1372639 - Add stats for time spent performing redirects whilst fetching logs for parsing

Status: RESOLVED FIXED

(Tree Management :: Treeherder, enhancement, P1)

Description

[Ed Morley [:emorley]]

It is believed that one of the (several) causes of the spikes in log parsing times is due to queue.taskcluster.net being under high load and taking a while to return the HTTP 303 redirect to the log's actual location on public-artifacts.taskcluster.net.

It would be preferable to make a data-based decision about this, however at present, New Relic only shows the overall request time (ie the combined duration of the redirect + the final HTTP 200) and does not break it down further.

The Python requests library exposes the time for each individual request, which we can add as custom New Relic annotations/events:
http://docs.python-requests.org/en/latest/api/#requests.Response.elapsed

Comment 1

[GitHub Autolander Bot]

Attachment: [treeherder] mozilla:newrelic-redirect-stats > mozilla:master

Comment 2

[Treeherder GitHub Bugbot]

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

https://github.com/mozilla/treeherder/commit/58daa2c6ecf5289b23599beecda997be0703f518
Bug 1372639 - Track the time spent redirecting when making requests

By default New Relic only tracks the overall time spent making an
external HTTP request, when this could in fact cover multiple
sub-requests, if the first redirected.

This adds a New Relic custom event that tracks the time spent
redirecting, to help assess the priority of bug 1348072.