neo-agent

README

SignalFx Smart Agent

The SignalFx Smart Agent is a metric agent written in Go for monitoring infrastructure and application services in a variety of different environments. It is meant as a successor to our previous collectd agent, but still uses that internally -- so any existing Python or C-based collectd plugins will still work without modification.

• Concepts
• Installation
• Configuration
• Logging
• Proxy Support
• Diagnostics
• Development

Concepts

The agent has three main components:

1. Observers that discover applications and services running on the host
2. Monitors that collect metrics from the host and applications
3. The Writer that sends the metrics collected by monitors to SignalFx.

Observers

Observers watch the various environments that we support to discover running services and automatically configure the agent to send metrics for those services.

For a list of supported observers and their configurations, see Observer Config.

Monitors

Monitors collect metrics from the host system and services. They are configured under the monitors list in the agent config. For application-specific monitors, you can define discovery rules in your monitor configuration. A separate monitor instance is created for each discovered instance of applications that match a discovery rule. See Auto Discovery for more information.

Many of the monitors are built around collectd, an open source third-party monitor, and use it to collect metrics. Some other monitors do not use collectd. However, either type is configured in the same way.

For a list of supported monitors and their configurations, see Monitor Config.

The agent is primarily intended to monitor services/applications running on the same host as the agent. This is in keeping with the collectd model. The main issue with monitoring services on other hosts is that the host dimension that collectd sets on all metrics will currently get set to the hostname of the machine that the agent is running on. This allows everything to have a consistent host dimension so that metrics can be matched to a specific machine during metric analysis.

Writer

The writer collects metrics emitted by the configured monitors and sends them to SignalFx on a regular basis. By default it batches metrics for 1 second before sending. There are a few things that can be configured in the writer, but this is generally unnecessary.

Installation

The agent is available for Linux in both a containerized and standalone form. Whatever form you use, the dependencies are completely bundled along with the agent, including a Java JRE runtime and a Python runtime, so there are no additional dependencies required. This means that the agent should work on any relatively modern Linux distribution (kernel version 2.6+).

Deployment

We support the following deployment/configuration management tools to automate the installation process. See Bundles for a list of underlying packages for the agent.

Installer Script

For non-containerized environments, there is a convenience script that you can run on your host to install the agent package. This is useful for testing and trials, but for full-scale deployments you will probably want to use a configuration management system like Chef or Puppet. You can view the source for the installer script and use it on your hosts by running:

curl -sSL https://dl.signalfx.com/signalfx-agent.sh > /tmp/signalfx-agent.sh
sudo sh /tmp/signalfx-agent.sh YOUR_SIGNALFX_API_TOKEN

Chef

We offer a Chef cookbook to install and configure the agent. See the cookbook source and on the Chef Supermarket.

Puppet

We also offer a Puppet manifest to install and configure the agent. See the manifest source and on the Puppet Forge.

Ansible

We also offer an Ansible Role to install and configure the Smart Agent. See the role source.

Salt

We also offer a Salt Formula to install and configure the Smart Agent. See the formula source.

Kubernetes

See our Kubernetes setup instructions and the documentation on Monitoring Kubernetes for more information.

Bundles

We offer the agent in the following forms:

Docker Image

We provide a Docker image at quay.io/signalfx/signalfx-agent. The image is tagged using the same agent version scheme.

If you are using Docker outside of Kubernetes, you can run the agent in a Docker container and still gather metrics on the underlying host by running it with the following flags:

$ docker run \
    --name signalfx-agent \
    --pid host \
    --net host \
    -v /:/hostfs:ro \
    -v /var/run/docker.sock:/var/run/docker.sock:ro \
    -v /etc/signalfx/:/etc/signalfx/:ro \
    quay.io/signalfx/signalfx-agent:<version>

This assumes you have the agent config in the conventional directory (/etc/signalfx) on the root mount namespace.

If you have the Docker API available through the conventional UNIX domain socket, you should mount that in to be able to use the docker-container-stats monitor.

It is necessary to mount in the host root filesystem at /hostfs in order to get disk usage metrics for the host filesystems using the collectd/df. You will need to set the hostFSPath: /hostfs config option on that monitor to make it use this non-default path.

The only other special config you will need is the etcPath: /hostfs/etc option under the collectd/signalfx-metadata monitor config. This tells it where to find certain files like /etc/os-release that are used to generate host metadata such as the Linux distro and version.

You may also want to use the Docker observer to automatically discover other containers running in the same Docker engine.

