Closed
Bug 739994
Opened 13 years ago
Closed 12 years ago
Remove Leopard (10.5) as a build platform
Categories
(Infrastructure & Operations Graveyard :: CIDuty, task, P2)
Tracking
(Not tracked)
RESOLVED
FIXED
People
(Reporter: coop, Assigned: coop)
References
Details
(Whiteboard: [cleanup])
Attachments
(3 files, 1 obsolete file)
3.04 KB,
patch
|
bear
:
review+
coop
:
checked-in+
|
Details | Diff | Splinter Review |
42.27 KB,
patch
|
jhford
:
review+
coop
:
checked-in+
|
Details | Diff | Splinter Review |
949 bytes,
patch
|
jhford
:
review+
coop
:
checked-in+
|
Details | Diff | Splinter Review |
The only builds still running on Leopard buildslaves (10.5==darwin9) are:
* mozilla-1.9.2 on-change and nightlies (EOL on April 24)
* release builders for mozilla-1.9.2 (EOL on April 24, no further releases planned)
* spidermonkey builds on various branches (also run on 10.6 already)
* mozilla-release hg bundle (other slaves types can also do this already)
* mozilla-release leak test build
The last one (mozilla-release leak test build) is the only one that concerns me at all. Is that builder (or any of the others) important enough that we couldn't just remove 10.5 from our configs now (this bug) and then turn off all the darwin9 machines (bug 731458)?
Comment 1•13 years ago
|
||
* mozilla-release leak test build
is still using 10.5 builders but in the next merge we will have 10.6 machines.
Even then, we should not remove the builder until we officially stop supporting 10.5 debug builds (which we are not AFAIK).
Comment 2•13 years ago
|
||
Spidermonkey does *not* run on 10.6. We waste a ton of resources pretending to run it on 10.6, but since last July/August, we've built a 32bit NSPR and a 64bit JS shell and failed to run the results.
Assignee | ||
Updated•12 years ago
|
Assignee | ||
Comment 3•12 years ago
|
||
This is the first pass to remove spidermonkey. We can remove everything else (1.9.2) on April 24.
Attachment #615411 -
Flags: review?(bear)
Updated•12 years ago
|
Attachment #615411 -
Flags: review?(bear) → review+
Assignee | ||
Comment 4•12 years ago
|
||
Comment on attachment 615411 [details] [diff] [review]
Turn off spidermonkey builds on 10.5
https://hg.mozilla.org/build/buildbot-configs/rev/779810d377bc
Attachment #615411 -
Flags: checked-in+
Comment 5•12 years ago
|
||
deployed in production
Assignee | ||
Comment 6•12 years ago
|
||
Also turns off 10.5 builds for Thunderbird which were just added, but I've verified with Standard8 that they're not required.
Attachment #618722 -
Flags: review?(jhford)
Assignee | ||
Comment 7•12 years ago
|
||
Comment on attachment 618722 [details] [diff] [review]
Remove references to 10.5 platform and associated slaves in configs
Review of attachment 618722 [details] [diff] [review]:
-----------------------------------------------------------------
::: mozilla/config.py
@@ +1592,4 @@
> }
> for b in BRANCHES.keys():
> if b in lion_branches:
> + for p in ('macosx64', 'macosx64-debug',):
This is the part I'm most concerned about, John. Should I not be removing "OS X 10.7 32-bit" here? I don't *think* it's actually being used.
Comment 8•12 years ago
|
||
Comment on attachment 618722 [details] [diff] [review]
Remove references to 10.5 platform and associated slaves in configs
Review of attachment 618722 [details] [diff] [review]:
-----------------------------------------------------------------
This patch looks good, but it does turn off 32-bit debug builds for OS X on both 10.6 slaves and 10.7 slaves for all branches. Is that a desired outcome? If so, we should also remove the leopard debug builders on on mozilla-tests masters since nothing will be triggering them. Removing the debug builders on mozilla-tests should reduce the builder count on the masters substantially.
If we don't want to change the coverage of 32-bit debug builds and unit tests for OS X, I think we want to change the patch to only remove things related to the 'macosx' platform and leave 'macosx-debug' entries alone. Marking as r- because I don't think we want to turn 32-bit debug builds off. If we can confirm that stopping 32-bit debug builds on all branches is OK, this patch looks good to me.
::: mozilla/config.py
@@ -649,5 @@
> 'enable_unittests': False,
> 'enable_checktests': True,
> 'talos_masters': GLOBAL_VARS['talos_masters'],
> },
> - 'macosx-debug': {
Even though this is shown as OS X 10.5.2, these builds are done on 10.6 or 10.7 slaves exclusive of 1.9.2. I am not sure why the name wasn't changed to match the change in slave list as this is confusing. We should confirm whether or not we should turn off 10.6 or 10.7 based 32-bit debug builds.
@@ +1592,4 @@
> }
> for b in BRANCHES.keys():
> if b in lion_branches:
> + for p in ('macosx64', 'macosx64-debug',):
It is used to produce the 32-bit debug builds which we run tests on.
If we want to continue producing 32-bit debug builds, we do need to leave macosx-debug in here otherwise lion_branches will revert to using 10.6 slaves to produce their 32bit OS X debug builds.
Attachment #618722 -
Flags: review?(jhford) → review-
Assignee | ||
Comment 9•12 years ago
|
||
Similar to previous except I've only removed the macosx platform, leaving the macosx-debug platform that gets overridden by the lion builders. I've also left an empty entry for SLAVES['macosx'] because this is also used/overridden by the lion builders.
Attachment #618722 -
Attachment is obsolete: true
Attachment #619104 -
Flags: review?(jhford)
Updated•12 years ago
|
Attachment #619104 -
Flags: review?(jhford) → review+
Comment 10•12 years ago
|
||
Comment on attachment 619104 [details] [diff] [review]
Remove references to 10.5 platform and associated slaves in configs, v2
I backed out the patch:
http://hg.mozilla.org/build/buildbot-configs/rev/a19aeead40b9
There is a syntax error and it doesn't pass test-master even with syntax error fixed.
Attachment #619104 -
Flags: checked-in-
Assignee | ||
Comment 11•12 years ago
|
||
With this patch and the missing comma added to the production_config.py, the preproduction tests pass for me.
Attachment #619150 -
Flags: review?(jhford)
Assignee | ||
Comment 12•12 years ago
|
||
(In reply to Chris Cooper [:coop] from comment #11)
> With this patch and the missing comma added to the production_config.py, the
> preproduction tests pass for me.
I would note that this was all running fine otherwise in staging this morning, but I obviously wasn't using the production config in staging, and already had the 1.9.2 release branch removed locally.
Comment 13•12 years ago
|
||
Comment on attachment 619150 [details] [diff] [review]
Remove 1.9.2 release from production-masters.json
it happens
Attachment #619150 -
Flags: review?(jhford) → review+
Assignee | ||
Comment 14•12 years ago
|
||
Comment on attachment 619104 [details] [diff] [review]
Remove references to 10.5 platform and associated slaves in configs, v2
https://hg.mozilla.org/build/buildbot-configs/rev/24ff655719fa
Attachment #619104 -
Flags: checked-in- → checked-in+
Assignee | ||
Comment 15•12 years ago
|
||
Comment on attachment 619150 [details] [diff] [review]
Remove 1.9.2 release from production-masters.json
https://hg.mozilla.org/build/tools/rev/e1eab1525b3d
Attachment #619150 -
Flags: checked-in+
Assignee | ||
Updated•12 years ago
|
Status: ASSIGNED → RESOLVED
Closed: 12 years ago
Resolution: --- → FIXED
Updated•11 years ago
|
Product: mozilla.org → Release Engineering
Updated•6 years ago
|
Component: Platform Support → Buildduty
Product: Release Engineering → Infrastructure & Operations
Updated•5 years ago
|
Product: Infrastructure & Operations → Infrastructure & Operations Graveyard
You need to log in
before you can comment on or make changes to this bug.
Description
•