network-observer

README

Network Observer

The network observer is an application that attaches to the skupper router network in order to expose skupper network telemetry. When installed alongside a skupper site it will collect operational data from ALL sites in the network and expose them via the API and metrics that back the Skupper Console web application.

Deployment

The Network Observer can be deployed with the network-observer Helm Chart.

API

The Skupper Network Console HTTP API is described in an openapi 3.0 specification file inside the spec directory. To view the API, either import this document into an online swagger editor, or start the collector locally and access the Swagger UI at http://localhost:8080/swagger/.

To edit the specification using a hosted (swagger editor)[https://editor-next.swagger.io/], open the editor in your browser, and import the spec by URL (File -> Import URL) from https://raw.githubusercontent.com/skupperproject/skupper/v2/cmd/network-observer/spec/openapi.yaml.

Metrics

The network console collector exposes a set of Prometheus metrics alongside the API at /metrics.

Operational Site Metrics

This set of metrics exposes coarse site-level details pertaining to the operation and topology of the skupper network.

• skupper_site_info: Metadata about the active sites in the network. Labels are site_id, name and version.
• skupper_routers_total: Number of active routers in each site. Labels are site_id and mode.
• skupper_site_links_total: Number of links from each site. Counts all router links so may not map directly to the number of links.skupper.io resources in the case of HA deployments. Labels are site_id, status and role.
• skupper_site_link_errors_total: Count of link connection errors from routers in a site. Labels are site_id and role.
• skupper_site_listeners_total: Number of listeners on each site. Counts all listeners so may not map directly to the number of listeners.skupper.io resources in the case of HA deployments. Label is site_id.
• skupper_site_connectors_total: Number of connectors on each site. Counts all connectors so may not map directly to the number of connectors.skupper.io resources in the case of HA deployments or connectors with a selector matching any number of pods. Label is site_id.

Application Network Traffic Metrics

This set of metrics exposes details about service traffic though the skupper network. These metrics have a shared set of labels that allow us to expose traffic patterns on a per client/service basis.

Signals:

All metric names are prefixed with the skupper namespace.

metric name	description
connections_opened_total	Number of connections opened through the application network
connections_closed_total	Number of connections opened through the application network that have been closed
sent_bytes_total	Bytes sent through the application network from client to service
received_bytes_total	Bytes sent through the application network back from service to client

Dimensions:

label name	description
source_site_id	ID of the site where the connection was established
source_site_name	Name of the source site
dest_site_id	ID of the site where the connection exited the skupper network through a connector
dest_site_name	Name of the destination site
source_component_id	ID of the process group for the client process establishing the connection
source_component_name	Name of the source process group
dest_component_id	ID of the process group for the server process
dest_component_name	Name of the destination process group
routing_key	The routing key of the service
protocol	The protocol used in the exchange (TCP)
source_process_name	The name of the client process as reflected in the collector API
dest_process_name	The name of the server process as reflected in the collector API

Application Network Request Traffic Metrics (Application Layer)

A proposed set of metrics that expose details about HTTP and HTTP/2 exchanges when enabled in the router.

Signals:

Like application traffic metrics, these are prefixed with skupper.

metric name	description
requests_total	Counter incremented for each request handled through the skupper network

Dimensions:

Request metrics share the same set of dimensions as the network traffic metrics above, but includes several additional labels.

label name	description
method	HTTP request method
code	HTTP response code class (for example, a response code 201 would be counted towards code='2xx')

Internal Metrics

We expose a set of metrics prefixed skupper_internal to help us observe the collector itself. Moving forward we may end up recommending anyone scraping skupper metrics from the collector exclude these when storage space is a concern. All internal metrics should be considered unstable.

notable internal metrics to the user

Related to Network Traffic metrics (and sharing the same label set.)

metric name	description
latency_seconds	Histogram of connection latency observations - the difference between TTFB on the listener and connector sides. TCP only.

Primarily for backward compatibility with the skupper v1 network console. New users should prefer the latency_seconds signal. The legacy_flow_latency_microseconds metric exposes the Time to First Byte (TTFB) as observed by skupper from the client (listener) and server (connector) sides on TCP connections. This metric shares the same standard set of labels described in the Application Network Traffic section, but adds a direction label to differentiate the listener and connector side TTFB.

metric name	description
legacy_flow_latency_microseconds	Histogram of connection time to first byte observations in microseconds. TCP only.