So according to the datastore timestamp, the anomaly object was generated early May 30, and the actual regression occurred on May 27. Is that the difference you're seeing?

There's an anomaly config associated with Chrome OS Graphics that sets the min_segment_size to 12, which as I understand it means that it needs to collect 12 points before/after a change point before it considers it.

OK, I see. 

It seems like the reason that we get the graphics_Idle alerts but graphics_Sanity is that somehow graphics_Idle generates more points than the other tests.

+sullivan for how to test settings

From offline discussion, Po-Hsien would like to try out different settings but I'm a little unfamiliar with the process of trying those out.

Yes, many tests we run at most once a day and waiting 2+ weeks for an alert is a long time.

You can experiment with changing the alert config settings in the debug page. Screenshot of the link to get there attached.

You'll likely want to tweak "min_segment_size". The tradeoff is that with a lower segment size, you get alerts faster, but if there are short jumps due to noise, you may be alerted on those.

Once you have a config you like, use the "Report Issue" menu on dashboard to request a config change.

Deleted: debug-alerts.png 53.0 KB

	debug-alerts.png 53.0 KB View Download

Very cool page!

Is it possible to set different min_segment_size for different paths/wild cards like ChromeOS_Graphics/*/Failures or is it per master only (same value for all of ChromeOS_Graphics)?

You can set different min_segment_size for different paths/wildcards as long as they don't overlap (if they do, the first one alphabetically will be chosen, which is a bit confusing)

Thanks everyone!