471006 - Setup new mac OSX systems

Status: RESOLVED DUPLICATE of bug 455089

(Release Engineering :: General, defect)

Description

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

Our intel-based mac build machines create 32bit universal builds (intel32 + ppc). We then run unittest and talos on more 32bit intel-based mac machines. 


During bug triage, we've found bug#465835 on testing builds with PPC, bug#455089c#4 on testing with OSX10.4. Based on recent discussions in dev.planning, it seems like there will be 10.6, 64bit OSX requests coming soon. I've filed this bug to figure out what variants of Mac O.S. people want infrastructure for, on what hardware, and what sequence we should tackle these in. 

To set context, today I believe we're claiming support for all these O.S./CPU combinations on FF2.x, FF3.0.x, FF3.1 (this is afaict, please adjust if I'm mistaken):
10.4 on intel32
10.4 on ppc
10.5 on intel32
10.5 on ppc
...and additionally, proposed for FF3.2:
10.6 on intel32
10.6 on intel64


1) Is anything missing from this list?

2) We believe most users are on 32bit intel-based 10.5 macs, so have focused infrastructure attention there. If this assumption is incorrect, please let us know asap.

3) We have always had very few PPC machines. Most builds are created as universal builds on 32bit intel machines. All unittest/perf are run on 32bit intel machines. A summary:
FF2.0: 2 build, 0 unittest, 0 talos machines
FF3.0: 0 build, 0 unittest, 0 talos machines
FF3.1: 0 build, 0 unittest, 0 talos machines
FF3.2/m-c/tracemonkey: 0 build, 0 unittest, 0 talos machines

3a) The two PPC machines I know of are in use by FF2+TB2. bm-xserve01 was recently freed up from other work, and is now a standby replacement in case bm-xserve02 dies and closes the FF2+TB2 trees. Current plan is for these to remain there until EOL of TB2. If there are other PPC machines available, please let me know.

3b) Apple desupported PPC xserves in August 2006. Buying and supporting this hardware will be "interesting". For Talos, we typically have 3 identical machines per set to deal with test variance. Its unlikely we could get ever get that here, so Talos test variance is a concern, although rebootability might help.

3c) How many existing users do we have on PPC? Do we have breakdowns of 10.4PPC vs 10.5PPC? 


4) Which of these variants needs builds, or can we continue with universal builds produced on 10.5-on-intel32? For 10.6-on-intel64, will we need specific 64bit builds or can we use the same 10.5-on-intel32 builds?

5) Which of these variants need unittests?

6) Which of these variants need talos?

Comment 2

[Samuel Sidler (old account; do not CC)]

John, did this ever make it to dev-planning? It should probably get there so others can provide input.

Comment 3

[Samuel Sidler (old account; do not CC)]

Just another ping to make sure this gets to dev-planning so people can discuss. I'm not sure when you're planning to move forward with adding new machines, but it might take time to reach conclusions on which priority platforms/OSes are.

Comment 4

[Nick Thomas [:nthomas] (UTC+12)]

Bug 475199 is PPC only and would have been caught by talos (as a crash on startup).

Comment 5

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

Linking in bugs for recycling old Firefox2 PPC machines as 10.5 PPC unittest machines.

Comment 6

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

Taking during triage; not sure what still needs to be done here.

Comment 7

[John O'Duinn [:joduinn] (please use "needinfo?" flag)]

afaict, all the work is being done in other bugs, namely bug#455089, bug#463262, bug#521888.

Not sure how best to close this, so closing as a dup of oldest bug that has real work in it.