968199 - AWS machines should run b2g emulator reftests with GALLIUM_DRIVER=softpipe

Status: RESOLVED FIXED

(Infrastructure & Operations Graveyard :: CIDuty, task)

Description

[Jeff Muizelaar [:jrmuizel]]

This causes a lot more of them to pass.

Comment 1

[Jeff Muizelaar [:jrmuizel]]

We don't want to make this change to http://mxr.mozilla.org/mozilla-central/source/build/mobile/b2gautomation.py#83 because that will affect all machines running the reftests. We only want this on the AWS machines.

Comment 2

[Armen [:armenzg]]

Let me see where we can make the change on mozharness.

We have this script and this config:
http://hg.mozilla.org/build/mozharness/file/default/scripts/b2g_emulator_unittest.py
http://hg.mozilla.org/build/mozharness/file/default/configs/b2g/emulator_automation_config.py

We have the reftest-options:
http://hg.mozilla.org/build/mozharness/file/default/configs/b2g/emulator_automation_config.py#l67

    67     "reftest_options": [
    68         "--adbpath=%(adbpath)s", "--b2gpath=%(b2gpath)s", "--emulator=%(emulator)s",
    69         "--emulator-res=800x1000", "--logcat-dir=%(logcat_dir)s"
    70         "--remote-webserver=%(remote_webserver)s", "--ignore-window-size",
    71         "--xre-path=%(xre_path)s", "--symbols-path=%(symbols_path)s", "--busybox=%(busybox)s",
    72         "--total-chunks=%(total_chunks)s", "--this-chunk=%(this_chunk)s",
    73         "%(test_manifest)s",
    74     ],

Is there a way that we could make runreftestb2g.py be able to take parameters to adjust the environment variables?

Comment 3

[Armen [:armenzg]]

<rail> btw, which driver is broken? maybe it's worth to fix it?
<jrmuizel> rail: llvmpipe
<rail> ok, I have I crazy idea :)
<jrmuizel> rail: but even if we fix it, we'll have to deploy a new driver to the whole ubuntu platform which will be an even more invasive change
<rail> we can set OPENGL_IS_BROKEN_HERE=1 on those machines, and use it in mozharness to set GALLIUM_DRIVER
<rail> so we don't even touch other pieces
* Tomcat|sheriffduty is now known as Tomcat|afk
<rail> if env.get("OPENGL_IS_BROKEN_HERE"): env["GALLIUM_DRIVER"] = "softpipe"
<armenzg> it works for me
* glob is now known as glob|away
<jrmuizel> rail: yeah, that seems reasonable
<rail> ship it! :)
<armenzg> the day that we fix the drivers situation we can unset that env
<rail> yeah

Comment 4

[Rail Aliiev [:rail]]

env for AWS Ubuntu machines can be set here: http://hg.mozilla.org/build/puppet/file/5f54cb661801/modules/gui/templates/xvfb.conf.erb#l7

Please add bug number and some comments to make sure we don't delete it a year later. :)

Comment 5

[Armen [:armenzg]]

I saw this fly by when syncing with puppet:
--- /etc/init/xvfb.conf	2014-02-05 13:19:48.279608001 -0800
+++ /tmp/puppet-file20140205-2419-670grd-0	2014-02-05 13:24:14.415607999 -0800
@@ -5,6 +5,7 @@
 description "start Xvfb server"
 
 export USER=cltbld
+export OPENGL_IS_BROKEN_HERE=1
 setuid cltbld

However, I can't seem to see the environment being set under the cltbld or root users.
I tried rebooting.
I tried pinning to my environment; I'm not sure if I did it properly.

Comment 6

[Armen [:armenzg]]

Attachment: add OPENGL_IS_BROKEN_HERE=1 to EC2 test machines

I still have to test the machine through buildbot on staging.

Comment 7

[Armen [:armenzg]]

I can't see the variable :(
http://dev-master01.build.scl1.mozilla.com:8042/builders/b2g_emulator_vm%20cedar%20opt%20test%20crashtest-1/builds/0/steps/run_script/logs/stdio

Comment 8

[Rail Aliiev [:rail]]

tst-linux64-ec2-001?

[[e-mail] ~]# cat /etc/X11/Xsession.d/98-broken-opengl
cat: /etc/X11/Xsession.d/98-broken-opengl: No such file or directory

Comment 9

[Armen [:armenzg]]

I was being lazy.
It is tst-linux64-ec2-armenzg disguised as tst-linux64-ec2-001 rather than having to add it to slavealloc + my master's configs.
The buildbot.tac is static on disk.

