- 1 Build Engineers
- 2 Signed-off Revision(s)
- 3 Tags
- 4 Build data
- 5 Notes
- 5.1 Build 1
- 5.1.1 Tag
- 5.1.2 Source
- 5.1.3 Build/Repack
- 5.1.4 Sign
- 5.1.5 L10nVerify
- 5.1.6 Generate updates and push betatest snippets
- 5.1.7 Update verify
- 5.1.8 Update Bouncer
- 5.1.9 Final checks before push
- 5.1.10 Push to mirrors
- 5.1.11 Final verification
- 5.1.12 Fix up releasetest
- 5.1.13 Publish updates to beta channel
- 5.1 Build 1
release tracking bug
Build1: rev 5b61f163f2fd
All l10n repositories listed in l10n-changesets were also relbranch'd to GECKO191b4_20090423_RELBRANCH, and tagged with FIREFOX_3_5b4_RELEASE and FIREFOX_3_5b4_BUILD1 for build1.
|Type||Build ID||Build machine|
- moved UPDATE_PACKAGING_R7 to 0f80ac6c75d8 (tip at the time) in build/tools repo to pick up bug 487677
- patch buildbot-configs/mozilla2/release_config.py, get review, land
- checkin latest copy of l10n-changesets from tracking bug
- reconfig production-master
- do sendchange
buildbot sendchange --username=nthomas --master=localhost:9010 --branch=releases/mozilla-1.9.1 -m "Firefox 3.5b4 build1" goforit
No problems. Landed the version bump to 3.5b5pre, need push -f to create a new head (make sure you are on the default branch!).
Slave for Windows build was interrupted by mistake, stopped linux & mac en-US builds and cleaned up all three slaves. Added a dummy_factory and used it for tag and source steps, reconfig'd master and did sendchange. No problems with mac or linux. Windows running.
Discovered localised complete mars missing, filed bug 489954 and put in a temporary fix.
Assorted more sendchanges to deal with
- moz2-win32-slave22 still having a 3.1b3 checkout and getting stuck on the relbranch for that (need newer hg or more careful cleanup or better source pull logic)
- moz2-win32-slave05 getting a busted mozilla-1.9.1 repo
- getting some connection resets on upload to stage
- not thinking as clearly as is useful
We have 70 complete mars for each platform now. It would be _really_ great if we had a way to trigger only some locale & platform combinations.
- Failed out when stub files couldn't be checked out.
- Had to re-install the Equifax root certificate (because of the recent cygwin update):
cp equifax.pem /usr/ssl/certs echo "ca_certificate = /usr/ssl/certs/equifax.pem" >> /etc/wgetrc
- Restarted with:
make sign PRODUCT=$PRODUCT VERSION=$VERSION BUILD=$BUILD TAG=$TAG REPO=$REPO EMAILemail@example.com
make stage verify fake-upload PRODUCT=$PRODUCT VERSION=$VERSION BUILD=$BUILD TAG=$TAG REPO=$REPO EMAILfirstname.lastname@example.org
- Removed the file en-US.xpi and entries from the two SUMS files; bug 485860 for the automation, bug 485935 for the build side
- verify-asc failed because it got 212 asc's vs 215 files - needs to ignore *SUMS and KEY, bug 492909)
Nice to have:
- Output from fake-upload could be much more useful, eg xpi's are shown
Plus the real upload of the builds and the logfile, as documented here.
- not all slaves have 7z to unpack windows installers (hit this with moz2-darwin9-slave06). Reran when only bm-xserve19 was available. Filed bug 485935
- no problems - looks like string updates to crashreporter.ini & updater.ini, hu fixing up searchplugins, some bookmark string changes. Plus "Only in" results for locales added/removed from shipped-locales
Generate updates and push betatest snippets
- No problems
- on moz2-linux-slave16 (to cleanup a big hunk of disk)
cd /builds/slave/updates/build/temp/firefox rm -rf 3.*/ftp
This leaves behind the snippets, source tree for mar & mbsdiff utilites, and so on, but removes all the mar files.
- Got race in pushing config back to hg repo, bug 486454. Cleaned up failed slaves on linux & windows and forced sequential builds
- All three failed because
- no ml for b4
- cy came back after missing b3, but patcher doesn't generate a snippet from b2 to b4. May need to manually correct this
- Hungarian (hu) and Mongolian (mn) changed their list of default searchplugins (bug 478878,bug 477989) and we get this on complete update. Can't remove these because they're in use by other locales, and removed-files is not locale specific
Only in source/firefox/searchplugins: amazondotcom.xml Only in source/firefox/searchplugins: answers.xml Only in source/firefox/searchplugins: eBay.xml Only in source/firefox/searchplugins: yahoo.xml
- On win32 the usual freebl3.chk and softokn3.chk mismatch, .autoreg and removed-files added on complete update
Final checks before push
Each of the find calls should return no output.
cd /home/ftp/pub/firefox/nightly/3.5b4-candidates/build1/ find . ! -user ffxbld find . ! -group firefox find . -type f ! -perm 644 find . -type d -mindepth 1 ! -perm 755 ! -name 'contrib*' find contrib contrib-localized ! -perm 2775
Push to mirrors
# ffxbld@stage rsync -av --exclude=*.log --exclude=*.txt --exclude=*unsigned* \ /home/ftp/pub/firefox/nightly/3.5b4-candidates/build1/ \ /home/ftp/pub/firefox/releases/3.5b4/
- Use 'Force Build to start the final_verification builder, chase down any 503 or 404 errors to find bad mirrors - lowered mozilla.ntua.gr from 329 to 150 in bouncer.
Fix up releasetest
Didn't get releasetest for 3.1a1/a2/b1. Copied beta snippets, removed b2 & b3, renamed beta to releasetest. Pushed live.
Fixed with change to patcher config.
Publish updates to beta channel
~/bin/backupsnip 20090424-Firefox-3.5b4 ~/bin/pushsnip 20090424-Firefox-3.5b4