init: Update boot-complete to run at login-prompt-visible.

init: Update boot-complete to run at login-prompt-visible.

This saves us 2-3 seconds at boot.  It appears that the jobs
that were started when boot-complete finishes were greatly
slowing down Chrome while it was painting the login windows
after mapping them.  The window manager now notifies the
session manager after the windows are painted, at which
point the login-prompt-visible Upstart signal is emitted.

Here are some timings showing the seconds since power-on at
which the various signals are fired without and with this
change.

before:
login-prompt-ready     6.74   6.77   6.49
login-prompt-visible  10.69  10.61  10.21

after:
login-prompt-ready     6.37   6.78   6.65
login-prompt-visible   7.54   7.94   7.61

This also makes it write a boot-complete event (I have a
separate change to make the session manager log
login-prompt-ready and login-prompt-visible) and fixes a
typo in the logic that was supposed to prevent the job from
running a second time if the user logged out.

Change-Id: Ib11cac2ccba2a54180e1d9d8849352e1cf7b99af

BUG=chromium-os:10349
TEST=tried it.  checked that boot-complete runs and that /tmp/uptime-* files are written correctly

Committed: http://chrome-svn/viewvc/chromeos?view=rev&revision=ec11517

[Daniel Erat, 2010-12-17 03:28:17]

I'll send out changes to update the factory scripts that patch this file before
checking in the change.

[Chris Masone, 2010-12-17 04:48:24]

I wonder what impact this will have on login time for users who put in their
password right away.  Also, is wifi still pretty much up by the time the login
UI is on screen?

[Daniel Erat, 2010-12-17 19:26:56]

I'm seeing wifi consistently come up before the UI is onscreen.  I
haven't noticed any difference in login time from this change, but I
plan to examine that after this is in (if it were slower, the fix
would be to push the stuff that's delaying login back still further,
not leave it happening before the login UI is up).

On Thu, Dec 16, 2010 at 8:48 PM,  <cmasone@chromium.org> wrote:
> I wonder what impact this will have on login time for users who put in their
> password right away.  Also, is wifi still pretty much up by the time the
> login
> UI is on screen?
>
> http://codereview.chromium.org/5868004/
>

[Chris Masone, 2010-12-17 19:38:20]

On Fri, Dec 17, 2010 at 11:26 AM, Daniel Erat <derat@chromium.org> wrote:

> I'm seeing wifi consistently come up before the UI is onscreen.  I
> haven't noticed any difference in login time from this change, but I
> plan to examine that after this is in (if it were slower, the fix
> would be to push the stuff that's delaying login back still further,
> not leave it happening before the login UI is up).
>

Well, unless it impacts our ability to perform login -- like, we were
delaying some TPM setup that then made login->browser take longer.  But, it
sounds like this change is nothing but good :-)  I was more concerned about
the wifi bringup, and it sounds like that's no problem.

Thanks for looking at this, Dan.  LGTM


>
> On Thu, Dec 16, 2010 at 8:48 PM,  <cmasone@chromium.org> wrote:
> > I wonder what impact this will have on login time for users who put in
> their
> > password right away.  Also, is wifi still pretty much up by the time the
> > login
> > UI is on screen?
> >
> > http://codereview.chromium.org/5868004/
> >
>

[jrbarnette, 2010-12-17 20:11:14]

On 2010/12/17 03:28:17, Daniel Erat wrote:
> I'll send out changes to update the factory scripts that patch this file
before
> checking in the change.

You also need to update the platform_BootPerfTest autotest; it depends on
the name login-prompt-ready to determine boot times.

[Daniel Erat, 2010-12-17 20:15:15]

Yes, there are probably a bunch of things that are still using
login-prompt-ready that will need to be updated.  Since the signal is
still getting emitted with this change, they should continue being
just as wrong as they already are, right?

On Fri, Dec 17, 2010 at 12:11 PM,  <jrbarnette@chromium.org> wrote:
> On 2010/12/17 03:28:17, Daniel Erat wrote:
>>
>> I'll send out changes to update the factory scripts that patch this file
>
> before
>>
>> checking in the change.
>
> You also need to update the platform_BootPerfTest autotest; it depends on
> the name login-prompt-ready to determine boot times.
>
>
> http://codereview.chromium.org/5868004/
>

[jrbarnette, 2010-12-17 20:35:51]

On 2010/12/17 20:15:15, Daniel Erat wrote:
> Yes, there are probably a bunch of things that are still using
> login-prompt-ready that will need to be updated.  Since the signal is
> still getting emitted with this change, they should continue being
> just as wrong as they already are, right?
> 
Ah, yes.  I hadn't internalized the changes elsewhere.

LGTM!


> On Fri, Dec 17, 2010 at 12:11 PM,  <mailto:jrbarnette@chromium.org> wrote:
> > On 2010/12/17 03:28:17, Daniel Erat wrote:
> >>
> >> I'll send out changes to update the factory scripts that patch this file
> >
> > before
> >>
> >> checking in the change.
> >
> > You also need to update the platform_BootPerfTest autotest; it depends on
> > the name login-prompt-ready to determine boot times.
> >
> >
> > http://codereview.chromium.org/5868004/
> >

[jrbarnette, 2010-12-17 20:35:53]

On 2010/12/17 20:15:15, Daniel Erat wrote:
> Yes, there are probably a bunch of things that are still using
> login-prompt-ready that will need to be updated.  Since the signal is
> still getting emitted with this change, they should continue being
> just as wrong as they already are, right?
> 
Ah, yes.  I hadn't internalized the changes elsewhere.

LGTM!


> On Fri, Dec 17, 2010 at 12:11 PM,  <mailto:jrbarnette@chromium.org> wrote:
> > On 2010/12/17 03:28:17, Daniel Erat wrote:
> >>
> >> I'll send out changes to update the factory scripts that patch this file
> >
> > before
> >>
> >> checking in the change.
> >
> > You also need to update the platform_BootPerfTest autotest; it depends on
> > the name login-prompt-ready to determine boot times.
> >
> >
> > http://codereview.chromium.org/5868004/
> >

[jrbarnette, 2010-12-17 20:35:54]

On 2010/12/17 20:15:15, Daniel Erat wrote:
> Yes, there are probably a bunch of things that are still using
> login-prompt-ready that will need to be updated.  Since the signal is
> still getting emitted with this change, they should continue being
> just as wrong as they already are, right?
> 
Ah, yes.  I hadn't internalized the changes elsewhere.

LGTM!


> On Fri, Dec 17, 2010 at 12:11 PM,  <mailto:jrbarnette@chromium.org> wrote:
> > On 2010/12/17 03:28:17, Daniel Erat wrote:
> >>
> >> I'll send out changes to update the factory scripts that patch this file
> >
> > before
> >>
> >> checking in the change.
> >
> > You also need to update the platform_BootPerfTest autotest; it depends on
> > the name login-prompt-ready to determine boot times.
> >
> >
> > http://codereview.chromium.org/5868004/
> >

[jrbarnette, 2010-12-17 21:33:58]

On 2010/12/17 20:35:54, jrbarnette wrote:
> On 2010/12/17 20:15:15, Daniel Erat wrote:
> > Yes, there are probably a bunch of things that are still using
> > login-prompt-ready that will need to be updated.  Since the signal is
> > still getting emitted with this change, they should continue being
> > just as wrong as they already are, right?
> > 
> Ah, yes.  I hadn't internalized the changes elsewhere.
> 
> LGTM!
> 
Ah, I just realized one other small issue:  send_boot_metrics
uses the login-prompt-ready stat rather than the new boot-complete
stat.  It should probably be updated now, if it's not too late.

OTOH, it can also wait, if necessary.

[Daniel Erat, 2010-12-17 22:02:24]

On Fri, Dec 17, 2010 at 1:33 PM,  <jrbarnette@chromium.org> wrote:
> On 2010/12/17 20:35:54, jrbarnette wrote:
>>
>> On 2010/12/17 20:15:15, Daniel Erat wrote:
>> > Yes, there are probably a bunch of things that are still using
>> > login-prompt-ready that will need to be updated.  Since the signal is
>> > still getting emitted with this change, they should continue being
>> > just as wrong as they already are, right?
>> >
>> Ah, yes.  I hadn't internalized the changes elsewhere.
>
>> LGTM!
>
> Ah, I just realized one other small issue:  send_boot_metrics
> uses the login-prompt-ready stat rather than the new boot-complete
> stat.  It should probably be updated now, if it's not too late.

Good catch.  I've added this to the change.

> OTOH, it can also wait, if necessary.
>
>
> http://codereview.chromium.org/5868004/
>