Victoriametrics: FAQ: impact of constant values

[jelmd]

Victoria Metrics DB supports the option--selfScrapeInterval to collect metrics from the DB itself. Right now it will collect 598 values for each interval (v1.48.0). So the first question I ask myself is: do I really need all of them? Certainly not and thus the next question: How can I reduce, what gets actually collected? And finally: there are lot of metrics, which probably change its value never ever, or very rarely, e.g. once a year, only (like most of the 108 flag* values). Should I care about them? How many space gets really consumed in memory and on disk lets say per year with a scrape interval of 5s for such constant values?

[valyala]

VictoriaMetrics provides very efficient compression for time series with constant values - they are usually compressed by 1000 times and more according to rough estimations. Let's calculate how much storage space is needed for storing a constant time series with 5s scrape interval. Such a time series will contain 3600*24*365/5=6.3 millions of samples. Each uncompressed sample occupies 16 bytes - 8 bytes for timestamp in milliseconds and 8 bytes for floating-point value. This means that uncompressed time series would occupy 6.3M*16b=100Mb of storage space. The time series will occupy less than 100Kb of storage space after the compression.

[valyala]

As for reducing the number of stored time series, this can be done with relabeling. See this article for details on how to remove unneeded time series via relabeling.

[jelmd]

Excellent. This helps a lot, especially when starting to get into touch/plan with vm{db,agent} (prometheus terms/configs).
Thanx a lot!