As PromQL has evolved, there are some functions that should no longer be used.

Note: As of Prometheus 2.0 count_scalar, drop_common_labels, and keep_common have been removed.

The first of these is count_scalar(), which returns the number of series in an instant vector. Unlike the count aggregator, this returns a 0 if the vector is empty. If you are trying to alert on missing time series, absent() is a better way to do it.

Next up is delta() which calculates the difference between the first and last samples in a time range for a gauge, with some extrapolation. As it only uses two samples it's highly susceptible to outliers, which is not desirable. If you want to know how fast a gauge is changing over time, deriv() is a better choice as it uses a least squares regression. If you want to compare a gauge to a previous time the offset modifier allows for that, though you'll still be susceptible to outliers.

Finally there is drop_common_labels(). A key part of working with labelled time series is knowing which labels apply to what you're currently doing. drop_common_labels() and its counterpart the keep_common aggregation modifier will produce different labels depending on what the input is, which is thus undesirable. If there's labels you don't want showing up, adjust your by/without clause accordingly.

 

Need help with PromQL? Contact us.