1279193 - Make autoclassification queries use a sliding time window for better performance.

Status: RESOLVED FIXED

(Tree Management :: Treeherder, defect)

Description

[James Graham [:jgraham]]

Sometimes autoclassification is slow because we are sorting a lot of data. This can happen for a couple of reasons:

1) An intermittent that is short-lived but has a lot of instances in that time
2) An intermittent that is longer-lived with a relatively high rate, but not high enough that the corresponding test (if any) was disabled (this case is quite likely for intermittent crashes or other things that don't correspond well to a specific test)

For the second case, starting by only looking at recent data, and then working back in time, can help, as long as we stop looking once we find a good enough match, or hit some predefined time limit.

Comment 1

[GitHub Autolander Bot]

Attachment: [treeherder] mozilla:autoclassify_time_window > mozilla:master

Comment 2

[Treeherder GitHub Bugbot]

Commit pushed to master at https://github.com/mozilla/treeherder

https://github.com/mozilla/treeherder/commit/b7617832818eae7afc86cb5bf92824a1b4682c51
Bug 1279193 - Use a sliding time window, with a maximum execution time, for autoclassification. (#1580)

First try to select only failure lines from the last 7 days. If that
didn't find a good enough match, select from the 7 days before that, and
so on. If running the queries has taken more than a set time limit,
stop looking at further time ranges.