Werk #16150: Forecasting: Stricter checks for missing data
| Component | Metrics system | ||||||||
| Title | Forecasting: Stricter checks for missing data | ||||||||
| Date | Sep 26, 2023 | ||||||||
| Level | Trivial Change | ||||||||
| Class | Bug Fix | ||||||||
| Compatibility | Compatible - no manual interaction needed | ||||||||
| Checkmk versions & editions |
|
When computing forecasts for forecast graphs or views, Checkmk now strictly validates the input RRD data and produces an error in case there are missing data points. The produced error message reads:
Historic metric data is incomplete. You can try to mitigate this by shortening the considered
history. Latest missing data point:
2023-09-26T00:00:00+00:00
Before this werk, missing data was handled inconsistently and partially incorrectly. Note that this change also solves crashes such as
Traceback (most recent call last):
...
y_hat = y_hat[display_time]
IndexError: boolean index did not match indexed array along dimension 0; dimension is 176 but corresponding boolean dimension is 177
These crashes were caused by handling missing data incorrectly. Instead, we now display the error message mentioned above.
Note that depending on the exact configuration, certain forecasts will now show the error message mentioned above, whereas before this change, a forecast was computed. We explicitly prefer this over showing potentially incorrect forecasts. Also note that we explicitly chose to not fill in missing data points eg. by interpolating, since this can lead to unreliable forecasts.