metric: windowed histograms and polling discrepancy #98621
Labels
A-observability-inf
C-bug
Code not up to spec/doc, specs & docs deemed correct. Solution expected to change code/behavior.
Describe the problem
Our current windowed histogram implementation does not match its intended purpose. Ideally, we want it to capture the sample data in the polling interval of its only user which is tsdb. This should result in every datapoint in tsdb having its data collected for the interval
[T-t, T]
whereT
is the current time, andt
is the time of last poll.Currently, we have windows that are set to 60s by default and are configurable. This means that when tsdb polls the histogram for data it gets data in an interval that could be any duration from 0-60s. This would mean that every subsequent poll calculates the moving average over a period that was already polled and then after a tick, it resets that histogram, potentially losing values that came after the previous poll.
Reference discussion in #98266.
Proposed Solution
From #98266:
Jira issue: CRDB-25368
The text was updated successfully, but these errors were encountered: