1165789 - [e10s] Closing a hangup-ed tab causes Chrome hang too

Status: RESOLVED DUPLICATE of bug 1171708

(Firefox :: General, defect)

Description

[Alice0775 White]

Attachment: 675417 hung.html

I think there is no advantage of e10s.

Reproducible : always

Steps to Reproduce:
1. Save attached html to local
2. Start Browser with e10s
3. Close tabs other than about:home 
4. Open the saved file in New tab
5. Close the tab while loadding spinner is spinning
6. Attempt to close browser

Actual Results:
Browser hangs

Comment 1

[Johannes Pfrang [:johnp]]

I'm sorry to disappoint you, but I can reproduce the hang on non-e10s just by clicking on your attachment.

Comment 2

[Alice0775 White]

Yes I know. the attachement was made from Bu67541.
So, I am thinking non-e10s === e10s.

Comment 3

[Tracy Walker [:tracy]]

Doesn't block e10s if it's reproducible on non-e10s

Comment 4

[Johannes Pfrang [:johnp]]

Well, I think it's a valid point that this probably should only hang the content, but not the chrome process.
So that's two issues here:

1. An e10s/non-e10s hang
2. e10s doesn't prevent the content process from hanging the chrome process (I'm not sure if that's in-scope for e10s, but afaik it should)

Comment 5

[Johannes Pfrang [:johnp]]

Correction of my previous comment:

Comment #1 indicates, that the hang occurs after closing the loading-spinner-tab, so the hang may be related to some e10s tab-close code waiting endlessly on some answer from the hanging content process (just a theory without knowledge of any of the code).

Comment 6

[Tracy Walker [:tracy]]

ok, tossing it back into triage.

Comment 7

[Johannes Pfrang [:johnp]]

/s/"tab-close"/"(on-)tab-close or spinner-related"
Also re-add blocking e10s-meta-bug.

Comment 8

[:Felipe Gomes (needinfo for replies!)]

I managed to reproduce this OSX (Yosemite).

Upon loading the page I closed the tab, and the tab closed properly in the UI but switching to any other tab would only show the spinner.

After that I tried to Quit firefox with Cmd+Q and then both processes ended up hanging. I sampled them with Mac's activity monitor, i'll attach the files

Comment 9

[:Felipe Gomes (needinfo for replies!)]

Attachment: hang-parent.txt

Comment 10

[:Felipe Gomes (needinfo for replies!)]

Attachment: hang-child.txt

Comment 11

[Johannes Pfrang [:johnp]]

(In reply to :Felipe Gomes from comment #8)
> Upon loading the page I closed the tab, and the tab closed properly in the UI but switching to any other
> tab would only show the spinner.

With dom.ipc.processCount=1 that seems to be expected, as the hanging content process cannot handle any other tab.

This seems related to bug 1118517 & bug 1116884. Reading those I now realize, it looks like we do kill content processes and have a hang detector, but it doesn't seem to work in this instance...

Comment 12

[Blake Kaplan (:mrbkap) (inactive)]

Felipe's parent stack shows that we're blocked on a CPOW, possibly trying to close the tab. Bill, we fixed the sessionstore CPOW on tab close, right?

Comment 13

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

Session store no longer uses CPOWs when tabs are closed, but it still uses them when you close a window or quit the app. I suspect the nsObserverService::NotifyObservers call in the parent stack is for quit-application-requested. I filed bug 1171708 about this since I don't think we have a bug on it yet.

Also, I suspect the hang detector is not kicking in because we're probably not hanging in JS or plugins. Bug 1171700 would help with that.