Amazon Kinesis Output for Telegraf
This is an experimental plugin that is still in the early stages of development. It will batch up all of the Points in one Put request to Kinesis. This should save the number of API requests by a considerable level.
This is not the place to document all of the various Kinesis terms however it maybe useful for users to review Amazons official documentation which is available here.
This plugin uses a credential chain for Authentication with the Kinesis API endpoint. In the following order the plugin will attempt to authenticate.
- Assumed credentials via STS if
role_arnattribute is specified (source credentials are evaluated from subsequent rules)
- Explicit credentials from
- Shared profile from
- Environment Variables
- Shared Credentials
- EC2 Instance Profile
For this output plugin to function correctly the following variables must be configured.
The region is the Amazon region that you wish to connect to. Examples include but are not limited to
The streamname is used by the plugin to ensure that data is sent to the correct Kinesis stream. It is important to note that the stream MUST be pre-configured for this plugin to function correctly. If the stream does not exist the plugin will result in telegraf exiting with an exit code of 1.
This is used to group data within a stream. Currently this plugin only supports a single partitionkey. Manually configuring different hosts, or groups of hosts with manually selected partitionkeys might be a workable solution to scale out.
When true a random UUID will be generated and used as the partitionkey when sending data to Kinesis. This allows data to evenly spread across multiple shards in the stream. Due to using a random paritionKey there can be no guarantee of ordering when consuming the data off the shards. If true then the partitionkey option will be ignored.
This is used to group data within a stream. Currently four methods are supported: random, static, tag or measurement
This will generate a UUIDv4 for each metric to spread them across shards. Any guarantee of ordering is lost with this method
This uses a static string as a partitionkey. All metrics will be mapped to the same shard which may limit throughput.
This will take the value of the specified tag from each metric as the paritionKey. If the tag is not found an empty string will be used.
This will use the measurement's name as the partitionKey.
The format configuration value has been designated to allow people to change the format of the Point as written to Kinesis. Right now there are two supported formats string and custom.
String is defined using the default Point.String() value and translated to byte for the Kinesis stream.
Custom is a string defined by a number of values in the FormatMetric() function.