Consider setting background attributionsrc request priority to low

[yoavweiss]

In Make a background attributionsrc request, we're defining a request, but it doesn't seem like we're setting its priority to low, nor setting it to a "keepalive" one. What makes it a background request? Any guardrails against such requests competing on bandwidth with more critical ones?

[apasel422]

It's true that the term background doesn't have much meaning here, but we do set keepalive true on these requests.

[yoavweiss]

You're right, I missed the keepalive bit. Might also be worthwhile to set their priority to low.

[apasel422]

Might also be worthwhile to set their priority to low.

What would the consequences of that be? The Fetch spec seems to propagate "priority" and "internal priority" but doesn't explain how they're used.

[yoavweiss]

They are used internally by the browser to set an implementation defined priority both in internal queues as well as on the network (eventually translated to H3 priorities)

[apasel422]

Thanks. Perhaps the Fetch spec could have some kind of non-normative guidance on when to use each value.

[csharrison]

@yoavweiss do you expect setting the priority to low could impact things like network loss or dropped requests? I want to understand the consequences a bit deeper.

[yoavweiss]

^^ @pmeenan

I wouldn't expect it to result in dropped requests other than cases where the reports would be currently sent right before the sender is killed. Depending on the implementation, the sender could be either the renderer or the browser process.

The tradeoff here is that if the requests are not marked as low priority, they could get in the way of requests for user visible content.

Another question that impacts priority, CSP, etc is - what should be these request's destination?

[apasel422]

Another question that impacts priority, CSP, etc is - what should be these request's destination?

Based on https://fetch.spec.whatwg.org/#destination-table the empty string seems most relevant.

[csharrison]

Another question that impacts priority, CSP, etc is - what should be these request's destination?

I'm not sure how that should be chosen but maybe following the ping attribute makes sense which uses the empty string: https://html.spec.whatwg.org/multipage/links.html#hyperlink-auditing

edit: oops race condition with @apasel422 :)

[csharrison]

I wouldn't expect it to result in dropped requests other than cases where the reports would be currently sent right before the sender is killed. Depending on the implementation, the sender could be either the renderer or the browser process.

Ack, this may impact us depending on the implementation given that it isn't unusual for these background requests to be made as the context is being navigated away (e.g. the user clicks on an ad). Before making this change I'd want to double check the impact here.

[dmdabbs]

Regarding priority I've noticed that the 'background' fetch is always High priority. And the order in which Chrome fetches (judging by DevTools Network) depends on how the DOM element is added.

As inline markup, the 'foreground' is always first. <img src="https://www.google.com/gen_204?ara=foreground" attributionsrc="https://www.google.com/gen_204?ara=background">

As injected markup the 'background' is first, again judging by Network listing.

document.getElementById('somediv').innerHTML = '<img src="https://www.google.com/gen_204?ara=foreground" attributionsrc="https://www.google.com/gen_204?ara=background">'

This may vary by element type, but I didn't compare supported elements.

[yoavweiss]

I don't think the compat label is required here, as this is something we can easily change after shipping, and have no reason to believe would result in site breakage.

[csharrison]

I don't think the compat label is required here, as this is something we can easily change after shipping, and have no reason to believe would result in site breakage.

I added this out of an abundance of caution because we would want to roll this out carefully and monitor potential loss in the APIs. I agree the impact is probably minimal so it might be overkill, but it's worth being a bit more careful IMO.