_oc_if_gauges{}.

oc_lldp

This plugin is based on the openconfig-lldp data model.
Subscribe to this schema path:

1. /lldp/interfaces/interface/neighbors/neighbor/state/

Produces one Prometheus metrics:

1. <configured_metric_prefix>_oc_lldp_if_nbr_gauges{}.
   LLDP must be enabled on the target devices.

Self-Monitoring Services

In addition to the schema plugins, GtExporter emits several self-monitoring metrics to keep track of the app's health and operational state.
These metrics are:

1. <configured_metric_prefix>_gnmi_client_total{}: These counters describe the state of the underlying gNMI client instances.
2. <configured_metric_prefix>_gnmi_client_gauges{}: These gauges describe the state of the underlying gNMI client instances.
3. <configured_metric_prefix>_plugin_formatter_gauges{}: These gauges describe the operational state of the running plugin's formatters.
4. <configured_metric_prefix>_plugin_parser_total{}: These counters describe the operational state of the running plugin's parsers.
5. The default Go Runtime Metrics exported by the Prometheus client library.

Caveats

The global:scrape_interval setting

This config key plays an important role. Together with the device:oversampling it is used to compute the actual sample interval of the gNMI subscription. It must match the configured Prometheus scrape interval.
Prometheus scrapes and gNMI sample interval are two asynchronous loops that can't be synchronized. To keep acquisition errors within acceptable levels, the gNMI subscription loop should run at least two times faster than Prometheus. This ensures that even if a scrape occurs in the middle of a gNMI batch delivery, the collected samples always fall within the Prometheus scrape interval.
The formula used to compute the gNMI sample interval is: sample_interval=scrape_interval/oversampling. The default device:oversampling value is 2.

Cache mode and max_life

By default, GtExporter does not cache any data. The device:mode key can be used to force persistence of
the yGot GoStruct over time. This setting implies that the monitored device must implement the gNMI delete messages mechanism, to avoid a continuously growing GoStruct.
The device:max_life config sets a time limit on the gNMI subscription. When max_life expires, the session is torn down and re-established, forcing a cache flush event. This setting can be useful in keeping the GoStruct size under control.

The device:desc_sanitize setting

Descriptions are user defined strings contained into the device configuration. Since descriptions are often used as Prometheus labels, not all characters are valid. The device:desc_sanitize config key is a regexp pattern used to remove unsupported characters by Prometheus.

License

Licensed under MIT license. See LICENSE.