1013452 - [B2G][Tarako][FTE] FTU is force closing after the user imports contacts and loads external links

Status: VERIFIED FIXED

(Firefox OS Graveyard :: Gaia::First Time Experience, defect)

Description

[Josh Schmitt [Joshs]]

Attachment: log.txt

Description:
FTU is force closing to the homescreen.

Repro Steps:
1) Update a Tarako to BuildID: 20140520014006
2) Proceed to the data/wifi in FTU and enable Cell Data and connect to a Wifi network
3) Proceed to 'Import Contacts' and import 'SIM Contacts'
4) Select 'Facebook', sign in and import all contacts
5) Select 'Next' and select 'Your Privacy' on the About FFOS page
6) Select one of the links and load the webpage
7) Repeat step 6 for all three links

Actual:
The FTU closes and user is unable to complete the FTU.

Expected:
FTU does not close and user is able to complete the FTU.

1.3T Environmental Variables:
Device: Tarako 1.3T
BuildID: 20140520014006
Gaia: 917174ee3812a43758bf43f7ba5f9416dcdb2ca8
Gecko: a3a14fb5ad51
Version: 28.1
Firmware Version: SP6821a-Gonk-4.0-5-12

Notes:
Repro frequency: 100%
See attached: logcat

Comment 1

[Josh Schmitt [Joshs]]

Issue does not repro on 1.3 Buri

1.3 Environmental Variables:
Device: Buri 1.3 MOZ
BuildID: 20140520024003
Gaia: 0ce948e378cab7ed3db20231281dd7ca2eb99779
Gecko: ee779e37aca1
Version: 28.0
Firmware Version: v1.2-device.cfg

Comment 2

[Naoki Hirata :nhirata (please use needinfo instead of cc)]

FTU does get killed.  Noming because it's a FTU bug.

Comment 3

[Joe Cheng [:jcheng] (please needinfo)]

triage: 1.3T+ as its FTU

Comment 4

[Joe Cheng [:jcheng] (please needinfo)]

ni? Danny for initial investigation

Comment 5

[Danny Liang [:dliang]]

It look like ftu was killed by LMK.
05-20 13:07:23.206: E/OomLogger(86): [Kill]: lowmem_shrink select 337 (Communications), adj 2, size 12325, to kill
05-20 13:07:23.206: E/OomLogger(86): [Kill]: lowmem_shrink send sigkill to 337 (Communications), adj 2, size 12325

Hi Josh, could you provide dmesg or slog for us to check memory usage?

Comment 6

[Danny Liang [:dliang]]

I can repo this issue, FTU was killed by LMK when I select the 2nd link of step 6.

Comment 7

[Gregor Wagner [:gwagner]]

We shouldn't crate a new process when we navigate. gmarty can you fix this?

Comment 8

[Gregor Wagner [:gwagner]]

Fcampo, can you help out here as well?

Comment 9

[Gregor Wagner [:gwagner]]

It doesn't reproduce for me :( I can open all 3 links multiple times. The only interesting thing I see is that when I open the marketplace link I see the browser process starting up and disappearing right away. Maybe that's the reason we kill the communication app?

Comment 10

[Gregor Wagner [:gwagner]]

Ok it reproduces maybe 1 out of 10 times with the marketplace link when it tries to open the browser app for a second. Bill, any idea who maintains this site and if we can avoid opening the browser app for a sec?

Comment 11

[[:fabrice] Fabrice Desré]

(In reply to Gregor Wagner [:gwagner] from comment #10)
> Ok it reproduces maybe 1 out of 10 times with the marketplace link when it
> tries to open the browser app for a second. Bill, any idea who maintains
> this site and if we can avoid opening the browser app for a sec?

Is it really opening the browser app or just a browser iframe? I think the browser iframe is due to the use of the mozId api.

Comment 12

[:gerard-majax]

I've flashed a tarako build from today (from pvtbuilds). I cannot reproduce so far, but I can confirm that memory is very low during FTU:

$ adb -s 19761202 shell b2g-info
                         |     megabytes     |
           NAME PID PPID CPU(s) NICE  USS  PSS  RSS VSIZE OOM_ADJ USER   
            b2g  84    1   88.9    0 17.9 21.5 28.0 143.6       0 root   
         (Nuwa) 316   84    0.9    0  0.2  1.7  5.6  45.7       0 root   
 Communications 334  316   49.3    1 22.9 26.9 33.9  96.3       2 app_334
(Preallocated a 621  316    0.9   18  5.4  7.5 12.4  50.7       4 root   

System memory info:

            Total 98.9 MB
     Used - cache 76.6 MB
  B2G procs (PSS) 57.5 MB
    Non-B2G procs 19.1 MB
     Free + cache 22.3 MB
             Free  1.5 MB
            Cache 20.8 MB

Low-memory killer parameters:

  notify_trigger 14336 KB

  oom_adj min_free
        0  4096 KB
        2  6144 KB
        6  7168 KB
        8 16384 KB
       10 18432 KB

Comment 13

[:gerard-majax]

I've reproduced, and .... I saw a Browser process for a small moment!

