1126867 - Investigate memory consumption for duplicated histogram data

Status: RESOLVED FIXED

(Toolkit :: Telemetry, defect)

Description

[Georg Fritzsche [:gfritzsche]]

Per todays discussion on the FHR/Telemetry unification concerns [0] we want to go with client-side submission of both "reset per subsession" and "old style telemetry" histograms.

That means we'll have to duplicate histograms and need to investigate whether that has acceptable memory costs.

0:"Suggestions for the new unified FHR/Telemetry/Experiment ping", https://mail.mozilla.org/pipermail/fhr-dev/2015-January/thread.html#382

Comment 1

[Vladan Djeric (:vladan)]

Dupe of bug 1126986?

Comment 2

[Georg Fritzsche [:gfritzsche]]

Bug 1126986 seems a result of this bug.

Closing as fixed per [0]:

> In my Firefox 35 (release) session on Windows 7 with an uptime of 18 hours, I noted the following entries for Telemetry in about:memory:
> * 1MB TelemetryFile for single-process
>   * This is completely unexpected, I filed bug 1126986
> * 270 kB for Telemetry heap data allocated by C++ code
>   * Pretty much all the Telemetry data (histograms and other data types) are stored in C++ code
> * 106 kB for TelemetryPing.jsm
> * 74 kB for BrowserUITelemetry.jsm
> * 34 kB for TelemetryLog.jsm
> * 34 kB for TelemetryStopwatch.jsm
> * 29 kB for TelemetryTimestamps.jsm
> * 27 kB for UITelemetry.jsm
> * 23KB for TelemetryStartup.js
> 
> The actual Telemetry ping for this session was 112 KB in uncompressed JSON form.
> An 8 day session from Nightly 38 (non-E10S) on Windows 8.1 had very similar numbers.
> Each process in an E10S session had roughly the same cost as the numbers above.
> 
> I think an additional ~300KB memory overhead on desktop machines is acceptable.

And:
> A note on implementation:
> 
> Measurements such as slowSQL, chrome-hangs, and thread-hangs are implemented in C++ and TelemetryPing.jsm simply queries the relevant C++ code for the measurements when creating a ping.
> 
> We will have to modify each implementation to support two versions of the measurements (whole-session and sub-session).
> We might want to report only the whole-session version of these measurements initially.

[0] https://mail.mozilla.org/pipermail/fhr-dev/2015-January/thread.html#402