Access metering data to run reports


As discussed in Metering of resource usage, metering data is collected into a Thanos instance. Thanos provides a query API which is guaranteed to be compatible with the Prometheus 2.x API. The collected data is intended to be used to generate invoices. In addition, the data may also be used to show cost reports on the APPUiO Portal.

It’s important to note that the Prometheus data is missing information, for example prices and discounts. Therefore, the collected metering data has to be augmented with information provided by another source.


  1. Directly use the Prometheus API

    The Prometheus API allows to get results for a single point in time. It also allows to query all results within a given time range with a data point as defined by a given interval.

    Data collect by Prometheus don’t contain information about prices and discounts. Adding them though an exporter is not a suitable solution. Sometimes we need to do changes/fixes to prices and discounts in retrospect. This is not possible on Prometheus time series. Consequently, they would have to be fetched from a different location and merged into the data at runtime. The missing data could be fetched from the ERP system or any other location that seems suitable for the task.

  2. Use of an intermediate storage

    The Prometheus data is queried regularly, aggregated, enriched with product data and stored into an intermediate database. That intermediate storage could then also contain data about pricing and discounts.


Use of an intermediate storage


Prometheus queries are restricted in how many data points can be processed. The queries required for the billing model are quite complex. It’s not possible to run them once for a whole billing interval (currently one month). Experiments have shown, that the query exceeds the maximum execution time. In addition, even with a fairly small data set, memory usage of the Thanos Querier went up as high as 20GiB.

Even if the query was run for smaller time ranges, the execution time was in the order of several seconds. While this is not an issue for invoicing, it most certainly is for reports accessible to a user.

Simplifying those queries might be possible by using recording rules. Retention time for invoicing outlines the need to change queries after the fact. This prohibits the use of recording rules.