Closed Bug 1280229 Opened 8 years ago Closed 8 years ago

Add telemetry to XHR::Constructor in workers

Categories

(Core :: DOM: Workers, defect)

49 Branch
defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla50
Tracking Status
firefox50 --- fixed

People

(Reporter: baku, Assigned: baku)

References

Details

Attachments

(1 file)

      No description provided.
Attached patch xhr.patchSplinter Review
Attachment #8762872 - Flags: review?(bugs)
And what are we trying to count here? The number of XHR doesn't look too useful by itself.
XHR in workers is a feature that, at some point, we should rewrite from scratch. I say that for several reasons:

1. the current setup is not great: we proxy everything to the main-thread.
2. the XHR implementations in workers and main-thread don't share anything and it's hard to keep them in sync.
3. etc etc etc

In order to evaluate how urgent is this rewriting, I would like to know how much this feature is used on the web.
Do we already have these numbers? Or do we care about these numbers?
Flags: needinfo?(bugs)
Well, what does counting XHRs tell you, other than XHR is being used.
Don't you want to know like how many pages use XHR in workers or something like that?
Flags: needinfo?(bugs)
Comment on attachment 8762872 [details] [diff] [review]
xhr.patch

Hmm, ok, maybe this gives some information. 
But it is still a bit unclear to me how we'd use the information.
Is it bad if there are lots of XHRs? And what is lots? Does it matter what kind of XHRs? Like, if transferring just small amount of data, then certain issues in XHR impls don't show up.
So, think about if the telemetry probe is actually answering to the question you're asking ;)
Attachment #8762872 - Flags: review?(bugs) → review+
Pushed by amarchesini@mozilla.com:
https://hg.mozilla.org/integration/mozilla-inbound/rev/b850837edec2
Add telemetry to XHR::Constructor in workers, r=smaug
https://hg.mozilla.org/mozilla-central/rev/b850837edec2
Status: NEW → RESOLVED
Closed: 8 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla50
bsmeberg, were you approached for a data review for this probe? It seems to not meet any of the criteria for data collection[1].

Also, :baku, now it has an alert [2] and I'm not exactly sure what to do with that since everything's a little shy on description.

[1]: https://wiki.mozilla.org/Firefox/Data_Collection
[2]: http://alerts.telemetry.mozilla.org/index.html#/detectors/1/metrics/1885/alerts/?from=2016-06-30&to=2016-06-30
Flags: needinfo?(benjamin)
Flags: needinfo?(amarchesini)
Note that all new Telemetry probes need data collection review per:
https://wiki.mozilla.org/Firefox/Data_Collection
Thanks, Chris for your comment. What I want to see from this telemetry is this:

Many XHR steps in workers are synchronous: they block the worker thread, a runnable is sent to the main-thread, something is done, we go back to the worker and we continue the operation. With this telemetry information I want to know which of these operations is slow so that we can optimize it.
Flags: needinfo?(amarchesini)
I'm afraid I don't understand how getting counts of how many worker XHRs are constructed provides that information. What does the information mean if it's mostly 1? What action can you take if there are a lot of workers that use XHR only a prime number of times (which seems to be happening, sorta [1])?

Is the alert [2] a signal of a good thing or a bad thing?

[1]: https://mzl.la/29ef0Tu
[2]: http://alerts.telemetry.mozilla.org/index.html#/detectors/1/metrics/1885/alerts/?from=2016-06-30&to=2016-06-30
Flags: needinfo?(amarchesini)
Ah no sorry. I totally confused what this telemetry ID was used for. So, yes, here I wanted to count of often XHR API is used in workers. There are a lot of things to do for this API: from optimize it to rewrite it from scratch.
I need data to see how urgent this is.
Flags: needinfo?(amarchesini)
Do you actually need opt-out data from release users for this?
Data from the beta channel should be enough for prioritizing and doesn't require the "opt-out" part.
Based on comment 6 and others here, I think we clearly have not met the high user-value bar for opt-out data collection here. Georg or Andrea, can you please immediately remove the opt-out tag from this?

I also expect that this should not be expires: never unless there is a much-clearer problem statement of how this will be monitored. The automated alerts that we currently have are clearly not measuring what you care about.
Flags: needinfo?(benjamin)
I file a follow up bug for removing the opt-out flag.
Blocks: 1284644
So we have now some data about the usage in Nightlies. Is that enough?
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: