Issue 640860  has been merged into this issue.

ABle to reproduce the issue on win10 chrome version 52.0.2743.116 - memory keeps on increasing

Waited for 4 hors and memory increased to 81MB approx

Confirming the issue. This is working fine on firefox

Will work and update the regression details soon

Observed that the memory is increasing from M30 builds to latest canary. Hence can be considered a non regression issue.

I observed memory usage increase even with the following code:

<script>
function update() {
    requestAnimationFrame(update);
}
requestAnimationFrame(update);
</script>

Also, confirmed that node count was correctly decreased by GC.

Running Linux Chrome stable 53.
Observed that the size of a V8 heap snapshot didn't change from 3.3MB as the tab process memory usage increased by 1MB.

What's the update here? A memory leak like this is pretty serious.

Tested using Linux Chrome 56 with the original test case and --enable-heap-profiling.
Recorded a trace over the course of 5 minutes taking a memory trace before and after (but not during as it adds too much overhead).

The test was in the renderer with pid 12785.

Memory increased by ~4MB which was consistent with what I saw in Chrome's task manager.
Between the before and after traces:
 - blink_gc free_size increased by 1.4MB, mostly in HashTable and NormalPage4.
 - malloc's size increased by ~3MB, 1MB to <unspecified> and 2MB to memory metadata_fragmentation_caches.

Deleted: trace_raf_before.json.gz 662 KB

trace_raf_before.json.gz 662 KB Download

Deleted: trace_raf_after.json.gz 963 KB

trace_raf_after.json.gz 963 KB Download

I'm not sure how to interpret these numbers or whether I'm looking into the most useful data.

Could someone with experience in memory usage investigation share what they think of this?

hajimehoshi@: Would you take a look at this?

Bumping up the priority to P1.