Metrictank implements tag ingestion, storage, and querying to be compatible with the graphite tags feature.
Metrictank has a feature called "Meta Tags" which allows a user to dynamically assign virtual tags to metrics based on given criteria.
To use it the following two feature flags need to be turned on: memory-idx.meta-tag-support, memory-idx.tag-support (default)
The necessary API calls to setup meta records are documented here.
When using Meta Tags in a clustered setup the replication of Meta Tag Rules (Meta Records) happens via the backend store (currently only implemented for Cassandra). When rules are modified on a Metrictank that has index updating of the backend store enabled the modifications get persisted in the store. At a given interval (default 10s) all other Metrictanks which are using this index and which have the Meta Tags feature enabled poll the store for changes, and if they detect a change they load it.
This means that for changes to get replicated across a cluster, they need to be made on a Metrictank which has index updates enabled (for cassandra cassandra-idx.update-cassandra-index=true).
metrics2.0
Note that tags are modeled as an array of strings, where you can have key:value tags but also simply value tags.
We want to go further with our tag support and adopt metrics2.0 fully, which brings naming conventions and leveraging semantics.
Here are some goals:
- automatically setting the right unit and axis labels in grafana
- automatically converting available data to the requested unit for displaying
- automatically merging series (if you send a series first as a time in ms and then s, we can intelligently merge)
- automatically setting consolidation parameters based on the mtype tag