Comment 14

[:gerard-majax]

(In reply to Fabrice Desré [:fabrice] from comment #11)
> (In reply to Gregor Wagner [:gwagner] from comment #10)
> > Ok it reproduces maybe 1 out of 10 times with the marketplace link when it
> > tries to open the browser app for a second. Bill, any idea who maintains
> > this site and if we can avoid opening the browser app for a sec?
> 
> Is it really opening the browser app or just a browser iframe? I think the
> browser iframe is due to the use of the mozId api.

I can confirm Gregor's finding: the Preallocated process mutates to a Browser app for less than 2 secs, and then it gets killed.

> $ grep '395' ../../../../../../devices/tarako_logcat.log
> 05-30 06:08:42.530    84    84 I Gonk    : Setting nice for pid 395 to 18
> 05-30 06:08:42.530    84    84 I Gonk    : Changed nice for pid 395 from 0 to 18.
> 05-30 06:08:44.420   395   395 I Gecko   : ###################################### forms.js loaded
> 05-30 06:08:44.440   395   395 I Gecko   : ############################### browserElementPanning.js loaded
> 05-30 06:08:44.470   395   395 I Gecko   : ######################## BrowserElementChildPreload.js loaded
> 05-30 12:14:43.053    84    84 I Gonk    : Setting nice for pid 395 to 18
> 05-30 12:14:43.063    84    84 I Gonk    : Changed nice for pid 395 from 18 to 18.
> 05-30 12:14:43.063    84    84 I Gonk    : Setting nice for pid 395 to 1
> 05-30 12:14:43.063    84    84 I Gonk    : Changed nice for pid 395 from 18 to 1.
> 05-30 12:14:43.743   395   395 I Gecko   : ###################################### forms.js loaded
> 05-30 12:14:43.913   395   395 I Gecko   : ############################### browserElementPanning.js loaded
> 05-30 12:14:44.123   395   395 I Gecko   : ######################## BrowserElementChildPreload.js loaded
> 05-30 12:14:47.533    84   285 I Gecko   : [Parent 84] WARNING: waitpid failed pid:395 errno:10: file ../../../gecko/ipc/chromium/src/base/process_util_posix.cc, line 254
> 05-30 12:14:49.533    84   285 I Gecko   : [Parent 84] WARNING: waitpid failed pid:395 errno:10: file ../../../gecko/ipc/chromium/src/base/process_util_posix.cc, line 254
> 05-30 12:14:49.533    84   285 I Gecko   : [Parent 84] WARNING: Failed to deliver SIGKILL to 395!(3).: file ../../../gecko/ipc/chromium/src/chrome/common/process_watcher_posix_sigchld.cc, line 118

Comment 15

[:gerard-majax]

Well I do reproduce the appearance of Browser, but the FTU was not killed, in fact.

Comment 16

[:gerard-majax]

(In reply to Josh Schmitt from comment #0)
> Created attachment 8425640 [details]
> log.txt
> 
> Description:
> FTU is force closing to the homescreen.
> 
> Repro Steps:
> 1) Update a Tarako to BuildID: 20140520014006
> 2) Proceed to the data/wifi in FTU and enable Cell Data and connect to a
> Wifi network
> 3) Proceed to 'Import Contacts' and import 'SIM Contacts'
> 4) Select 'Facebook', sign in and import all contacts
> 5) Select 'Next' and select 'Your Privacy' on the About FFOS page
> 6) Select one of the links and load the webpage
> 7) Repeat step 6 for all three links
> 
> Actual:
> The FTU closes and user is unable to complete the FTU.
> 
> Expected:
> FTU does not close and user is able to complete the FTU.
> 
> 1.3T Environmental Variables:
> Device: Tarako 1.3T
> BuildID: 20140520014006
> Gaia: 917174ee3812a43758bf43f7ba5f9416dcdb2ca8
> Gecko: a3a14fb5ad51
> Version: 28.1
> Firmware Version: SP6821a-Gonk-4.0-5-12
> 
> Notes:
> Repro frequency: 100%
> See attached: logcat

I'm far from being able to repro. I may have reproduced it once over 20 retries, doing the same steps. Importing 13 contacts from SIM one, 108 from Gmail and 84 from Facebook. Opening all links several times. But no success with either builds from may 29th or may 30th.

Comment 17

[:gerard-majax]

Attachment: Logscale profile of the free memory while performing FTU

Attached is a screenshot of plotting the result of b2g-info ran every 2 secs on tarako while performing FTU steps. The Y is logscale, values are in MB.

Comment 18

[Gregor Wagner [:gwagner]]

Dale mentioned that persona uses an iframe and that loads the browser app.
The marketplace team has a version of the privacy page without the persona login:
 https://marketplace.cdn.mozilla.net/media/docs/privacy/en-US.html

We just have to figure out the locale and load https://marketplace.cdn.mozilla.net/media/docs/privacy/{LOCALE}.html

Comment 19

[Dale Harvey (:daleharvey)]

The iframe thing is a guess, but could be related, its certainly not the Browser app that is being started, but a Browser frame (they have the same name in b2g-ps)

Comment 20

[Josh Schmitt [Joshs]]

(In reply to Alexandre LISSY :gerard-majax from comment #16)
> (In reply to Josh Schmitt from comment #0)
> > Created attachment 8425640 [details]
> > log.txt
> > 
> > Description:
> > FTU is force closing to the homescreen.
> > 
> > Repro Steps:
> > 1) Update a Tarako to BuildID: 20140520014006
> > 2) Proceed to the data/wifi in FTU and enable Cell Data and connect to a
> > Wifi network
> > 3) Proceed to 'Import Contacts' and import 'SIM Contacts'
> > 4) Select 'Facebook', sign in and import all contacts
> > 5) Select 'Next' and select 'Your Privacy' on the About FFOS page
> > 6) Select one of the links and load the webpage
> > 7) Repeat step 6 for all three links
> > 
> > Actual:
> > The FTU closes and user is unable to complete the FTU.
> > 
> > Expected:
> > FTU does not close and user is able to complete the FTU.
> > 
> > 1.3T Environmental Variables:
> > Device: Tarako 1.3T
> > BuildID: 20140520014006
> > Gaia: 917174ee3812a43758bf43f7ba5f9416dcdb2ca8
> > Gecko: a3a14fb5ad51
> > Version: 28.1
> > Firmware Version: SP6821a-Gonk-4.0-5-12
> > 
> > Notes:
> > Repro frequency: 100%
> > See attached: logcat
> 
> I'm far from being able to repro. I may have reproduced it once over 20
> retries, doing the same steps. Importing 13 contacts from SIM one, 108 from
> Gmail and 84 from Facebook. Opening all links several times. But no success
> with either builds from may 29th or may 30th.

I was able to repro the bug after importing 151 sim contacts, 4 gmail, and 93 facebook.

I had to select all the external links multiple times in order for the bug to occur.

It appears the repro rate is a lot lower than when I initially filed the bug.

Environmental Variables:
Device: Tarako
Build ID: 20140530014002
Gaia: e68858693b71d917c9c5ee7e215f7ceea04635f7
Gecko: 1945abae19ff
Version: 28.1 (1.3) 
Firmware Version: SP6821a-Gonk-4.0-5-12

Comment 21

[Cody Roesch [:croesch]]

Removing QAWanted per comment 20.

Comment 22

[Aus Lacroix [:aus]]

So, I'm attempting to help with this and there's one snag here.

We don't seem to have very many version of the marketplace privacy policy here: https://marketplace.cdn.mozilla.net/media/docs/privacy/

I've only hit two that exist... en-US.html and es.html. They're not consistent with the use of the language culture name code (eg. en-US, fr-FR, es-ES.html) either.

Would it be at all acceptable to link to the en-US privacy policy?

Comment 23

[Andy McKay]

I think we should only link to the en-US.html when something better doesn't exist. Chances are something better does for some languages. All the privacy policies are here:

https://github.com/mozilla/legal-docs/tree/master/marketplace_privacy_policy

Comment 24

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

The simplest solution here seems to be to link to open an <iframe> to a very simple page hosted somewhere on mozilla servers. I would recommend linking to a static URL no matter what language the user has. The http request will include the user's chosen language in the accept-language header, and so the marketplace can redirect to different URLs depending on what languages we have translations for (which might change over time).

This way we can place the redirecting page somewhere where we are comfortable having dynamic pages, whereas the actual content can be hosted on a CDN serving static content. If that's what we want.

Comment 25

[Aus Lacroix [:aus]]

Attachment: Pull Request - Update Privacy Policy Link when Language Changes

Comment 26

[Aus Lacroix [:aus]]

(In reply to Jonas Sicking (:sicking) from comment #24)
> The simplest solution here seems to be to link to open an <iframe> to a very
> simple page hosted somewhere on mozilla servers. I would recommend linking
> to a static URL no matter what language the user has. The http request will
> include the user's chosen language in the accept-language header, and so the
> marketplace can redirect to different URLs depending on what languages we
> have translations for (which might change over time).
> 
> This way we can place the redirecting page somewhere where we are
> comfortable having dynamic pages, whereas the actual content can be hosted
> on a CDN serving static content. If that's what we want.

That's definitely ideal but I believe we're trying to work around it on our side for 1.3t specifically. Other releases will retain the current URL and allow the Marketplace team time to create the appropriate fix to make the page accessible without Persona.

Comment 27

[Aus Lacroix [:aus]]

Yep, looks like this one is mine...

Comment 28

[Fernando Campo (:fcampo)]

Comment on attachment 8431970 [details] [review]
Pull Request - Update Privacy Policy Link when Language Changes

Looks nice, few nits on github. 
Add some more tests, and a comment about the danger hardcoding default language.

Ask for r? again when addressed :)

Comment 29

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

We just have to make sure that whatever website we use to host the pages that we directly link to, is prepared to keep those links up and working for a couple of years. Including not making any changes to them that can cause tarako to run out of memory again.

Comment 30

[Aus Lacroix [:aus]]

(In reply to Jonas Sicking (:sicking) from comment #29)
> We just have to make sure that whatever website we use to host the pages
> that we directly link to, is prepared to keep those links up and working for
> a couple of years. Including not making any changes to them that can cause
> tarako to run out of memory again.

That's a very good point. I believe the Marketplace folks are aware of this, but, let's double check.

:andym, is this something you folks are comfortable with?

Comment 31

[Fernando Campo (:fcampo)]

Comment on attachment 8431970 [details] [review]
Pull Request - Update Privacy Policy Link when Language Changes

Code looks good, but Travis is complaining with jshint and tests not working.

Please merge when green

Comment 32

[Andy McKay]

It's just plain HTML and we'll keep it like that. We do change providers, but the domain and URL should stay the same.

Comment 33

[Aus Lacroix [:aus]]

Commit (v1.3t ONLY): https://github.com/mozilla-b2g/gaia/commit/fbead7b3bafb865fbbf269c6cd53ac1866f64f27

Fixed!

Comment 34

[Rachel Pribble [:rpribble]]

Verified fixed on the Tarako v1.3T MOZ ril.

Environmental Variables:
Device: Tarako 1.3T
BuildID: 20140602014001
Gaia: 335486c42498fa7a93c21e4d6121199728602ab8
Gecko: 55e4d83019e5
Version: 28.1
Firmware Version: SP6821a-Gonk-4.0-4-29

FTE completes as intended.

Comment 35

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Will: I just want to double check that the market place is aware that the set of URLs that we'll open is something that the marketplace needs to support for at least 2-3 years.

I.e. the client is hard-coding the set of languages that marketplace needs to support. Marketplace can neither add nor remove languages until tarako devices are no longer in use. Nor can we change the URL used by the various languages. And we'll need to make sure that those URLs will not consume enough memory that they run tarako out of memory.

The set of languages that you guys will need to support is listed here:
https://github.com/mozilla-b2g/gaia/commit/fbead7b3bafb865fbbf269c6cd53ac1866f64f27#diff-5e8933357bf1088e2d0aac0d20cf9b87R267

Is this solution workable?

Comment 36

[Wil Clouser [:clouserw]]

We agreed that we'll maintain the following URLs:

https://marketplace.cdn.mozilla.net/media/docs/privacy/bn-IN.html
https://marketplace.cdn.mozilla.net/media/docs/privacy/en-US.html
https://marketplace.cdn.mozilla.net/media/docs/privacy/hi.html
https://marketplace.cdn.mozilla.net/media/docs/privacy/ru.html
https://marketplace.cdn.mozilla.net/media/docs/privacy/ta.html

Those are the languages shipping with 1.3t (and maybe not even Russian) so those should be the only ones required.  I don't plan on removing any of the others, but we'll certainly be adding more to the website.