Closed Bug 1381603 Opened 3 years ago Closed 2 years ago

Random ghost windows

Categories

(Core :: DOM: Core & HTML, defect, P3)

x86_64
Windows 7
defect

Tracking

()

RESOLVED INCOMPLETE

People

(Reporter: guijoselito, Assigned: mccr8)

References

(Blocks 1 open bug)

Details

(Whiteboard: [MemShrink:P2])

I was seeing high memory usage with the latest Nightly and checking about:memory I noticed some ghost windows.
I don't have specific steps to reproduce, unfortunately. Don't know if it matters, but I have only 2 content process set in about:preferences.

I think the easiest way to trigger the problem is to navigate to http://motorsport.com/ and load some news, scroll the pages and the comment section and don't close the tabs very fast (let them open for a few minutes).

Clicking in 'minimize memory usage' in about:memory make all go away.
Possibly related to bug 1383553 given the timing of the report? The offending patch was backed out yesterday and should be in today's nightly. Can you see if it still reproduces after updating?
Flags: needinfo?(guijoselito)
Bug 1159003 landed 4 days after I created this ticket.
I still see this problem with today's Nightly (07-28-2017).

56 (100.0%) -- ghost-windows
├──32 (57.14%) ── https://www.motorsport.com/f1/news/red-bull-ricciardo-verstappen-hungary-updates-935532/ [32]
├──23 (41.07%) ── https://www.motorsport.com/f1/news/gallery-key-f1-tech-shots-at-hungarian-gp-935334/ [23]
└───1 (01.79%) ── https://www.motorsport.com/f1/news/gallery-key-f1-tech-shots-at-hungarian-gp-935334/#image-renault-sport-f1-team-rs17-in-the-garage-14005248

Since I only had 1 tab opened for each news, I think it has to do with the comments section.
Flags: needinfo?(guijoselito)
Can you try to reproduce this problem?
Flags: needinfo?(continuation)
Sure, I'll take a look tomorrow.
Blocks: GhostWindows
Component: Untriaged → DOM
Assignee: nobody → continuation
Flags: needinfo?(continuation)
I can't reproduce the problem. Can you reproduce it without any addons? Could you give more specific steps? Thank you.
Flags: needinfo?(guijoselito)
Hmm, ok, it looks like I did manage to reproduce it one time, some how. But if you can come up with more specific steps on how to reproduce it, it would be appreciated.
Steps to reproduce (probably not minimal):
1 - Have set "2" in "Content Process Limit" in about:preferences and restart
2 - open this bug on Tab 1
3 - open http://forums.mozillazine.org/viewforum.php?f=23 on Tab 2 (different process from the Tab 1)
4 - open http://motorsport.com/ on Tab 3
5 - ctrl+click the five news right in the center of the page, so now you have 8 tabs
6 - in this third tab, scroll down to the "Giorgio Piola - F1 Technical Analysis" and click to open the first gallery in this tab
7 - scroll all the way down in all the tabs from Motorsport.com and wait 2 minutes
8 - close all the motorsport.com tabs, having left open only this bug and mozillazine
9 - wait 2 minutes and open about:memory and click on "Measure"

Result: you now have some ghost windows on the content processes.
Flags: needinfo?(guijoselito)
Thanks for the detailed steps.

When I follow them, I don't actually see any ghost windows, but motorsport.com is still listed as an active window, which is very strange:

204.26 MB (100.0%) -- explicit
├───63.54 MB (31.11%) -- window-objects
│   ├──38.43 MB (18.82%) -- top(https://www.motorsport.com/, id=2147484830)
│   │  ├──26.10 MB (12.78%) -- active
│   │  │  ├──22.38 MB (10.96%) -- window(https://www.motorsport.com/)

If I then open another about:memory tab and measure again, then motorsport is gone. I suspect that by opening a new tab it triggers a GC somehow.
(That was with a Nightly from July 1.)
Olli, do you have any ideas why a window might still be listed as "active" even though the tab has been closed? I'm not too familiar with the "active" status and how that works.
Flags: needinfo?(bugs)
If your investigation in bug 1388111 reveals that something really regressed, this bug has been filled right in that time window.
Whiteboard: [MemShrink]
Whiteboard: [MemShrink] → [MemShrink:P2]
Priority: -- → P2
Flags: needinfo?(bugs)
Moving to p3 because no activity for at least 1 year(s).
See https://github.com/mozilla/bug-handling/blob/master/policy/triage-bugzilla.md#how-do-you-triage for more information
Priority: P2 → P3
I couldn't reproduce this and it was filed a while ago, so I'm just going to mark it as incomplete.
Status: NEW → RESOLVED
Closed: 2 years ago
Resolution: --- → INCOMPLETE
Nothing personal, but I am not a dev in this project, so I can do this noob remark: Since some 40s Version I have only one Issue with Firefox: "Memory Leaks", heavy memory leaks, was way better in between. I've been reporting, many seem to have to and it is ghost windows very often. And I am here watching for literally years that most of these have been closed without fix, often due to age, but nothing has improved. I really hesitate to even start Firefox to check, if something has changed, I hardly do it every other release. I speak for few people from my world, who might well return to Firefox, if this memory stuff would be settled. Forget anything else, we do not care, fix the leaks or lose.
I spend most of my time fixing memory leaks. Unfortunately, it is not possible to fix specific leaks without being able to reproduce them. This bug did have specific steps, but they did not trigger the leak for me. I've been trying to come up with more general fixes for some of these issues, but I haven't come up with anything workable yet.
Component: DOM → DOM: Core & HTML
You need to log in before you can comment on or make changes to this bug.