Open Bug 1352706 Opened 7 years ago Updated 2 years ago

[e10s-multi] Firefox 55 breaks tabs & session with high number of processes and causes critical dataloss

Categories

(Core :: General, defect, P2)

Product:
Core
Component:
General
Version:
55 Branch
Type:
defect

Tracking

()

Status:
UNCONFIRMED

People

(Reporter: u589863, Unassigned)

Details

(Keywords: dataloss, multiprocess, perf, Whiteboard: [e10s-multi:+])

Attachments

(1 file)

Support
10.70 KB, text/plain
Details 
User Agent: Mozilla/5.0 (Windows NT 10.0; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/57.0.2987.111 Safari/537.36 Vivaldi/1.8.770.46

Steps to reproduce:

if any one has high processes say 32 process enabled then tabs and data loss occurs.




eg set 32 processes and 

click folder to open many/all tabs from 1 folder 

then quickly 

click folder to open many/all tabs from 2 folder.


Tabs of 1 folder are replaced/overwritten by tabs from folder 2


This also causes a lot of delay in opening tabs, and most of the time crashes the tabs or sometimes the browser.

Also tabs appear to open very slow *one by one* when opening a few bookmarks in new profile with e10s=on, e10s off = instant
why and will it be fixed as it feels browser is slow or using high cpu/ram/sdd

P.s: sorry for bad english and not so good STR
Blocks: e10s-multi-aurora, e10s-multi-beta, 1304548
Has STR: --- → yes
Keywords: dataloss, multiprocess, perf
OS: Unspecified → All
Hardware: Unspecified → All
Component: Activity Streams: General → General
Product: Firefox → Core
Whiteboard: [e10s-multi:?]
:mrbkap is going to deep dive and figure out what the action(s) required are.
Flags: needinfo?(mrbkap)
Sunil,  I have a few questions;

1) Can you reproduce the bug with just 4 content processes?
2) Are you running any add-ons, in particular any tab related add-ons? If so, can you reproduce this bug with them disabled?
3) About how many tabs are you opening from each folder?
(In reply to Tracy Walker [:tracy] from comment #2)
> Sunil,  I have a few questions;
> 
> 1) Can you reproduce the bug with just 4 content processes?
> 2) Are you running any add-ons, in particular any tab related add-ons? If
> so, can you reproduce this bug with them disabled?
> 3) About how many tabs are you opening from each folder?

HI will try.

1) NO , it has to be higer side try 32 or so to regularly reproduce, happensoften with more than 10 processes.
2) new profile.
3) 2 or 5 bookmarks in each
(In reply to Tracy Walker [:tracy] from comment #2)
> Sunil,  I have a few questions;
> 
> 1) Can you reproduce the bug with just 4 content processes?
> 2) Are you running any add-ons, in particular any tab related add-ons? If
> so, can you reproduce this bug with them disabled?
> 3) About how many tabs are you opening from each folder?

HI will try.

1) NO , it has to be higher side try 32 or so to regularly reproduce, happens often with more than 10 processes.
2) new profile.
3) 2 or 5 bookmarks in each
Give me a few time to find appropriate STR 100% working one.
tried to reproduce the problem since Tuesday
seems it occurs randomly but here's how

set dom.ipc.processCount=-1 
which is basically one process per tab(broken recently but works on build 31032017)

When did it break? 
when will it be fixed soon as it helps  a lot for testing with process per tab instead of manually setting it to 8-32 etc.
Please fix it , don't want to file a new bug just for it.

so set it dom.ipc.processCount=32 processes

Follow STR in comment 1 and out of 50+ times got to produce it 10 or so times.
Fresh profile with no extensions.

But found another way to reproduce it exactly with an add-on

install this add-on 

https://addons.mozilla.org/firefox/downloads/file/480890/stay_open_menu-3.0.2b1-fx.xpi

set the process to 8 or higher(6 processes still work as do 4 )

follow STR 1

You can now reproduce the problem.


P.S sorry for bad English and not clear
Flags: needinfo?(twalker)
Do it quickly to see the problem, try to open many tabs and close them.
Its very slow like slow motion.
Hi Sunil,

Thank you very much for your work tracking down a solid STR and initial conditions for this bug. It's been a huge help for us.

For the moment, we're concentrating on getting 4 content processes shipping, so bugs when there are higher numbers of content processes aren't going to block our release. That being said, if we have time, we'll definitely work on fixing this bug.
No longer blocks: e10s-multi-beta, 1304548, e10s-multi-aurora
Flags: needinfo?(mrbkap)
Whiteboard: [e10s-multi:?] → [e10s-multi:+]
(In reply to Blake Kaplan (:mrbkap) from comment #8)
> Hi Sunil,
> 
> Thank you very much for your work tracking down a solid STR and initial
> conditions for this bug. It's been a huge help for us.
> 
> For the moment, we're concentrating on getting 4 content processes shipping,
> so bugs when there are higher numbers of content processes aren't going to
> block our release. That being said, if we have time, we'll definitely work
> on fixing this bug.

thanks, can you please confirm the problem and set the right flags,
(In reply to Blake Kaplan (:mrbkap) from comment #8)
> Hi Sunil,
> 
> Thank you very much for your work tracking down a solid STR and initial
> conditions for this bug. It's been a huge help for us.
> 
> For the moment, we're concentrating on getting 4 content processes shipping,
> so bugs when there are higher numbers of content processes aren't going to
> block our release. That being said, if we have time, we'll definitely work
> on fixing this bug.

Hi this again surfaced in build 

17042017

with default 4 process, could some bug landing/back-out cause this??
Flags: needinfo?(mrbkap)
(In reply to Mefoster from comment #10)
> (In reply to Blake Kaplan (:mrbkap) from comment #8)
> > Hi Sunil,
> > 
> > Thank you very much for your work tracking down a solid STR and initial
> > conditions for this bug. It's been a huge help for us.
> > 
> > For the moment, we're concentrating on getting 4 content processes shipping,
> > so bugs when there are higher numbers of content processes aren't going to
> > block our release. That being said, if we have time, we'll definitely work
> > on fixing this bug.
> 
> Hi this again surfaced in build 
> 
> 17042017
> 
> with default 4 process, could some bug landing/back-out cause this??

How is the build from five days ago relevant? What about today's build?
(In reply to avada from comment #11)
> How is the build from five days ago relevant? What about today's build?
I doubt today's build will change anything. I reproduced this bug even before 4 content processes were enabled by default in Nightly. If anyone wants to reproduce this bug reliably, overload CPU and hard drive first, then use STR from the description.
(In reply to avada from comment #11)
> (In reply to Mefoster from comment #10)
> > (In reply to Blake Kaplan (:mrbkap) from comment #8)
> > > Hi Sunil,
> > > 
> > > Thank you very much for your work tracking down a solid STR and initial
> > > conditions for this bug. It's been a huge help for us.
> > > 
> > > For the moment, we're concentrating on getting 4 content processes shipping,
> > > so bugs when there are higher numbers of content processes aren't going to
> > > block our release. That being said, if we have time, we'll definitely work
> > > on fixing this bug.
> > 
> > Hi this again surfaced in build 
> > 
> > 17042017
> > 
> > with default 4 process, could some bug landing/back-out cause this??
> 
> How is the build from five days ago relevant? What about today's build?

Did you not get it?
Started since that build again!

(In reply to ajfhajf from comment #12)
> (In reply to avada from comment #11)
> > How is the build from five days ago relevant? What about today's build?
> I doubt today's build will change anything. I reproduced this bug even
> before 4 content processes were enabled by default in Nightly. If anyone
> wants to reproduce this bug reliably, overload CPU and hard drive first,
> then use STR from the description.

Yup this bug is very myterious
(In reply to Mefoster from comment #13)
> Did you not get it?
> Started since that build again!
> 
There was nothing to get! He didn't write that...
(In reply to avada from comment #14)
> (In reply to Mefoster from comment #13)
> > Did you not get it?
> > Started since that build again!
> > 
> There was nothing to get! He didn't write that...

> Hi this again surfaced in build 
> 
> 17042017

Are you daft?


Devs this is seriously a major bug for photon & e10s multi.
I spent some time last week trying to reproduce this locally and I was unable to. I was using the bookmark bar with a folder with several bookmarks in it, dom.ipc.processCount set to 100, and the "Open all in tabs" option. In all of my tests, I got the expected number of tabs and other behavior. Mefoster, could you attach your about:support output to this bug?

Thanks.
Flags: needinfo?(wrvwuevas)
Flags: needinfo?(twalker)
Flags: needinfo?(mrbkap)
Attached file Support — 
(In reply to Blake Kaplan (:mrbkap) from comment #16)
> I spent some time last week trying to reproduce this locally and I was
> unable to. I was using the bookmark bar with a folder with several bookmarks
> in it, dom.ipc.processCount set to 100, and the "Open all in tabs" option.
> In all of my tests, I got the expected number of tabs and other behavior.
> Mefoster, could you attach your about:support output to this bug?
> 
> Thanks.

Hi adding the file,
please remember this problem comes and goes randomly,
another way is this

(In reply to ajfhajf from comment #12)
> (In reply to avada from comment #11)
> > How is the build from five days ago relevant? What about today's build?
> I doubt today's build will change anything. I reproduced this bug even
> before 4 content processes were enabled by default in Nightly. If anyone
> wants to reproduce this bug reliably, overload CPU and hard drive first,
> then use STR from the description.
Flags: needinfo?(wrvwuevas)
I wonder if this is related to bug 1363240. There the steps are quite similar (a bunch of possibly lazy tabs open and then another bunch of tabs are overwriting them). The difference is the add-ons installed in the other bug (and there the work is driven by add-on scripts instead of manual opening).
(In reply to Gabor Krizsanits [:krizsa :gabor] from comment #18)
> I wonder if this is related to bug 1363240. There the steps are quite
> similar (a bunch of possibly lazy tabs open and then another bunch of tabs
> are overwriting them). The difference is the add-ons installed in the other
> bug (and there the work is driven by add-on scripts instead of manual
> opening).

see bug 1366477
Priority: -- → P2
(In reply to Gabor Krizsanits [:krizsa :gabor] from comment #18)
> I wonder if this is related to bug 1363240. There the steps are quite
> similar (a bunch of possibly lazy tabs open and then another bunch of tabs
> are overwriting them). The difference is the add-ons installed in the other
> bug (and there the work is driven by add-on scripts instead of manual
> opening).

Wasn't there a big fix for "lazy tabs" recently? Did that fix this bug?
Flags: needinfo?(epinal99-bugzilla2)
Sorry, I haven't followed BMO since June, so I don't really know about recent features.
Flags: needinfo?(epinal99-bugzilla2)
"WONTFIX"? "INVALID"? "WORKSFORME"?
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: