Oilpan: Request an Oilpan's GC when a V8's major GC is triggered

Oilpan: Request an Oilpan's GC when a V8's major GC is triggered

This CL schedules an Oilpan's GC when a V8's major GC is triggered to avoid the following scenario:

(1) A DOM object X holds a v8::Persistent to a V8 object. Assume that X is small but the V8 object is huge.
(2) X's DOM wrapper is created.
(3) The DOM wrapper becomes unreachable.
(4) V8 triggers a GC. The V8's GC collects the DOM wrapper. However, X is not collected until a next Oilpan's GC is triggered.
(5) If a lot of such DOM objects are created, we end up with a situation where V8's GC collects the DOM wrappers but the DOM objects are not collected forever. (Note that Oilpan's GC is not triggered unless Oilpan's heap gets full.)
(6) V8 hits OOM.

BUG=423047
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=183861

[haraken, 2014-10-16 08:20:48]

haraken@chromium.org changed reviewers:
+ keishi@chromium.org, oilpan-reviews@chromium.org

[haraken, 2014-10-16 08:20:48]

This CL is not yet for review.

keishi-san: Would you try to apply this CL and see if the requestAnimationFrame
leak is fixed?

[haraken, 2014-10-16 11:49:11]

haraken@chromium.org changed reviewers:
+ erikcorry@chromium.org, vegorov@chromium.org

[haraken, 2014-10-16 11:52:18]

Slava, Erik: Would you take a look at this change?

The underlying issue is that Oilpan's GC is not aware of the amount of memory
allocated behind Oilpan objects. V8's GC has the same problem and V8 solves the
problem by providing AdjustAmountOfExternalAllocatedMemory.

It's possible for Oilpan's GC to provide similar APIs but it seems too
complicated. I think it would be more reasonable to take the approach in this CL
in short term and revisit the problem when integrate V8's GC and Oilpan's GC in
long term.

[keishi, 2014-10-17 05:36:17]

This fixes the bug crbug.com/423047

May I land this?

[haraken, 2014-10-17 05:38:49]

On 2014/10/17 05:36:17, keishi wrote:
> This fixes the bug crbug.com/423047
> 
> May I land this?

Yeah, I guess this is a reasonable fix in short term at least. In long term the
issue should be solved by integrating Oilpan's GC with V8's GC and Blink
scheduler.

[keishi, 2014-10-17 05:51:55]

lgtm

[keishi, 2014-10-17 05:52:01]

The CQ bit was checked by keishi@chromium.org

[tkent, 2014-10-17 05:52:33]

tkent@chromium.org changed reviewers:
+ tkent@chromium.org

[tkent, 2014-10-17 05:52:34]

lgtm

[commit-bot: I haz the power, 2014-10-17 05:52:42]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/660753002/20001

[commit-bot: I haz the power, 2014-10-17 07:48:21]

Committed patchset #2 (id:20001) as 183861