Comment 10

[Andrew Halberstadt [:ahal]]

Fwiw it would be relatively easy to add the env to the harness, but if you already have it solved in mozharness that wfm :). Having the fix be as close to the slave environment as possible makes sense too.

Comment 11

[Armen [:armenzg]]

Comment on attachment 8371857 [details] [diff] [review]
add OPENGL_IS_BROKEN_HERE=1 to EC2 test machines

Live in puppet.
https://hg.mozilla.org/build/puppet/rev/aa0fde115097

The EC2 machines will be picking this change up over the day.
We still have to work on the mozharness patch to set the GALLIUM_DRIVER env variable.

https://tbpl.mozilla.org/?tree=Cedar&jobname=b2g_emulator.*test

Comment 12

[Armen [:armenzg]]

Attachment: add GALLIUM_DRIVER=1 to b2g emulator jobs when running on EC2 test machines

Comment 13

[Armen [:armenzg]]

Jeff, can you please review this log and see if it does what you want? I see the GALLIUM_DRIVER variable.
https://tbpl.mozilla.org/php/getParsedLog.php?id=34295871&tree=Ash&full=1

Comment 14

[Aki Sasaki (not active)]

Comment on attachment 8372431 [details] [diff] [review]
add GALLIUM_DRIVER=1 to b2g emulator jobs when running on EC2 test machines

I think

Comment 15

[Armen [:armenzg]]

Comment on attachment 8372431 [details] [diff] [review]
add GALLIUM_DRIVER=1 to b2g emulator jobs when running on EC2 test machines

http://hg.mozilla.org/build/mozharness/rev/1ac44c9974e7

Comment 16

[Armen [:armenzg]]

I spoke with Jeff at the office.

Comment 17

[Aki Sasaki (not active)]

I'm wondering if this is causing the ICS emulator crashtests on Cypress to hang 7200 seconds with no output:
https://tbpl.mozilla.org/?tree=Cypress

Comment 18

[John Hopkins (:jhopkins)]

in production

Comment 19

[Armen [:armenzg]]

Backing out. It is making b2g emulator crashtests time out.

Comment 20

[Armen [:armenzg]]

I've requested jobs being re-triggered for this changeset:
https://tbpl.mozilla.org/?tree=Ash&jobname=b2g_emulator&rev=53bb6beeb1dc

They should be testing this:
http://hg.mozilla.org/users/asasaki_mozilla.com/ash-mozharness/rev/649b6f5c150e

+        if suite_name == 'reftest' and os.environ.get("OPENGL_IS_BROKEN_HERE"):
+            env["GALLIUM_DRIVER"] = "softpipe"

We should see results in less than 2 hours.

Comment 21

[Armen [:armenzg]]

Attachment: add GALLIUM_DRIVER=1 to b2g *reftests* emulator jobs when running on EC2 test machines

It is working now.

Comment 22

[Armen [:armenzg]]

Comment on attachment 8373491 [details] [diff] [review]
add GALLIUM_DRIVER=1 to b2g *reftests* emulator jobs when running on EC2 test machines

http://hg.mozilla.org/build/mozharness/rev/2ba270c1536c

Comment 23

[Armen [:armenzg]]

Pushed to the production branch:
https://hg.mozilla.org/build/mozharness/rev/7853577f5492

Comment 24

[Armen [:armenzg]]

Now that we have deployed a patched mesa we can back this out to take advantage of it:
<jrmuizel> armenzg: we should be able to backout bug 968199 now
<jrmuizel> armenzg: that will be needed for us to take advantage of the new mesa

Comment 25

[Armen [:armenzg]]

Done: https://hg.mozilla.org/build/mozharness/rev/ef33850170e

We should see the newer results in here:
https://tbpl.mozilla.org/?tree=Elm&jobname=b2g_emulator_vm%20elm

Comment 26

[Armen [:armenzg]]

Comment on attachment 8371857 [details] [diff] [review]
add OPENGL_IS_BROKEN_HERE=1 to EC2 test machines

Backed out as well since we don't need it:
https://hg.mozilla.org/build/puppet/rev/2122b8cb3302

Comment 27

[Dustin J. Mitchell [:dustin] (he/him)]

in production