Releases/Firefox 3.1b3/BuildNotes
Build Engineers
nthomas
release tracking bug
Signed-off Revision(s)
Build1: revision df865af82400 Build2: revision 3d9704097cd8
Tags
On release/mozilla-1.9.1:
| Build # | Tag | Changeset |
| 1 | GECKO191b3_20090304_RELBRANCH | df865af82400 |
| FIREFOX_3_1b3_RELEASE, FIREFOX_3_1b3_BUILD1 | f2472bc0091b | |
| 2 | FIREFOX_3_1b3_RELEASE, FIREFOX_3_1b3_BUILD2 | 3d9704097cd8 |
All l10n repositories listed in l10n-changesets were also relbranch'd to GECKO191b3_20090304_RELBRANCH, and tagged with FIREFOX_3_1b3_RELEASE and FIREFOX_3_1b3_BUILD1 for build1. Tagging for build 2 adds a FIREFOX_3_1b3_BUILD2 tag, while FIREFOX_3_1b3_RELEASE is unchanged.
Build data
| Type | Build ID | Build machine |
| [Windows installer/zip] | ||
| [Mac compressed] | ||
| [Linux compressed] |
Notes
- Did some disk cleanup before starting, making sure these dirs were all empty across the farm (/builds/*slave*/): tag source linux_build linux_repack macosx_build macosx_repack win32_build win32_repack l10n_verification updates linux_update_verify macosx_update_verify win32_update_verify final_verification
- Make sure any changes to files like l10n-changesets, release_config.py, and release_master.py have been updated on the master, and it has been reconfig'd to pick them up
Build 1
Close Firefox3.1 tree, and kicked off automation with: (TBC)
buildbot sendchange --master=localhost:9010 --username=nthomas --branch=releases/mozilla-1.9.1 -m "Firefox 3.1b3 build1 try#1" gogogo
Tag
Automation successfully did all the locales, then tried to push changes to releases/mozilla-1.9.1 and failed because of the CLOSED TREE hook. (bug 482269)
pushing to ssh://hg.mozilla.org/releases/mozilla-1.9.1 searching for changes remote: adding changesets remote: adding manifests remote: adding file changes remote: added 3 changesets with 6 changes to 5 files remote: Tree mozilla-1.9.1 is CLOSED! (http://tinderbox.mozilla.org/Firefox3.1/) remote: To push despite the closed tree, include "CLOSED TREE" in your push comment remote: transaction abort! remote: rollback completed remote: abort: pretxnchangegroup.a_treeclosure hook failed abort: unexpected response: empty string
Which seems bogus because pushes don't have comments, checkins do. Our version bump does have "CLOSED TREE" in it, but the two tagging operations don't. Fixed by adding a new line before <span id="extended-status"> on the tinderbox page, which disables the hook, and rerunning
hg push -e 'ssh -l ffxbld -i ~cltbld/.ssh/ffxbld_dsa' -f ssh://hg.mozilla.org/releases/mozilla-1.9.1
on moz2-linux-slave08.
Then made tag_factory a dummy, and did another sendchange - got same slave and died trying to reclone af locale, no fallout. Then reconfig'd master to pickup dummy factory, and did sendchange.
bug 482649 non-release builders are also triggered by sendchange, reducing slave availability
Landed version bump for nightlies and reopened tree. You have to push -f because you're creating a new head (the relbranch doesn't create one but the first push after branching does).
Source
Automated, no problems
Build/Repack
- Linux: No problems
- Mac: Looks like the first build on each slave failed to update to the tag on m-1.9.1 example. Luckily we do this twice, and the second time uses the -C that the v0.9.5 mercurial needs to span the branch
- Win32: No problems
Sign
instructions
Status: dll/exe signing running
Respin was declared at this point. Pushed out builds with signed exe/dll (WITHOUT signed installers) so QA can test build1 while waiting for build2.
Build 2
Cleaned up build1 dirs for tag, source, linux_build/repack, macosx_build/repack, win32_build/repack.
Updated automation config, removed Dummy step for Tag and reconfig'd master.
Close Firefox3.1 tree, and kicked off automation with: (TBC)
buildbot sendchange --master=localhost:9010 --username=nthomas --branch=releases/mozilla-1.9.1 -m "Firefox 3.1b3 build2 try#1" gogogo
Tag
Automated, no problems this time.
