DEPRECATED : this was a proof-of-concept hacked together quite quickly, and had some pretty serious performance issues when dealing with a lot of metrics. Use https://github.com/Cleafy/promqueen for a much better solution.

A prototype tool for pre-initializing a Prometheus on-disk data store.

Useful for setting up standalone Prometheus instances that should be preloaded with data. Be careful loading very deep time-series without setting an appropriate retention time in the Prometheus instance when it starts.

$ for i in {1..1000} ; \
    do d=$(date +%s) ; \
    echo test_metric $i $(($(($d*1000 - 8640000 )) + $(($i * 15000)) )) ; \
    echo test_metric_b $i $(($(($d*1000 - 8640000 )) + $(($i * 15000)) ))  ; \
    done > test.prom
$ cat test.prom | ./prometheus-prefiller --log.level debug
$ echo "global: {}" > prometheus.yml
$ prometheus

You will now be able to query test_metric and test_metric_b over the last day.

Metrics are ingested in the prometheus text ingestion format, described here and visible from any Prometheus exporter endpoint.

Basic format:

metric_name value timestamp

With labels (probably what you want):

metric_name{label="value", label2="value2"} value timestamp

Unlike regular prometheus metrics, you should be appending a unix timestamp in milliseconds to backfill the data - Prometheus's memory storage engine will only accept monotonically increasing timestamps, so your data needs to be sorted

• this is not a backfilling solution. Identical timestamps are allowed but they must never go backwards.

The Prometheus engine writes a lot of logging by default - I've not supressed it since this is a quick and dirty tool to deal with quite a specific requirement for the time being.