NOTE: This repository is deprecated. Please see examples in metrics-discovery-release

CF-pushable Telegraf

Overview

This repo provides a reference architecture allowing Cloud Foundry users to leverage agent based architecture for collecting metrics. It is provided for reference only and is not intended to be used in production.

This examples utilizes the metrics-discovery-release to gather all the component metrics endpoints published to the NATs queue "metrics.scrape_targets".

Minimum Requirements

• BOSH CLI v6.1.1
• CF CLI v6.47.0
• CF Deployment v12.21.0 NOTE: This example will not with in a TAS environment due how certificates are managed in Ops Manager
• Credhub CLI version 2.5.3

Components

There are two components: telegraf and telegraf-config-sidecar.

The telegraf-config-sidecar will generate the Prometheus input config from the metrics endpoints it gets from the NATs queue and add that to the telegraf configuration. The NATs queue is checked every 15 seconds to see if any updates are needed to the input config. Every 45 seconds, the the sidecar restarts telegraf to pick up these new metrics endpoints.

Usage

1. Add output plugin(s) to telegraf.conf
2. cf, credhub, and bosh target the desired environment
3. ./push.sh

Scaling

Scaling of telegraf can be handled directly by Diego. That said, this implementation will not ensure "only once" delivery. Scaling to two instances will result in duplicate metrics, three instances will triple the metrics etc.

Security Group Restrictions

Due to application security groups, Telegraf cannot scrape the Diego Cell it is running on. This means there must be at least 2 instances of Telegraf (on different diego cells) in order to ingest all metrics.

Dropping metrics

This promQL query will allow you to determine if a specific output is not keeping up. A good number to shoot for is 99% of metrics getting through. Just replace my-output-plugin with the name of your output e.g. datadog.

100 * (1 -
  rate(internal_write_metrics_dropped{output="my-output-plugin"}[1m]) /
  rate(internal_write_metrics_written{output="my-output-plugin"}[1m]))

If this number is below 99%, try increasing the metric_buffer_limit.