Closed Bug 673837 Opened 13 years ago Closed 13 years ago

Telemeterize js-compartment-count

Categories

(Toolkit :: Telemetry, defect)

defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla8

People

(Reporter: justin.lebar+bug, Assigned: sandervv)

References

Details

(Whiteboard: [MemShrink:P2])

Attachments

(1 file)

Split off from bug 672731 (see bug 672731 comment 4).

That bug is a follow-up to bug 672694, which added the js-compartment-count reporter.  From bug 672694 comment 3.

This bug is about hooking this count up to telemetry.  njn, do you want to do it?
Depends on: 672731
Whiteboard: [MemShrink]
Whiteboard: [MemShrink] → [MemShrink:P2]
I chose {min: 1, max: 1000, scale: LINEAR} for both compartment statistics. Do you agree? I'm not sure about the scale. Note that low numbers will dominate for these statistics.

What happens if the gathered statistics exceeds the maximum value (1000 in this case); will the data be lost?
Attachment #550583 - Flags: review?(tglek)
1000 is way too high. I'd browse around for a bit see what kind of numbers you get then base the range on that. If the stats later show too much on the maximum end of the scale can always increase it later
I think the data just ends up in the largest bucket, so you don't lose it altogether, you just don't see exactly how much it was.
I actually think 1000 is reasonable.  A decent number of people have hundreds of tabs open, and it's not hard to have multiple compartments per tab.  Also, JetPack-based add-ons create compartments like it's going out of style (bug 672443);  I saw 305 compartments one time with Bugzilla Tweaks installed when I had only 12 tabs open.

But I suspect it should be EXPONENTIAL, since there's a strong bias towards the lower numbers.  That's what EXPONENTIAL is for, right?
Hmm, but maybe they're not biased enough for EXPONENTIAL... the expected range is only 10^3.  So maybe LINEAR is best after all.
I am guessing that the number of tabs will cluster at <= 100 so EXPONENTIAL seems like a better fit, but that's just a guess. I'll leave this up to the domain experts.
If we have a min of 1 and max of 1000, what do the bucket ranges look like for LINEAR vs. EXPONENTIAL?
Comment on attachment 550583 [details] [diff] [review]
Added the histograms for "js-compartments-{user,system}"

I'm going to approve this with LINEAR changed to EXPONENTIAL.  I'll land it.  Thanks, Sander!
Attachment #550583 - Flags: review?(tglek) → review+
http://hg.mozilla.org/integration/mozilla-inbound/rev/30bdc4e72672

That's your first code in a Mozilla repo, Sander :)  Assuming there are no test failures, it'll be merged into mozilla-central in the next 24 hours or so.
Whiteboard: [MemShrink:P2] → [MemShrink:P2][inbound]
Merged:
http://hg.mozilla.org/mozilla-central/rev/30bdc4e72672
Status: NEW → RESOLVED
Closed: 13 years ago
Resolution: --- → FIXED
Whiteboard: [MemShrink:P2][inbound] → [MemShrink:P2]
Target Milestone: --- → mozilla8
Version: unspecified → Trunk
Assignee: nnethercote → sandervv
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: