touch-preview-images

ueventd in a busy loop on container-flipped image

Bug #1190792 reported by Steve Langasek on 2013-06-13

This bug affects 4 people

Affects		Status	Importance	Assigned to	Milestone
	touch-preview-images	Fix Released	Critical	Ricardo Salveti

Bug Description

It's possible this is specific to the container-flipped images; I haven't checked it against the pre-flip images.

ueventd is in a busy loop here, using 100% of CPU. Strace shows:

poll([{fd=4, events=POLLIN}], 1, -1) = 1 ([{fd=4, revents=POLLIN|POLLERR}])

lsof shows:

ueventd 460 root 4u netlink 6784 KOBJECT_UEVENT

Killing ueventd let it respawn, after which it was no longer in this busy loop.

Tags:

Revision history for this message

Tony Espy (awe) wrote on 2013-06-24:

@Steve

What device did this occur on? Any ideas on the scenario that led to ueventd getting into this state? Can you reproduce this consistently?

Revision history for this message

Steve Langasek (vorlon) wrote on 2013-06-24: Re: [Bug 1190792] Re: ueventd in a busy loop on container-flipped image

On Mon, Jun 24, 2013 at 03:48:59PM -0000, Tony Espy wrote:
> What device did this occur on? Any ideas on the scenario that led to
> ueventd getting into this state? Can you reproduce this consistently?

This was on a mako device with the container-flipped image. So this is
almost certainly an adverse interaction with udev (though fundamentally, it
remains a ueventd bug).

Pat McGowan (pat-mcgowan) on 2013-06-27

Changed in touch-preview-images:
assignee:	nobody → Oliver Grawert (ogra)

Revision history for this message

Oliver Grawert (ogra) wrote on 2013-06-27:

this happens if udev is started while ueventd is still doing its job, as discussed in the meeting we will need some way for upstart to know when ueventd is done so it can hold back udev until then.

Ricardo Salveti (rsalveti) on 2013-06-28

Changed in touch-preview-images:
assignee:	Oliver Grawert (ogra) → Ricardo Salveti (rsalveti)
status:	New → Confirmed
importance:	Undecided → High

Ricardo Salveti (rsalveti) on 2013-07-25

Changed in touch-preview-images:
status:	Confirmed → In Progress

Revision history for this message

Omer Akram (om26er) wrote on 2013-08-06:

This affects our autopilot testing in UTAH, as at times the screen unlock would fail due to this bug.

tags:	added: qa-touch
Changed in touch-preview-images:
importance:	High → Critical

Revision history for this message

Alexander Sack (asac) wrote on 2013-08-14:

yeah, this bug seems to be very painful. System needs to be calm to have autopilot tests run in a predictable manner.

Revision history for this message

Alexander Sack (asac) wrote on 2013-08-14:

can we get an update? fama est, that there is a fix already in upstart trunk?

Revision history for this message

Steve Langasek (vorlon) wrote on 2013-08-14:

On Wed, Aug 14, 2013 at 11:17:13AM -0000, Alexander Sack wrote:
> can we get an update? fama est, that there is a fix already in upstart
> trunk?

There is a related change on upstart trunk, but we should not release
upstart trunk before we have useful integration tests in place for it.

Also, the upstart change only helps /in connection with/ a change to ueventd
itself (to run once, then exit). I'm not sure why the upstart changes to
make this perfectly reliable, blocks us from landing the ueventd change
which should fix the busy loop problem.

Revision history for this message

Sergio Schvezov (sergiusens) wrote on 2013-08-23:

http://phablet.ubuntu.com/gitweb?p=CyanogenMod/android_system_core.git;a=commit;h=07a1e9d5ec8067ed8cd2d5a46c2ed9caef087298