influxdata
Feature: non-indexed tags
Proposal: The ability to declare some tags as non-indexed. This will reduce the blot up of the series count greatly (OOM or timeouts). This feature is need not to relay on the timestamp to sperate 2 different points in a measurement.
Current behavior: Currently the only way to store values non-indexed is to put them as fields and not as tags into the measurement. But this behavior can create some collisions on writes and influxdb will overwrite/update the point, because the timestamp and all tags are equal. This collisions will increase with a lower precision like for seconds or minutes.
It is currently only possible to store externally aggregated samples into influxdb, but very hard to write all points directly into the measurement and not to blot up the series count.
Desired behavior: A feature to declare a tag as non-indexed. The different value of a non-indexed tag will create a new point and not update/overwrite one.
Use case:
-
In the example of a simple simple delivery-shipment application it would already have a big advantage to store the order-Id as a non-indexed tag and not as an field-value Influx would always write all order-points into the series and they could be aggregated with a simple query.
-
A simple http server single or cluster can write all requests with low precision into a measurement The request-Id and/or correlation-Id can be a non-indexed tag. A simple query with requests/seconds or requests/minutes could be executed.
This would be very nice. We need a way to uniquely identify telemetry runs that are happening concurrently from a time perspective in the data. Every way we have to do this increases the cardinality of our indices.
