992334 - Marketplace app is stuck on the loading screen upon launch

Status: RESOLVED WORKSFORME

(Marketplace Graveyard :: Consumer Pages, defect, P2)

Description

[krupa raj[:krupa]]

+++ This bug was initially created as a clone of Bug #990581 +++

Prerequisites:
FFOS phone, build 1.1, MP Stage app installed
FFOS phone, build 1.2 , MP Stage app installed
FFOS phone, build 1.4 , MP Stage app installed

Steps to reproduce:
1.On Android, load https://marketplace.allizom.org/
2.On FFOS devices, open MP-Stage app

Expected results:
MP stage is opened without any issues

Actual results:
FFOS: MP Stage loading screen is continuously displayed , but app is never opened

Logs id FFOS 1.1 : b606f

Comment 1

[krupa raj[:krupa]]

madalin, 

mind answering the questions asked at https://bugzilla.mozilla.org/show_bug.cgi?id=990581#c14 here?

I tried reproducing this issue on 1.4 and was able to see this once after trying ~10 times.

I see that on a fresh load, marketplace-stage takes a while but it does eventually load.

Comment 2

[Madalin Cotetiu]

(In reply to krupa raj[:krupa] from comment #1)
> madalin, 
> 
> mind answering the questions asked at
> https://bugzilla.mozilla.org/show_bug.cgi?id=990581#c14 here?
> 
> I tried reproducing this issue on 1.4 and was able to see this once after
> trying ~10 times.
> 
> I see that on a fresh load, marketplace-stage takes a while but it does
> eventually load.

On FF31 (Win7) MP works just fine (I can load -dev, stage and production every time)
On Inari FF OS 1.4 I can load production and -dev, but not stage. When I open the stage app the loading screen is continuosly displayed but the app is not opening. Please see this screenshot http://screencast.com/t/kAmkRj4N
When loading http://jsbin.com/tadezixa/2/ on my device this is the result: http://screencast.com/t/wRcBrjlw5F

As krupa mentioned after a fresh install marketplace-stage takes a while but it does eventually load.

I've come up with this steps to reproduce the issue after a fresh install of the mp-stage app on the Inari FF OS 1.4
1. Reinstall the stage app and confirm that it loads.
2. Restart the device.
3. Close any internet connection (wifi and data are off)
4. Open the stage app and confirm that it doesen't load, close stage app
5. Connect to wifi and reopen the stage app

Krupa can you please confirm that this steps work on your side? Thanks!

Comment 3

[krupa raj[:krupa]]

(In reply to Cotetiu Madalin from comment #2)
> (In reply to krupa raj[:krupa] from comment #1)
> > madalin, 
> > 
> > mind answering the questions asked at
> > https://bugzilla.mozilla.org/show_bug.cgi?id=990581#c14 here?
> > 
> > I tried reproducing this issue on 1.4 and was able to see this once after
> > trying ~10 times.
> > 
> > I see that on a fresh load, marketplace-stage takes a while but it does
> > eventually load.
> 
> On FF31 (Win7) MP works just fine (I can load -dev, stage and production
> every time)
> On Inari FF OS 1.4 I can load production and -dev, but not stage. When I
> open the stage app the loading screen is continuosly displayed but the app
> is not opening. Please see this screenshot http://screencast.com/t/kAmkRj4N
> When loading http://jsbin.com/tadezixa/2/ on my device this is the result:
> http://screencast.com/t/wRcBrjlw5F
> 
> As krupa mentioned after a fresh install marketplace-stage takes a while but
> it does eventually load.
> 
> I've come up with this steps to reproduce the issue after a fresh install of
> the mp-stage app on the Inari FF OS 1.4
> 1. Reinstall the stage app and confirm that it loads.
> 2. Restart the device.
> 3. Close any internet connection (wifi and data are off)
> 4. Open the stage app and confirm that it doesen't load, close stage app
> 5. Connect to wifi and reopen the stage app
> 
> Krupa can you please confirm that this steps work on your side? Thanks!

The STR doesn't work for me :(

Comment 4

[Wil Clouser [:clouserw]]

I couldn't reproduce this on 1.3.

Comment 5

[Wil Clouser [:clouserw]]

Please reopen with logcats if you can reproduce this.  I'm closing this since a P1/critical is making people nervous.

Comment 6

[krupa raj[:krupa]]

I just hit this issue on prod. Verbose logcat @ http://pastebin.mozilla.org/4777923

Comment 7

[Christopher Van Wiemeersch [:cvan]]

Yay!

