Which simulators need to be part of Xcode 9 beta 4?

9.3 and 10.3 and 11

Until the puppet changes land for iOS xcode9 hermetic, this includes builders as well.  Unfortunately, Xcode 9 broke how hermetic works for iOS builders.

Note that the iOS 9 simulator bots are currently testing on 9.0 not 9.3 so you'll have to upgrade those (see https://crbug.com/662689). And the iOS 10 simulator bots are testing 10.0 not 10.3 (but there's no bug for this).

Safest bet would be to include 9.0, 9.3, 10.0, 10.3, and 11.0 simulators.

Also it wasn't mentioned but I imagine we should include touch-swarming as well. touch-swarming is the only place tests actually run.

Thank you smut@  That all sounds correct.  If it's a problem I imagine we could skip 9.3 and 10.3?

Yes, indeed we also need to upgrade all the touch-swarming bots to 10.12.5+, and those are the ones that do require all the simulators.

I don't have an opinion on {9,10}.0 vs. {9,10}.3, but I believe devs should make a call what they want to test and we should make those available (or have all of them to be safe).

Thanks, Sana, for catching this.

Is it easier to also do the same upgrade on the upstream bots?  I don't know which set of machines iOS depends on upstream, or how this will affect other bots, but these machines will also need to move to 10.12.5+

We'll want the ios machines in chromium.mac, chromium.fyi, tryserver.chromium.mac, and the upstream swarming machines.

dba@ how many machines a day do you think we can upgrade?

It depends greatly on load and if its touch-swarming vs buildbot and bling vs non-bling waterfalls.

I've still not had time to create the Xcode tarball and verify it.

Seeing as 10.12.6 is out, I created a 10.12.6 image as well as the Xcode 9.0 Beta 4 tarball (with the simulators listed in #5).

I've re-deployed build{41..45}-m7 on touch-swarming as 10.12.6 with 8.0 + 9.0-B4 for verification purposes.

It like the iOS10 testers on main and iOS11 testers on FYI are successfully using build{41..45}-m7.  

I've upgraded all of the -m7 bots to 10.12.6 + XC8 + XC9-B4 on touch-swarming. This just leaves the -b7 bots remaining there, which I'll start work on tomorrow for the ones without devices.

sergeyberezin@ - I placed the Xcode 9.0-Beta4 tarball I'm using in the bling-archive bucket at: gs://bling-archive/xcode/Xcode9.0-Beta4-10.12-20170728.tar.gz

Thanks, Bryce! I'll get the (tar)ball going tomorrow.

The following revision refers to this bug:
  https://chrome-internal.googlesource.com/chrome-golo/chrome-golo/+/1f93ed09a4355a743f18f1e01c39837fac16e61a

commit 1f93ed09a4355a743f18f1e01c39837fac16e61a
Author: Bryce Albritton <dba@google.com>
Date: Tue Aug 01 16:30:59 2017

Done with the non-device touch-swarming bots in b-lab, started on the ones with devices and have 31 left to go (doing this remotely by backing up /var/db/lockdown/*.plist.. build{25,27,29,30}-b7 were done this way already).

Will continue the work tomorrow.

touch-swarming bots are upgraded. I'll pick through the buildbot waterfalls for bling and official next week.

Thanks dba@!

Can you do the following first for the next round, please?

build58-m7 material-components-autoroller
build59-m7 piper-autoroller
build60-m7 simulator-builder
build61-m7 device-builder
build94-m7 chromium-autoroller
build140-m7 smoke 
build141-m7 smoke 
build142-m7 smoke 
build143-m7 smoke 

internal.bling.main was finished earlier this morning as well as the smoke bots. I'm going to finish chipping away at bling.tryserver now.

I think we now have enough bots upgraded that we can move main development over to Xcode 9, downstream only.  Thank you!

This bug only mentioned downstream bots, but the next step will be to do the same upstream for iOS compile and test.  What are your thoughts for how best to upgrade builders and testers?  Per your doc it appears to be around 100 builders and 250 testers.  Is there anything we can do to ease the upgrade, or things we should be aware of to not break anyone else?  If we fix hermetic, would it be easier to do the upgrade to 10.12.6 and not install Xcode 9 as well -- and get hermetic to do the Xcode deployment?   Or does it make sense to just install everything?

Also, would you like me to file a different bug for upstream, or just track here?

> This bug only mentioned downstream bots, but the next step will be to do the same upstream for iOS compile and test.  What are your thoughts for how best to upgrade builders and testers?  Per your doc it appears to be around 100 builders and 250 testers.  Is there anything we can do to ease the upgrade, or things we should be aware of to not break anyone else?  If we fix hermetic, would it be easier to do the upgrade to 10.12.6 and not install Xcode 9 as well -- and get hermetic to do the Xcode deployment?   Or does it make sense to just install everything?

Re-install time with including the Xcode 9 beta only adds maybe 3 minutes to each re-deploy, it makes no difference to me if we include Xcode.

My spreadsheet does not include the chromium-swarm bots that run the upstream tests (there are ~240 of those) I just mapped out the buildbot bots (we're mostly concerned with the tryserver.chromium.mac and chromium.mac iOS bots here and its not easy to just see what those were without some work). I can add tabs for those bots if needed, but its pretty easy to drill down via the swarming UI.

I can just deploy Xcode 9 beta on the chromium-swarm bots as they're 10.12.5 now, but the main reason why I'm targeting 10.12.6 is because I'd rather not get caught with Xcode 9 final requiring 10.12.6 if a hermetic solution is ready by that time. To actually upgrade the chromium-swarm bots to 10.12.6 I'd need to get buy-in from the GPU team that it's clear for us to upgrade them (which usually means they need to test it on their unstable bots then give the green light to redeploy the swarming fleet).

We can do the work here in this bug, unless someone really feels like we need to split the work/discussion out into a different bug.

Agreed on 10.12.6

I think in the short term it would be better to have compilation coverage upstream over  test coverage, so doing tryserver.chromium.mac and chromium.mac first makes sense?

How do we initiate the request to the GPU team to get buy in on upgrading?

> I think in the short term it would be better to have compilation coverage upstream over  test coverage, so doing tryserver.chromium.mac and chromium.mac first makes sense?

I'd opt for upgrading those first either way, as they're going to take the longest.

I'm still needing to upgrade cronet and official waterfalls as well.

> How do we initiate the request to the GPU team to get buy in on upgrading?

Filed issue 753510 requesting if we can upgrade the GPU fleet to 10.12.6.

bling.tryserver is now done. This completes the bling buildbot masters and touch-swarming.

I'll start on the cronet bots, then official, then the upstream buildbot bots (tryserver.chromium.mac/chromium.mac). I'm going to target chromium-swarm bots last (as issue 753510 is in progress and that'll be a large batch of bots)

Maybe upgrade the two machine on master.chromium.fyi, build252-m1, vm112-m1 early as well.  We should be able to flip that config quickly.

Thanks!!

official bots are done (minus osx-31-m0 which looks to have suffered an SSD failure, however this is only 1 of 10 bots that back the 'ios' target).

I've also finished the 2 chromium.fyi bots.

All tryserver.chromium.mac ios* bots should now be 10.12.6 + XC 8 + XC 9.0-B4. 

I'll move onto upgrading the chromium-swarm fleet next.

Is this fixed?  All the relevant upstream and downstream iOS bots have been flipped over to XC9.  Anything else missing?

Yeah, this should be done now seeing as most of the bare metal chromium-swarm fleet is now upgraded.

Thanks!