You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Users report wanting to filter different metrics at different magnitudes (e.g. p99 vs p50). However, the --magnitude argument filters across all metrics.
One option could be to allow setting multiple metric=min_magnitude pairs with the --magnitude flag, or exclusively a single float. This would be pretty verbose... Ex:
Another option would be setting the min_magnitude in the config for each test's metric, with support for inheritance so each test doesn't need to define it. This option should also support overriding the config.
templates:
common_metrics:
metrics:
p50:
scale: 1.0e-6direction: -1min_magnitude: 0.05p99:
scale: 1.0e-6direction: -1min_magnitude: 0.2tests:
local2:
type: csvfile: tests/resources/sample.csvtime_column: timemetrics:
m1:
column: metric1direction: 1m2:
column: metric2direction: -1min_magnitude: 0.99# Be able to set min_magnitude for specific testsattributes: [ commit ]remote1:
inherit: [common_metrics] # min_magnitude set via inheritanceprefix: "performance_regressions.my_product.%{BRANCH}.test1"
I just want to state for the record, that one of the Nyrkiö 2024 patches contains a fix that improved the accuracy of Hunter for finding small changes and in my experience enabled going more in the direction where the user can rather use a really small p-value (0.0001) and then set the min threshold to zero or really small too. If other users have a similar experience (once they actually upgrade to what is now the head of the master branch, then I would not invest in making the min threshold more complicated.
Activity
henrikingo commentedon Feb 12, 2025
I just want to state for the record, that one of the Nyrkiö 2024 patches contains a fix that improved the accuracy of Hunter for finding small changes and in my experience enabled going more in the direction where the user can rather use a really small p-value (0.0001) and then set the min threshold to zero or really small too. If other users have a similar experience (once they actually upgrade to what is now the head of the master branch, then I would not invest in making the min threshold more complicated.