04-08 01:49:50.970 E/GeckoConsole( 1084): Content JS LOG at https://marketplace.cdn.mozilla.net/media/fireplace/js/include.js?b=1396521250112:5 in s/<: [login] Calling navigator.id.watch

Things get bad after calling Persona. Can we have someone on the Persona team look at this?

Comment 8

[Jared Hirsch [:jhirsch] (he/him) (Needinfo please)]

Paging jedp.

Currently there are no reliable STR here, and it's very very hard to reproduce at all, so it seems like some kind of race condition must be in play.

The pastebin from comment 6 contains this error towards the bottom:

04-08 01:49:54.830 E/GeckoConsole( 1303): [JavaScript Error: "uncaught exception: target window is same as present window -- not allowed"]

I haven't looked at Persona postmessage code in a long time, but the jschannel source[1] throws that error if a postmessage channel is opened from a window to itself.

[1] https://github.com/mozilla/persona/blob/18e5ee8/resources/static/include_js/_jschannel.js#L248

cvan mentioned in IRC that the marketplace code recently moved from the include snippet to the native implementation on device, so perhaps the native implementation is running into some kind of race condition setting up postmessage callbacks?

Comment 9

[Jed Parsons (use needinfo, please) [:jedp, :jparsons]]

(In reply to Jared Hirsch (PTO until 4/7) [:_6a68] from comment #8)
> Paging jedp.

ohai!

> I haven't looked at Persona postmessage code in a long time, but the
> jschannel source[1] throws that error if a postmessage channel is opened
> from a window to itself.

I see this message whenever the persona include.js is executed, whether on a desktop browser on FirefoxOS.  It's been there since the dawn of time.  It's alarming, for sure, but I don't think it's an indicator of anything truly amiss.

There's an ancient issue for it: https://github.com/mozilla/persona/issues/2904

> cvan mentioned in IRC that the marketplace code recently moved from the
> include snippet to the native implementation on device, so perhaps the
> native implementation is running into some kind of race condition setting up
> postmessage callbacks?

6a68 or cvan, can you give me more information about this change?

From Persona's point of view, I don't think this should make a difference.  include.js always bound to navigator.mozId if it was available.  Basically, all the native code for persona does is set up the dom bindings; all the postmessage logic is still handled by the same hosted code.  So whether you call navigator.id or navigator.mozId, it's the same in the end.

See https://github.com/mozilla/persona/blob/dev/resources/static/provisioning_api.js#L630
and https://github.com/mozilla/persona/blob/dev/resources/static/authentication_api.js#L11

Krupa, do you have any way to know what the memory usage on your device is when this fails?

I'm asking because I'm wondering whether marketplace is getting OOM'd by the OS after the Trusted UI loads persona.

Sorry to needinfo so many people, but, well, need info :)

Comment 10

[Jared Hirsch [:jhirsch] (he/him) (Needinfo please)]

Sadly, I've got not much to add here. cvan or krupa will know more about the bug :-)

Comment 11

[krupa raj[:krupa]]

I don't know how to check for memory usage on my phone. Will check with the client QA team and find out. I'll report back when i have something useful.

Comment 12

[Jed Parsons (use needinfo, please) [:jedp, :jparsons]]

Krupa, this may be useful:  If you can attach adb, run in a shell on your computer:

  watch -n.5 adb shell b2g-ps --oom

The --oom shows you the out-of-memory score for each process.  I run that when running manual tests on device to see what the OS is killing and what it allows to live another day.  Also, with b2g-ps, seeing just the process id for b2g processes, you can kill them yourself and see what happens (adb shell kill -9 <procid>).  Thanks for your help!

Comment 13

[krupa raj[:krupa]]

Attachment: memory-report.json.gz

I got the memory report while this bug was happening. Let me know if this is useful.

Comment 14

[Jared Hirsch [:jhirsch] (he/him) (Needinfo please)]

so was this solved by/related to bug 990560?

Comment 15

[Kumar McMillan [:kumar]]

Not sure if it's the same thing but I've seen this happen a lot where I only see the Marketplace loading screen forever. The workaround for me is to delete and re-install the app. Even swipe-killing it does not solve it.

Comment 16

[Christopher Van Wiemeersch [:cvan]]

Is this still happening?

Comment 17

[Christopher Van Wiemeersch [:cvan]]

Krupa, is this still happening?

Comment 18

[krupa raj[:krupa]]

This bug was hard to reproduce in the first place. I haven't see this in a while. Let close it as WFM for now and reopen if this issue resurfaces.

Comment 19

[Christopher Van Wiemeersch [:cvan]]

https://github.com/mozilla/zamboni/commit/30e8177

Removed the hack that forced Persona to always load even when we didn't need it. So expect better perf all around now.