Closed Bug 811513 Opened 13 years ago Closed 13 years ago

sum field for MOZ_STORAGE_ASYNC_REQUESTS_SUCCESS off by 1

Categories

(Toolkit :: Telemetry, defect)

x86_64
Windows 7
defect
Not set
normal

Tracking

()

RESOLVED WORKSFORME

People

(Reporter: dzeber, Unassigned)

Details

Attachments

(1 file, 2 obsolete files)

For a few of the pings containing MOZ_STORAGE_ASYNC_REQUESTS_SUCCESS histogram, the sum field seems to be off by 1 (sometimes above, sometimes below). I wonder if this has to do with the large number of observations (~1000s) per histogram. I've attached some output from some code to examine the inconsistencies I was seeing. Each row corresponds to one session measurement, and column "count" is the number of pings with the measurements listed in the row. The dataset had 44025 pings (from FF v18 Nightly Sep 3-9). Column "n.observations" is the number of values added to the histogram (ie. sum(h$values)) and sum.field is the value of h$sum which accompanied histogram h. Columns n.zeros/ones/twos are the number of observations in each respective bin. For booleans and flags we should have sum.field = n.ones + 2*n.twos and n.observations = n.zeros + n.ones + n.twos.
Attachment #681239 - Attachment is obsolete: true
The issue still seems to be present. I've attached some data from Nightly v21 from January. Each row corresponds to a MOZ_STORAGE_ASYNC_REQUESTS_SUCCESS histogram from a telemetry session. For each one, given are the total number of observations, the sum value, and the number of observations of each value. As this histogram is a boolean, the sum should equal the number of ones. However, in these examples, the sum is off by 1, above or below. I'm not sure what's causing this, and I was wondering if this pointed to a bug somewhere in the data collection, or if this is something client-specific.
Wrong attachment before - sorry - but same conclusions.
Attachment #716825 - Attachment is obsolete: true
Telemetry currently does not synchronize updates to the histograms. So if you have multiple threads (as might be expected with asynchronous storage requests) updating a single histogram, you might see some inconsistencies in that histogram's data. We have code to try to identify inconsistent histograms in Telemetry already: http://dxr.mozilla.org/mozilla-central/toolkit/components/telemetry/Telemetry.cpp#l1147 http://dxr.mozilla.org/mozilla-central/ipc/chromium/src/base/histogram.cc#l352 You can see from the second link that we're willing to send histograms even if sums are slightly off. I don't think this counts as a bug.
Ok - thanks for clarifying this. I don't think I'd seen that part of the code. I'll close the bug.
Status: NEW → RESOLVED
Closed: 13 years ago
Resolution: --- → WORKSFORME
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: