Closed Bug 1123439 Opened 9 years ago Closed 9 years ago

[e10s] All browser tabs crash simultaneously when one of the tabs loads heavy stuff

Categories

(Firefox :: Untriaged, defect)

38 Branch
x86_64
Linux
defect
Not set
normal

Tracking

()

RESOLVED WORKSFORME
Tracking Status
e10s ? ---

People

(Reporter: sks3286, Unassigned)

References

()

Details

(Keywords: regression)

Crash Data

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Firefox/38.0
Build ID: 20150119030222

Steps to reproduce:

I restarted my Nightly after the 18 Jan 2015 update and ever since the tabs (all) were crashing very frequently. Almost like every 15-20 minutes. One of the tabs crashes and a domino effect causes all other tabs to crash simultaneously (or I guess its probably because e10s is running the chrome element of all tabs in the same web process so when one crashes, all crash). Sometimes right before a crash, i see an information bar (not unlike IE8's) at the top of the page informing me that one of the pages is slowing down Firefox but before I can respond to it, all the browser tabs crash and the message obviously disappear.
I did some more testing. The one thing I noticed was that pages that load flash plugin are pushing the browser over the edge. Happened when loading facebook and a couple of other websites that had embedded flash in their header. Also, the number of tabs open matters. So if FB is the only tab open then the browser happily chugs along but once you go over maybe 7 or 8 tabs, then the crash starts happening. In fact if you have 15 or more tabs open (as I normally do) then simply switching back to the fb tab after a gap of maybe 10-15mins is sufficient to crash the browser tabs (when e10s tries to repaint the page). Rapidly switching between tabs also initiates the crash.
Turning off e10s improved the situation and as far as I could see the problem did not recur. Turning on safe mode also helped but then the flash content doesn't load at all.
I also tried creating a fresh profile but running it in a normal e10s window caused the crash symptoms to recur.


Actual results:

All browser tabs crashed.


Expected results:

Crash should not have happened
tracking-e10s: --- → ?
Florin, can your team have a look at this to see if steps to reproduce can be found? If you can find reliable steps to reproduce a regression window would be the next thing needed.

Sid, since you're able to reproduce this already, it would save us some time if you could find the regression window using http://mozilla.github.io/mozregression/. If you need help with that, please email me directly and I'll try to walk you through it.
Flags: needinfo?(florin.mezei)
I was able to reproduce this issue and obtain this signature  using Latest Nightly (build ID: 20150119030222) on Ubuntu 14.04 x64.

Report ID:
bp-ffdd7714-06b1-4b9b-8b61-645632150120

Steps used:
1. Open Nightly
2. Open facebook in a new tab and navigate for about one minute.
3. Open other several flash websites (See list below)
4. Load two games in two different websites.
5. Switch between tabs several times and you will notice that tabs will crash.


Flash websites used:
http://www.y8.com
http://www.flash-games.net
http://www.miniclip.com/
https//kongregate.com
(other opened websites) http://www.emaramures.ro/ and http://www.emag.ro/


Also encountered another crash while trying to reproduce this: bp-50a0dba3-a7a5-4b84-8fc5-5484a2150120


Will continue to work on this issue and update it as soon as I'll have some clearer results.
Status: UNCONFIRMED → NEW
Crash Signature: [@ libpthread-2.19.so@0xc414 ]
Ever confirmed: true
Flags: needinfo?(florin.mezei)
QA Contact: cornel.ionce
I have this issue even when no heavy stuff is loaded.
Cornel, do you have any addons?
Flags: needinfo?(cornel.ionce)
I have been trying to run the mozregression tool since the morning but something or the other always interrupts the network connection and I got frustrated. I had put good date as 17 Jan 2015 (which was when the nightly was last working perfectly) and bad date as 18 Jan 2015 (when things went pear shaped). I was able to reach the 4th iteration with no issues whatsoever, if that helps. I will stop regression testing for now since my day job work is being hampered a bit if I am running any regression testing in the background. Let me know if one of the developers needs me to absolutely work through to the end and I'll find a way to do it.
Sid, can you please clarify a couple of things?
* last known good build for you is January 17, 2015
* first known bad build for you is January 18, 2015
* do you have any addons installed?

Thanks
Flags: needinfo?(sks3286)
Hi Anthony

* last known good build for you is January 17, 2015: CORRECT
* first known bad build for you is January 18, 2015: CORRECT
* do you have any addons installed? AdBlock Plus and FlashGot are the only two addons enabled. Its been that way since e10s landed in the nightly trunk. Also, as I mentioned earlier I ran the browser in safe mode and the problem was apparently gone but then so was the flash plugin capability of the browser
Flags: needinfo?(sks3286)
(In reply to Sid from comment #8)
> Hi Anthony
> 
> * last known good build for you is January 17, 2015: CORRECT
> * first known bad build for you is January 18, 2015: CORRECT

Here is the pushlog for mozilla-central between those dates:
http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=369a8f14ccf8&tochange=0c454540fc2b

>> do you have any addons installed? 
> AdBlock Plus and FlashGot are the only two addons enabled. Its been that way since 
> e10s landed in the nightly trunk. Also, as I mentioned earlier I ran the browser 
> in safe mode and the problem was apparently gone but then so was the flash plugin 
> capability of the browser.

Did you receive any updates to your add-ons or Flash on January 18?
Keywords: regression
Nope no updates for either of the addons between those two dates
(In reply to Sid from comment #10)
> Nope no updates for either of the addons between those two dates

Thanks Sid.

Brad, does anything stand out to you in the pushlog from comment 9?
Flags: needinfo?(blassey.bugs)
I too have attached a crash dump if it helps. Check the url section of the bug
(In reply to Brad Lassey [:blassey] (use needinfo?) from comment #4)
> Cornel, do you have any addons?

Nope, I was using a clean profile.

Please let me know if I can provide further assistance here.
Flags: needinfo?(cornel.ionce)
Nothing sticks out.

Blake, the stack has cookie related stuff in it https://crash-stats.mozilla.com/report/index/20b8cb1c-da65-45bb-88bb-ed4312150119

Sid, can you reproduce with your addons disabled?
Flags: needinfo?(sks3286)
Flags: needinfo?(mrbkap)
Flags: needinfo?(blassey.bugs)
(In reply to Brad Lassey [:blassey] (use needinfo?) from comment #14)
> Nothing sticks out.
> 
> Blake, the stack has cookie related stuff in it
> https://crash-stats.mozilla.com/report/index/20b8cb1c-da65-45bb-88bb-
> ed4312150119
> 
> Sid, can you reproduce with your addons disabled?

Yes Brad. It seems to appear less frequently (probably because AdBlock Plus is not overloading the engine) but it does appear nonetheless.

P.S. sorry for the late response. Day job issues
Flags: needinfo?(sks3286)
This should be fixed when bug 1128457 is fixed. Let's re-check after that.
Depends on: 1128457
Flags: needinfo?(mrbkap)
Flags: needinfo?(mrbkap)
Mistakenly filed against Firefox 38 and should be instead 38 Branch. Sorry for the spam. dkl
Version: Firefox 38 → 38 Branch
Sid, now that bug 1128457 is fixed, can you see if you can still reproduce this?
Flags: needinfo?(mrbkap) → needinfo?(sks3286)
Cornel, you were able to reproduce this.  Can you do so with latest Nightly?
Flags: needinfo?(cornel.ionce)
Mozilla/5.0 (X11; Linux x86_64; rv:41.0) Gecko/20100101 Firefox/41.0

I could not reproduce this anymore on the same machine using Latest Nightly, build ID: 20150615030204.
Flags: needinfo?(cornel.ionce)
(In reply to Cornel Ionce [QA] from comment #21)
> Mozilla/5.0 (X11; Linux x86_64; rv:41.0) Gecko/20100101 Firefox/41.0
> 
> I could not reproduce this anymore on the same machine using Latest Nightly,
> build ID: 20150615030204.

Ok, thanks for checking Cornel.  Based on that, resolving this WFM
Status: NEW → RESOLVED
Closed: 9 years ago
Flags: needinfo?(sks3286)
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.