# Deployment Examples ## Deployment Options Overview This section provides a quick overview for a few common deployment options for Netdata. You can read about [Standalone Deployment](/docs/deployment-guides/standalone-deployment.md) and [Deployment with Centralization Points](/docs/deployment-guides/deployment-with-centralization-points.md) in the documentation inside this section. The sections below go into configuration examples about these deployment concepts. ## Deployment Configuration Details ### Stand-alone The stand-alone setup is configured out of the box with reasonable defaults, but please consult our [configuration documentation](/docs/netdata-agent/configuration/README.md) for details, including the overview of [common configuration changes](/docs/netdata-agent/configuration/common-configuration-changes.md). ### Parent – Child For setups involving Parent and Child Agents, they need to be configured for [streaming](/docs/observability-centralization-points/metrics-centralization-points/configuration.md), through the configuration file `stream.conf`. This will instruct the Child to stream data to the Parent and the Parent to accept streaming connections for one or more Child Agents. To secure this connection, both need a shared API key (to replace the string `API_KEY` in the examples below). Additionally, the Child can be configured with one or more addresses of Parent Agents (`PARENT_IP_ADDRESS`). An API key is a key created with `uuidgen` and is used for authentication and/or customization on the Parent side. For example, a Child can stream using the API key, and a Parent can be configured to accept connections from the Child, but it can also apply different options for Children by using multiple different API keys. The easiest setup uses just one API key for all Child Agents. #### Child config As mentioned above, we do not recommend to connect the Child to Cloud directly during your setup. This is done in order to reduce the footprint of the Netdata Agent on your production system, as some capabilities can be switched OFF for the Child and kept ON for the Parent. In this example, Machine Learning and Alerting are disabled for the Child, so that the Parent can take the load. We also use RAM instead of disk to store metrics with limited retention, covering temporary network issues. ##### netdata.conf On the child node, edit `netdata.conf` by using the [edit-config](/docs/netdata-agent/configuration/README.md#edit-a-configuration-file-using-edit-config) script and set the following parameters: ```yaml [db] # https://github.com/netdata/netdata/blob/master/src/database/README.md # none = no retention, ram = some retention in ram mode = ram # The retention in seconds. # This provides some tolerance to the time the child has to find a parent in # order to transfer the data. For IoT this can be lowered to 120. retention = 1200 # The granularity of metrics, in seconds. # You may increase this to lower CPU resources. update every = 1 [ml] # Disable Machine Learning enabled = no [health] # Disable Health Checks (Alerting) enabled = no [web] # Disable remote access to the local dashboard bind to = lo [plugins] # Uncomment the following line to disable all external plugins on extreme # IoT cases by default. # enable running new plugins = no ``` ##### stream.conf To edit `stream.conf`, use again the [edit-config](/docs/netdata-agent/configuration/README.md#edit-a-configuration-file-using-edit-config) script and set the following parameters: ```yaml [stream] # Stream metrics to another Netdata enabled = yes # The IP and PORT of the parent destination = PARENT_IP_ADDRESS:19999 # The shared API key, generated by uuidgen api key = API_KEY ``` #### Parent config For the Parent, besides setting up streaming, this example also provides configuration for multiple [tiers of metrics storage](/docs/netdata-agent/configuration/optimizing-metrics-database/change-metrics-storage.md), for 10 Children, with about 2k metrics each. This allows for: - 1s granularity at tier 0 for 1 week - 1m granularity at tier 1 for 1 month - 1h granularity at tier 2 for 1 year Requiring: - 25GB of disk - 3.5GB of RAM (2.5GB under pressure) ##### netdata.conf On the Parent, edit `netdata.conf` by using the [edit-config](/docs/netdata-agent/configuration/README.md#edit-a-configuration-file-using-edit-config) script and set the following parameters: ```yaml [db] mode = dbengine dbengine tier backfill = new storage tiers = 3 dbengine page cache size = 1.4GiB # storage tier 0 update every = 1 dbengine tier 0 retention space = 12GiB # storage tier 1 dbengine tier 1 update every iterations = 60 dbengine tier 1 retention space = 4GiB # storage tier 2 dbengine tier 2 update every iterations = 60 dbengine tier 2 retention space = 2GiB [ml] # Enabled by default # enabled = yes [health] # Enabled by default # enabled = yes [web] # Enabled by default # bind to = * ``` ##### stream.conf On the Parent node, edit `stream.conf` by using the [edit-config](/docs/netdata-agent/configuration/README.md#edit-a-configuration-file-using-edit-config) script and set the following parameters: ```yaml [API_KEY] # Accept metrics streaming from other Agents with the specified API key enabled = yes ``` ### Active–Active Parents In order to setup active–active streaming between Parent 1 and Parent 2, Parent 1 needs to be instructed to stream data to Parent 2 and Parent 2 to stream data to Parent 1. The Child Agents need to be configured with the addresses of both Parent Agents. An Agent will only connect to one Parent at a time, falling back to the next upon failure. These examples use the same API key between Parent Agents and for connections for Child Agents. On both Netdata Parent and all Child Agents, edit `stream.conf` by using the [edit-config](/docs/netdata-agent/configuration/README.md#edit-a-configuration-file-using-edit-config) script: #### stream.conf on Parent 1 ```yaml [stream] # Stream metrics to another Netdata enabled = yes # The IP and PORT of Parent 2 destination = PARENT_2_IP_ADDRESS:19999 # This is the API key for the outgoing connection to Parent 2 api key = API_KEY [API_KEY] # Accept metrics streams from Parent 2 and Child Agents enabled = yes ``` #### stream.conf on Parent 2 ```yaml [stream] # Stream metrics to another Netdata enabled = yes # The IP and PORT of Parent 1 destination = PARENT_1_IP_ADDRESS:19999 api key = API_KEY [API_KEY] # Accept metrics streams from Parent 1 and Child Agents enabled = yes ``` #### stream.conf on Child Agents ```yaml [stream] # Stream metrics to another Netdata enabled = yes # The IP and PORT of the parent destination = PARENT_1_IP_ADDRESS:19999 PARENT_2_IP_ADDRESS:19999 # The shared API key, generated by uuidgen api key = API_KEY ``` ## Further Reading We strongly recommend the following configuration changes for production deployments: 1. Understand Netdata's [security and privacy design](/docs/security-and-privacy-design/README.md) and [secure your nodes](/docs/netdata-agent/securing-netdata-agents.md) To safeguard your infrastructure and comply with your organization's security policies. 2. [Optimize the Netdata Agents system utilization and performance](/docs/netdata-agent/configuration/optimize-the-netdata-agents-performance.md) To save valuable system resources, especially when running on weak IoT devices. We also suggest that you: 1. [Use Netdata Cloud to access the dashboards](/docs/netdata-cloud/README.md) For increased security, user management and access to our latest features, tools and troubleshooting solutions. 2. [Change how long Netdata stores metrics](/docs/netdata-agent/configuration/optimizing-metrics-database/change-metrics-storage.md) To control Netdata's memory use, when you have a lot of ephemeral metrics. 3. [Use host labels](/docs/netdata-agent/configuration/organize-systems-metrics-and-alerts.md) To organize systems, metrics, and alerts.