Debian Package

We provide a Debian package repository that you can make use of with the following commands:

curl -sSL https://dl.signalfx.com/debian.gpg > /etc/apt/trusted.gpg.d/signalfx.gpg
echo 'deb https://dl.signalfx.com/debs/signalfx-agent/final /' > /etc/apt/sources.list.d/signalfx-agent.list
apt-get update
apt-get install -y signalfx-agent

RPM Package

We provide a RHEL/RPM package repository that you can make use of with the following commands:

cat <<EOH > /etc/yum.repos.d/signalfx-agent.repo
[signalfx-agent]
name=SignalFx Agent Repository
baseurl=https://dl.signalfx.com/rpms/signalfx-agent/final
gpgcheck=1
gpgkey=https://dl.signalfx.com/yum-rpm.key
enabled=1
EOH

yum install -y signalfx-agent

Standalone tar.gz

If you don't want to use a distro package, we offer a .tar.gz that can be deployed to the target host. This bundle is available for download on the Github Releases Page for each new release.

To use the bundle:

1. Unarchive it to a directory of your choice on the target system.

2. Ensure a valid configuration file is available somewhere on the target system. The main thing that the distro packages provide -- but that you will have to provide manually with the bundle -- is a run directory for the agent to use. Since you aren't installing from a package, there are three config options that you will especially want to consider:

• diagnosticsSocketPath - This is the full path to a UNIX domain socket that the agent will listen on so that the signalfx-agent status command can read diagnostic information from a running agent. It can be blank if you don't desire that functionality.

• internalMetricsSocketPath - This is the full path to a UNIX domain socket that the agent will listen on for requests from the internal-metrics monitor to gather internal metrics about the agent. It can also be blank to disable this socket.

• collectd.configDir - This is where the agent writes the managed collectd config, since collectd can only be configured by files. Note that this entire dir will be wiped by the agent upon startup so that it doesn't pick up stale collectd config, so be sure that it is not used for anything else. Also note that these files could have sensitive information in them if you have passwords configured for collectd monitors, so you might want to place this dir on a tmpfs mount to avoid credentials persisting on disk.

See the section on Privileges for information on the capabilities the agent requires.

1. Run the agent by invoking the archive path signalfx-agent/bin/signalfx-agent -config <path to config.yaml>. The agent logs only to stdout/err so it is up to you to direct that to a log file or other log management system if you wish to persist logs. See the signalfx-agent command doc for more information on supported command flags.

Privileges

When using the host observer, the agent requires the Linux capabilities DAC_READ_SEARCH and SYS_PTRACE, both of which are necessary to allow the agent to determine which processes are listening on network ports on the host. Otherwise, there is nothing built into the agent that requires privileges. When using a package to install the agent, the agent binary is given those capabilities in the package post-install script, but the agent is run as the signalfx-agent user. If you are not using the host observer, then you can strip those capabilities from the agent binary if so desired.

You should generally not run the agent as root unless you can't use capabilities for some reason.

Configuration

The agent is configured primarily from a YAML file (by default, /etc/signalfx/agent.yaml, but this can be overridden by the -config command line flag).

For the full schema of the config, see Config Schema.

For information on how to configure the agent from remote sources, such as other files on the filesystem or KV stores such as Etcd, see Remote Configuration.

Logging

Currently the agent only supports logging to stdout/stderr, which will generally be redirected by the init scripts we provide to either a file at /var/log/signalfx-agent.log or to the systemd journal on newer distros. The default log level is info, which will log anything noteworthy in the agent without spamming the logs too much. Most of the info level logs are on startup and upon service discovery changes. debug will create very verbose log output and should only be used when trying to resolve a problem with the agent.

Proxy Support

To use an HTTP(S) proxy, set the environment variable HTTP_PROXY and/or HTTPS_PROXY in the container configuration to proxy either protocol. The SignalFx ingest and API servers both use HTTPS. The agent will automatically manipulate the NO_PROXY envvar to not use the proxy for local services.

Diagnostics

The agent serves diagnostic information on a UNIX domain socket at the path configured by the diagnosticsSocketPath option. The socket takes no input, but simply dumps its current status back upon connection. As a convenience, the command signalfx-agent status will read this socket and dump out its contents.

The agent status output has the following sections:

• Version: The agent version and build time
• Agent Configuration: The current configuration in use by the agent, with secret values replaced by *s. Default values will be shown here if they were not set in the agent config file.
• Writer Status: The status and metrics about the writer component which writes datapoints to SignalFx.
• Observers: The active observers in the agent
• Monitor Configurations (Not necessarily active): A list of monitor configurations that are in place. If a configuration has a discovery rule but no discovered endpoints match that rule, there will not be any active instances of this monitor.
• Active Monitors: Monitors instances that are actively monitoring something. There may be multiple instances of these per configuration above if there is a discovery rule that matches multiple services.
• Discovered Endpoints: A list of the endpoints discovered by the agent's observers. The fields shown there will be the fields used when matching discovery rules to a discovered endpoint.
• Bad Monitor Configurations: This will be a set of monitor configurations that did not validate and the associated error. Bad monitor configuration generally does not prevent the agent from starting up, but will prevent that monitor from ever instantiating.

Also see our FAQ for more troubleshooting help.

Development

If you wish to contribute to the agent, see the Developer's Guide.