Turns out the new sampling algorithm introduced in https://trac.webkit.org/changeset/212853 can hide vast majority of points when zoomed out to a large time range because then almost all points are fall below the threshold.
Created attachment 302634 [details] Fixes the bug
Comment on attachment 302634 [details] Fixes the bug Nope, this still has a pathology where if the graph becomes more dense because bots become healthier, cycle time improved, etc... then we can simply drop an entire section of data points :(
Created attachment 302645 [details] Fixes the bug
Comment on attachment 302645 [details] Fixes the bug r=me
Committed r212946: <http://trac.webkit.org/changeset/212946>