Comparing the counts of UMA.Discarded Log Events with UMA.RecordCounts, it looks like we're discarding:
* 0.5% of Windows log entries
* 0.3% of Mac log entries
* 0.3% of Linux log entries
* 0.0% (rounded down) of Android log entries
* almost exactly 0% of iOS log entries
* and, to save the best for last, 13.5% of ChromeOS log entries!

Eying the distribution of UMA.Discarded Log Events, it looks like it we raise the limit to 10k, we'd drop the discard rate to 3% for ChromeOS and to 0.1% on Windows.  (Most other platforms are similar.)

I think we should do it.

By the way, I don't think this discarding explains the Android user action / histogram discrepancy discussed on bug 659336.

What's the motivation for the limit in the first place?  I thought it was to limit the upload size.  If so, are we sure that it's fine to bump the threshold significantly?  Can we figure out why ChromeOS is so action-heavy, and see if there's some abuse of the actions mechanism that we can identify and fix?

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/bb1aaf95ea8fd6a5815d512a690ac8587ded2e7d

commit bb1aaf95ea8fd6a5815d512a690ac8587ded2e7d
Author: jwd <jwd@chromium.org>
Date: Thu Jan 19 22:22:16 2017

Removing *Scroll user actions because they are extremely spammy.

Specifically TouchscreenScroll and TrackpadScroll

BUG= 682329 

Review-Url: https://codereview.chromium.org/2642903002
Cr-Commit-Position: refs/heads/master@{#444862}

[modify] https://crrev.com/bb1aaf95ea8fd6a5815d512a690ac8587ded2e7d/content/browser/renderer_host/render_widget_host_view_event_handler.cc
[modify] https://crrev.com/bb1aaf95ea8fd6a5815d512a690ac8587ded2e7d/tools/metrics/actions/actions.xml

There's a little bit more discussion from the (older) internal bug: b/34200439

We know chromeOS is spammy because of trackpad (jesse submitted a fix to clean it up a bit).

We think the original motivation was bandwidth but I'm not convinced it's the right model for solving that issue. There's logic to capture how much bandwitdh UMA is using (on cellular) and limiting it - so we should reuse that logic instead.

For now I want to experiment to see what happens if we remove the limit, see if we get users sending huge records (or not).

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/02e248436f3736e6b85ab4c47452ce13611cdd11

commit 02e248436f3736e6b85ab4c47452ce13611cdd11
Author: rkaplow <rkaplow@chromium.org>
Date: Fri Jan 20 23:39:41 2017

Add a feature to control if we drop UMA records due to excess events.

Enabled by default as this is current behavior.

BUG= 682329 

Review-Url: https://codereview.chromium.org/2646163004
Cr-Commit-Position: refs/heads/master@{#445210}

[modify] https://crrev.com/02e248436f3736e6b85ab4c47452ce13611cdd11/components/metrics/metrics_service.cc
[modify] https://crrev.com/02e248436f3736e6b85ab4c47452ce13611cdd11/tools/metrics/histograms/histograms.xml

We changed how the limit works (and the amount of data).
See  https://bugs.chromium.org/p/chromium/issues/detail?id=683934#c21 for more details