mimir
mimir copied to clipboard
Mixin: Add experimental Loki recording rules for query analysis.
Adds some experimental Loki recording rules (and some associated dashboards) to extract useful insights about costly queries from query-frontend log lines. The motivation to use Loki recording rules is as follows:
- If we added similar metrics on a per-
user
basis, the metrics would potentially be very high cardinality, especially if we wanted to add histograms (for computing p99, etc). - Querying the information from Loki over large time windows becomes prohibitively slow, particularly some interesting types of analysis, such as
(tenant wall time) / (total wall time)
ortopk()
over week.
I'm sure there is plenty to discuss here, including the necessity of the rules, but I would like to propose it anyway and perhaps we can merge it in some form, to get some feedback about how useful it is from users.
Screenshots: