Closed Bug 1037154 Opened 10 years ago Closed 10 years ago

b2g master no longer clears layers resources for offscreen processes

Categories

(Core :: Graphics: Layers, defect, P2)

All
Gonk (Firefox OS)
defect

Tracking

()

RESOLVED DUPLICATE of bug 1039883
blocking-b2g -

People

(Reporter: BenWa, Assigned: milan)

References

Details

(Keywords: memory-footprint, perf, regression, Whiteboard: [MemShrink][c=memory p= s= u=])

Similar to tabs on desktop, when a process is hidden/minimized/off the screen we used to delete the associated gralloc resources. Now they show up plain as day in about:memory and appear to pile up until presumably we OOM them.

We should fix this regression.
This seems very bad
Assignee: nobody → milan
blocking-b2g: --- → 1.4?
Can you give me example STR to use for the window here? I need end user STR to do a window against.
Flags: needinfo?(bgirard)
Whiteboard: [memshrink] → [MemShrink]
Keywords: footprint, perf
That would be very handy. I'm still a little but suspicious that the bug is not real and that I made a mistake because I'd imagine this wouldn't have slipped up for long (so maybe the window is small). Let's hope it's INVALID

The best way to reproduce is to:
1) Launch several app. Say: Settings, Gallery, Clock. (These shouldn't be killed, but in the foreground)
2) Return to the homescreen in the same default position (this should keep the reading as consistent as possible)
3) run tools/get_about_memory.py (cancel after it has printed out the URL, the GC stuff isn't required and hangs).
4) Measure the section under Gralloc. Actually you should also collect a data point with just the homescreen booted before stpe 0. The base gralloc number will fluctuate due to change/regressions in gaia and the layer tree. We're looking for final number (or delta) is high.

This might be a complex range as we regress and fix this over time.

If it's not much extra effort it might be interesting to label each about-memory report with the build id and the log of this before and after in case the regression range is complex.
Flags: needinfo?(bgirard)
Adding Sotaro in case this is related to the shared buffer manager parent memory leak discovered.
See Also: → 1037360
Was this found on 1.4?
Status: NEW → UNCONFIRMED
Ever confirmed: false
no, not AFAIK. We should get an exact window.
Priority: -- → P2
Whiteboard: [MemShrink] → [MemShrink][c=memory p= s= u=]
This bug might be dup of Bug 1039883.
See Also: → 1039883
Yes it is.
Status: UNCONFIRMED → RESOLVED
Closed: 10 years ago
Resolution: --- → DUPLICATE
issue has been marked as resolved, clearing qa-wanted keywords
Clearing nom based on Comment 9
blocking-b2g: 1.4? → -
You need to log in before you can comment on or make changes to this bug.