Skip to content

Latest commit

 

History

History
157 lines (119 loc) · 7.91 KB

ecs.md

File metadata and controls

157 lines (119 loc) · 7.91 KB
Raw

ecs

Queries the ECS Task Metadata Endpoint version 2 for running containers.

The Smart Agent needs to run in the same task with the containers to be monitored to get access to the data through ECS task metadata endpoint.

Configuration from Labels

You must put at least one special label to identify an exposed port on your Docker containers in the ECS task definition since ECS metadata does not contain any port information. You can either specify all of the configuration in container labels, or you can use the more traditional agent configuration with discovery rules and specify configuration overrides with labels.

The config labels are of the form agent.signalfx.com.config.<port number>.<config_key>: <config value>. The <config value> must be a string in a container label, but it will be deserialized as a YAML value to the most appropriate type when consumed by the agent. For example, if you have a Redis container and want to monitor it at a higher frequency than other Redis containers, you could have an agent config that looks like the following:

observers:
 - type: ecs
monitors:
 - type: collectd/redis
   discoveryRule: container_image =~ "redis" && port == 6379
   auth: mypassword
   intervalSeconds: 10

And then launch the Redis container with the label:

agent.signalfx.com.config.6379.intervalSeconds: 1

This would cause the config value for intervalSeconds to be overwritten to the more frequent 1 second interval.

You can also specify the monitor configuration entirely with Docker labels and completely omit monitor config from the agent config. With the agent config:

observers:
 - type: ecs

You can then launch a Redis container with the following labels:

  • agent.signalfx.com.monitorType.6379: collectd/redis
  • agent.signalfx.com.config.6379.auth: mypassword

Which would configure a Redis monitor with the given authentication configuration. No Redis configuration is required in the agent config file.

The distinction is that the monitorType label was added to the Docker container. If a monitorType label is present, no discovery rules will be considered for this endpoint, and thus, no agent configuration can be used anyway.

Multiple Monitors per Port

If you want to configure multiple monitors per port, you can specify the port name in the form <port number>-<port name> instead of just the port number. For example, if you had two different Prometheus exporters running on the same port, but on different paths in a given container, you could provide labels like the following:

 - `agent.signalfx.com.monitorType.8080-app`: `prometheus-exporter`
 - `agent.signalfx.com.config.8080-app.metricPath`: `/appMetrics`
 - `agent.signalfx.com.monitorType.8080-goruntime`: `prometheus-exporter`
 - `agent.signalfx.com.config.8080-goruntime.metricPath`: `/goMetrics`

The name that is given to the port will populate the name field of the discovered endpoint and can be used in discovery rules as such. For example, with the following agent config:

observers:
 - type: ecs
monitors:
 - type: prometheus-exporter
   discoveryRule: name == "app" && port == 8080
   intervalSeconds: 1

And given docker labels as follows (remember that discovery rules are irrelevant to endpoints that specify monitorType labels):

  • agent.signalfx.com.config.8080-app.metricPath: /appMetrics
  • agent.signalfx.com.config.8080-goruntime.metricPath: /goMetrics

Would result in the app endpoint getting an interval of 1 second and the goruntime endpoint getting the default interval of the agent.

Observer Type: ecs

Observer Source Code

Configuration

Config option Required Type Description
metadataEndpoint no string The URL of the ECS task metadata. Default is http://169.254.170.2/v2/metadata, which is hardcoded by AWS for version 2. (default: http://169.254.170.2/v2/metadata)
labelsToDimensions no map of strings A mapping of container label names to dimension names that will get applied to the metrics of all discovered services. The corresponding label values will become the dimension values for the mapped name. E.g. io.kubernetes.container.name: container_spec_name would result in a dimension called container_spec_name that has the value of the io.kubernetes.container.name container label.

Target Variables

The following fields are available on targets generated by this observer and can be used in discovery rules.

Name Type Description
container_name string The first and primary name of the container as it is known to the container runtime (e.g. Docker).
has_port string Set to true if the endpoint has a port assigned to it. This will be false for endpoints that represent a host/container as a whole.
ip_address string The IP address of the endpoint if the host is in the from of an IPv4 address
network_port string An alias for port
private_port string The port that the service endpoint runs on inside the container
public_port string The port exposed outside the container
alternate_port integer Used for services that are accessed through some kind of NAT redirection as Docker does. This could be either the public port or the private one.
container_command string The command used when running the container exposing the endpoint
container_id string The ID of the container exposing the endpoint
container_image string The image name of the container exposing the endpoint
container_labels map of string A map that contains container label key/value pairs. You can use the Contains and Get helper functions in discovery rules to make use of this. See Endpoint Discovery. For containers managed by Kubernetes, this will be set to the pod's labels, as individual containers do not have labels in Kubernetes proper.
container_names list of string A list of container names of the container exposing the endpoint
container_state string The container state, will usually be "running" since otherwise the container wouldn't have a port exposed to be discovered.
discovered_by string The observer that discovered this endpoint
host string The hostname/IP address of the endpoint. If this is an IPv6 address, it will be surrounded by [ and ].
id string
name string A observer assigned name of the endpoint. For example, if using the k8s-api observer, name will be the port name in the pod spec, if any.
orchestrator integer
port integer The TCP/UDP port number of the endpoint
port_labels map of string A map of labels on the container port
port_type string TCP or UDP
target string The type of the thing that this endpoint directly refers to. If the endpoint has a host and port associated with it (most common), the value will be hostport. Other possible values are: pod, container, host. See the docs for the specific observer you are using for more details on what types that observer emits.

Dimensions

These dimensions are added to all metrics that are emitted for this discovery target. These variables are also available to use as variables in discovery rules.

Name Description
container_id The container id of the container running this endpoint.
container_image The image name (including tags) of the running container
container_name The primary name of the running container -- Docker containers can have multiple names but this will be the first name, if any.