1569916 - schedule android-hw reftest/crashtest for webrender on integration branches when gfx/* files change

Status: RESOLVED FIXED

(Testing :: Reftest, task)

Description

[Joel Maher ( :jmaher ) (UTC -8)]

currently android-hw crashtest/reftest runs on mozilla-central only, we need this to be tier-1 and running on integration (autoland/inbound). We will do this for opt only (not debug), and only when gfx/* files are changed (which should be 6% of the time)

Comment 1

[Joel Maher ( :jmaher ) (UTC -8)]

Attachment: Bug 1569916 - run reftest/crashtest on android-hw when gfx/* files change. r=ahal

run reftest/crashtest on android-hw when gfx/* files change.

Comment 2

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/b4d1af8b3345
run reftest/crashtest on android-hw when gfx/* files change. r=ahal

Comment 3

[Natalia Csoregi [:nataliaCs]]

https://hg.mozilla.org/mozilla-central/rev/b4d1af8b3345

Comment 4

[Joel Maher ( :jmaher ) (UTC -8)]

as a note, this is doing the right thing:

• integration: opt only when gfx/* files change
• central: runs both opt/debug on all pushes (what was there before)

in terms of load, the pool of unittest pixel2 phones went from a peak backlog of 100-150 in the past 2 week, to a backlog of 300. Prior to this change we had a 6-8 hour cycle during peak to get the queue back to 0. What I saw yesterday was a 20 hour window to clear the backlog- still getting results, but not ideal. Once we have another week of data on this, it will be interesting to see our backlogs and compare those to the perf tests. That would help make a decision if we need more phones or need to get creative in other ways.

Comment 5

[Joel Maher ( :jmaher ) (UTC -8)]

revisiting this, it appears we have trouble keeping the queue for the unittest devices to zero, I believe we need to increase our pool by 2 or 3 devices. Looking at the perf pool, the queue rarely has a >12 hour >0 size, can we pull some devices from perf, or can we merge the pools?

Comment 6

[Andrew Erickson [:aerickson]]

Let's move devices for now. Merging pools seems like a larger discussion.

https://github.com/bclary/mozilla-bitbar-devicepool/pull/33

Comment 7

[Andrew Erickson [:aerickson]]

Change landed, devicepool host updated, and service restarted. The 3 additional p2 devices are working in the unit queue.