Source
Automated, no problems. Verified
hg diff -r FIREFOX_3_1b3_BUILD1 -r FIREFOX_3_1b3_BUILD2
is sane with local repo.
Build/Repack
Same mac hg issue as build1. Also had
Rebuilt these by hand by grabbing a slave and reproducing the commands in the buildbot factories. Basically (but I did all the steps, semi-redundant though they are) this for win32/da:
cd /e/builds/moz2-slave/win32_repack/build/releases/l10n-mozilla-1.9.1 hg clone http://hg.mozilla.org/releases/l10n-mozilla-1.9.1/da/ hg -R da up -C -r FIREFOX_3_1b3_RELEASE cd /e/builds/moz2-slave/win32_repack/build/mozilla-1.9.1/browser/locales ZIP_IN=/e/builds/moz2-slave/win32_repack/build/mozilla-1.9.1/firefox.zip \ WIN32_INSTALLER_IN=/e/builds/moz2-slave/win32_repack/build/mozilla-1.9.1/firefox.exe \ MOZ_MAKE_COMPLETE_MAR=1 MOZ_PKG_PRETTYNAMES=1 \ make installers-da POST_UPLOAD_CMD="/home/ffxbld/bin/post_upload.py -p firefox -v 3.1b3 -n 2 --release-to-candidates-dir" \ UPLOAD_HOST=stage.mozilla.org UPLOAD_SSH_KEY="~/.ssh/ffxbld_dsa" \ UPLOAD_TO_TEMP=1 UPLOAD_USER=ffxbld MOZ_PKG_PRETTYNAMES=1 \ MOZ_MAKE_COMPLETE_MAR=1 \ make l10n-upload-da
mac/ko:
cd /builds/slave/macosx_repack/build/releases/l10n-mozilla-1.9.1 hg clone http://hg.mozilla.org/releases/l10n-mozilla-1.9.1/ko/ hg -R ko up -C -r FIREFOX_3_1b3_RELEASE cd /builds/slave/macosx_repack/build/mozilla-1.9.1/browser/locales ZIP_IN=/builds/slave/macosx_repack/build/mozilla-1.9.1/firefox.dmg \ MOZ_MAKE_COMPLETE_MAR=1 MOZ_PKG_PRETTYNAMES=1 \ make installers-ko POST_UPLOAD_CMD="/home/ffxbld/bin/post_upload.py -p firefox -v 3.1b3 -n 2 --release-to-candidates-dir" \ MOZ_MAKE_COMPLETE_MAR=1 MOZ_PKG_PRETTYNAMES=1 \ UPLOAD_HOST=stage.mozilla.org UPLOAD_SSH_KEY="~/.ssh/ffxbld_dsa" \ UPLOAD_TO_TEMP=1 UPLOAD_USER=ffxbld \ make l10n-upload-ko
NB: path is slave dependent
XXX We need a better way to spot these errors, the waterfall isn't very well suited to it. verify-locales.pl perhaps.
Sign
instructions No problems.
bug 482270 Need something that grooms out un-needed locales (eg ja on mac, ja-JP-mac on linux/win32) based on shipped-locales. Did this manually on stage this time.
bug 482644 Need to also add KEY, contrib & contrib-localized, generate SHA1SUM/MD5SUMS, fix up permissions etc. Some sort of final staging step, most of which can probably run on signing box.
L10nVerify
Green - lots of changes but they seem fine.
bug 482880 - win32 diffs have duplication
Generate updates and push betatest snippets
First attempt died out trying to execute
['/usr/bin/cvs', '-d', ':ext:cltbld@cvs.mozilla.org:/cvsroot', '-z3', 'checkout', '-d', 'build', '-r', 'post_signing', '-D', 'Sun, 08 Mar 2009 23:28:48 -0000', 'mozilla/tools/patcher'] cvs [checkout aborted]: no such tag post_signing
The 'post_signing' branch has overridden patcherToolsTag in ReleaseUpdatesFactory. bug 482272: Bug to swap this over to a ShellCommand like all the other cvs operations.
Triggered manually on waterfall, leaving branch blank (no need to clean up moz2-linux-slave09 as it updates/ contains only tools/, and we delete & repull that on every step). No further problems.
XXX We need to generate releasetest snippets to test bouncer/mirrors on release day
Update verify
Hit problem with missing CVS_RSH=rsh on Mac, fallout from bug 428124. Updated plist on slaves.
All three are red from missing Welsh locale (cy) for b3, otherwise fine.