calling 'initctl set-env -g' from within an upstart job is lost

https://jenkins.qa.ubuntu.com/job/saucy-touch_ro-maguro-smoke-music-app-autopilot/101/artifact/clientlogs/dbus.log

could it be that the bus is just very busy with all these errors ?
:)

Perhaps, but I was unable to figure out who PID 1722 is to blame them :-)

Grabbing the environment from the crash file:

$1 = 0xbeed69b1 "PATH=/home/phablet/bin:/usr/lib/lightdm/lightdm:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
$2 = 0xbeed6a1e "TERM=linux"
$3 = 0xbeed6a29 "USER=phablet"
$4 = 0xbeed6a36 "LANGUAGE=en_US:en"
$5 = 0xbeed6a48 "GRID_UNIT_PX=18"
$6 = 0xbeed6a58 "XDG_SEAT=seat0"
$7 = 0xbeed6a67 "EXTERNAL_STORAGE=/mnt/sdcard"
$8 = 0xbeed6a84 "HOSTNAME=android"
$9 = 0xbeed6a95 "LOOP_MOUNTPOINT=/mnt/obb"
$10 = 0xbeed6aae "SSH_AGENT_PID=1221"
$11 = 0xbeed6ac1 "SHLVL=1"
$12 = 0xbeed6ac9 "HOME=/home/phablet"
$13 = 0xbeed6adc "DESKTOP_SESSION=ubuntu-touch-surfaceflinger"
$14 = 0xbeed6b08 "XDG_SEAT_PATH=/org/freedesktop/DisplayManager/Seat0"
$15 = 0xbeed6b3c "ANDROID_ASSETS=/system/app"
$16 = 0xbeed6b57 "QML2_IMPORT_PATH=/usr/lib/arm-linux-gnueabihf/qt5/imports"
$17 = 0xbeed6b91 "DBUS_SESSION_BUS_ADDRESS=unix:abstract=/tmp/dbus-tsYcRUEiJo"
$18 = 0xbeed6bcd "MANDATORY_PATH=/usr/share/gconf/ubuntu-touch-surfaceflinger.mandatory.path"
$19 = 0xbeed6c18 "LOGNAME=phablet"
$20 = 0xbeed6c28 "QT_SELECT=qt5"
$21 = 0xbeed6c36 "DEFAULTS_PATH=/usr/share/gconf/ubuntu-touch-surfaceflinger.default.path"
$22 = 0xbeed6c7e "QTWEBKIT_DPR=2.0"
$23 = 0xbeed6c8f "XDG_SESSION_ID=c1"
$24 = 0xbeed6ca1 "MKSH=/system/bin/sh"
$25 = 0xbeed6cb5 "ANDROID_DATA=/data"
$26 = 0xbeed6cc8 "GDM_LANG=en_US"
$27 = 0xbeed6cd7 "XDG_SESSION_PATH=/org/freedesktop/DisplayManager/Session0"
$28 = 0xbeed6d11 "XDG_RUNTIME_DIR=/run/user/32011"
$29 = 0xbeed6d31 "ANDROID_ROOT=/system"
$30 = 0xbeed6d46 "XDG_CURRENT_DESKTOP=Unity"
$31 = 0xbeed6d60 "LANG=en_US.UTF-8"
$32 = 0xbeed6d71 "SSH_AUTH_SOCK=/tmp/ssh-gL0Leo99g5Dl/agent.1246"
$33 = 0xbeed6da0 "SHELL=/system/bin/sh"
$34 = 0xbeed6db5 "GDMSESSION=ubuntu-touch-surfaceflinger"
$35 = 0xbeed6ddc "UPSTART_SESSION=unix:abstract=/com/ubuntu/upstart-session/32011/1229"
$36 = 0xbeed6e21 "XDG_VTNR=0"
$37 = 0xbeed6e2c "ASEC_MOUNTPOINT=/mnt/asec"
$38 = 0xbeed6e46 "QT_IM_MODULE=maliitphablet"
$39 = 0xbeed6e61 "ANDROID_PROPERTY_WORKSPACE=8,49152"
$40 = 0xbeed6e84 "PWD=/home/phablet"
$41 = 0xbeed6e96 "QT_QPA_PLATFORM=ubuntu"
$42 = 0xbeed6ead "XDG_CONFIG_DIRS=/etc/xdg/xdg-ubuntu-touch-surfaceflinger:/etc/xdg"
$43 = 0xbeed6eef "XDG_DATA_DIRS=/usr/share/ubuntu-touch-surfaceflinger:/usr/local/share/:/usr/share/"
$44 = 0xbeed6f42 "ANDROID_CACHE=/cache"
$45 = 0xbeed6f57 "ANDROID_BOOTLOGO=1"
$46 = 0xbeed6f6a "JOB=dbus"
$47 = 0xbeed6f73 "INSTANCE="
$48 = 0xbeed6f7d "UPSTART_EVENTS=started"
$49 = 0xbeed6f94 "UPSTART_JOB=url-dispatcher"
$50 = 0xbeed6faf "UPSTART_INSTANCE="
$51 = 0x0

Okay, discussion on IRC seems to believe that this is related to dbus' Upstart job believing that DBus is listening before it actually is (my wife says I have this problem as well). To fix that we need to change the session job to expect a fork, and have the daemon fork, as it will listen before forking.

<jodh> tedg: once you can recreate the problem, I wonder if it is worth tweaking the session dbus.conf job to "expect fork" and "--fork" like the system dbus job already does. That may give a "more correct" started event time.
<tedg> jodh, Why would that be more correct?
<xnox> tedg: dbus forks after it starts to listen, thus expect fork with --fork will make sure started is emitted only after dbus is ready.

This bug was fixed in the package dbus - 1.6.12-0ubuntu6

---------------
dbus (1.6.12-0ubuntu6) saucy; urgency=low

  * Specify --fork to dbus-daemon in upstart user-session mode, to get the
    daemon readiness information and emit started dbus, when dbus is
    actually ready to operate. (LP: #1234731)
 -- Dmitrijs Ledkovs <email address hidden> Thu, 03 Oct 2013 17:32:15 +0100

The actual problem here is not caused by dbus startup issues. /usr/share/upstart/sessions/dbus.conf calls 'initctl set-env -g DBUS_SESSION_BUS_ADDRESS' on startup, but even after login if I call 'initctl list-env -g', this variable is missing. If I call 'initctl list-env', it's present - possibly inherited from the dbus job's environment rather from the global environment, since gnome-session is 'start on started dbus [...]'.

Creating a test job, ~/.config/upstart/env-test.conf, that does the following:

pre-start script
 initctl set-env --global FOO=bar
end script

And starting this job with 'start env-test', the variable then shows up correctly in 'initctl list-env -g'.

I don't yet know why the variable exported by dbus is missing, but I can confirm that this is a common problem across all the variables being set in /usr/share/upstart/sessions/*.conf; GPG_AGENT_INFO, DBUS_SESSION_BUS_ADDRESS, SSH_AUTH_*, UBUNTU_MENUPROXY are all missing from the global env. GTK_MODULES is there, but presumably arrives by other means.

Although looking at my existing login session on my laptop let me reproduce this problem right off the bat, I've had difficulty reproducing it again with subsequent login sessions. The problem, once reproduced, is persistent within upstart, but it's not 100% reproducible between sessions. No idea why yet. I've done 10+ guest login sessions, 5+ logins with a second account on my system, and a restart of my regular user's session, and have yet to reproduce the bug again after adding debugging info to the upstart package here.

See bug 1238078.