All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=814763

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=7b36056ae0316736a5e0b8e01c96d98218e00786731d7cbfe4d177305231cd06


Bot(s) for this bug's original alert(s):

android-webview-nexus5X
android-webview-nexus6

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14ddc600440000

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14d7f8e1440000

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/16e37eb1440000

danakj: Not sure why the bisect's not updating the bug, but it looks like r538150, "Invalidate resources in ResourcePool when moving to/from gpu raster". Can you take a look? 

This will continue to report resources that are in use in the browser/display compositor. So it's an accounting change to keep ownership in the renderer, but not a usage change. Tracing to confirm.

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/16e37eb1440000

Invalidate resources in ResourcePool when moving to/from gpu raster by danakj@chromium.org
https://chromium.googlesource.com/chromium/src/+/1c8ab25a7851ba87c989a69874ab9b96b1cc8945

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

📍 Found significant differences after each of 8 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/14d7f8e1440000

Standardize the name to launch id in NotificationPlatformBridgeWin by chengx@chromium.org
https://chromium.googlesource.com/chromium/src/+/88f12f0b9748f9e9ea300b5b8fe2648d86b5a5d3

Quick Cleanup from https://chromium-review.googlesource.com/c/chromium/src/+/922799 by robliao@chromium.org
https://chromium.googlesource.com/chromium/src/+/b14da7b8517a3325b26267e68285bf8feb534e36

Fix mime sniffer, so OnReadCompleted(EOF) is replayed if needed. by lukasza@chromium.org
https://chromium.googlesource.com/chromium/src/+/7aa64c8089a805b1435a2ee9f768d8dc47d97b6b

Make memory_instrumentation service responsible for adding heap dumps. by erikchen@chromium.org
https://chromium.googlesource.com/chromium/src/+/67fff826658bea7a7f36dc7ea4c5dbcb27438636

Scroll to offscreen nodes when reading them from search+s mode. by katie@chromium.org
https://chromium.googlesource.com/chromium/src/+/0475b9d3f8a02b0455ecdc50cff267daa0ec0edd

Disable ProcessManagerBrowserTest.NestedURLNavigationsToExtensionBlocked on Mac and Linux by chongz@chromium.org
https://chromium.googlesource.com/chromium/src/+/1bdf9c8041e83302f352ce42ee58e03f8e6347c4

Disable ParallelDownloadTest.ResumptionLastSliceUnfinished on Windows. by xingliu@chromium.org
https://chromium.googlesource.com/chromium/src/+/959bc37c5d5aa34c9b8d8833fe761a488180c444

Changed std::find() to ContainsValue() in src/android_webview/* by ganesh.borle@samsung.com
https://chromium.googlesource.com/chromium/src/+/df3c9ce78232fb22fbdcdee955a7831b581f9023

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

Issue 815152 has been merged into this issue.

Before this change, resources would be deleted from cc immediately, which could cause flashing. After the change, they are kept alive in cc until returned from the display compositor, so the reporting shows more memory in cc, tho it would be held by the display compositor regardless. So the actual amount of memory in use is not changed. This is confirmed in the traces, as the amount of memory in use is not changed, only what is reported.

The graph becomes very noisy afterward, because the memory capture races with the resources being returned to cc and freed, with some runs returning all the resources and showing the old memory reporting levels, and others returning some of the resources before it captures the memory trace.