Remove artificial columns 'min' and 'max' from the per-worker data#6469
Remove artificial columns 'min' and 'max' from the per-worker data#6469Karakatiza666 wants to merge 1 commit into
Conversation
…ker data Signed-off-by: Karakatiza666 <bulakh.96@gmail.com>
mythical-fred
left a comment
There was a problem hiding this comment.
APPROVE.
Clean fix. Removing the synthetic Min/Max columns at the source is the right call — profiler-layout already computes overview aggregates in its own metrics views (the per-column MetricsDistributionBlock has its own Avg/Min/Max headers), so this was duplicated state at the wrong layer.
The downstream HierarchicalTable.createDisplay just iterates columnNames to build header cells; nothing special-cases the last two slots, so the consumer side stays clean. The MeasurementMatrix invariant (columnNames.length == measurements.length) keeps holding.
Import cleanup correctly drops the now-unused MissingValue symbol. Description ties the change to the upstream cleanup of aggregation moving into profiler-layout, which matches what the code shows.
Fix #6467
The issue was the profiler-lib was unconditionally augmenting the per-worker metrics with a min and max value for the statistic. This was originally done as a way to provide an overview of the metrics - and aggregation for the purposes of overview is now coordinated in the consumer of profiler-lib (profiler-layout).
The artificial
minandmaxcolumn are not an intrinsic part of reporting the per-worker metrics, so I decided to remove computing these columns in the first place, rather than e.g. just filtering them out or making them optional.Testing: manual