1642325 - Exclude referrer policy from the preload hash key to workaround bug 1637671

Status: RESOLVED FIXED

(Core :: DOM: Networking, task, P1)

Description

[Honza Bambas (:mayhemer)]

This is more for consideration to fix a non-small regression after preload has been enabled.

Bug 1637671 is a preexisting problem to preload.

Technically, the preload initiated at the prescan phase as a speculative load will never be used anyway, because any image tag later dynamically added to DOM will not be able to find it because of the different, updated referrer policy. I consider that a website bug, actually.

I'm not trying to workaround that web site bug here, though, I'm trying to prevent duplicate preloads and fix a performance regression or at least start a discussion what to do.

Comment 1

[Honza Bambas (:mayhemer)]

Attachment: Bug 1642325 - Exclude referrer policy from the preload hash key, r=ckerschb

Depends on D77664

Comment 2

[Emilio Cobos Álvarez (:emilio)]

So, I'm hitting a similar problem actually, with my patches for bug 1599160. We hit the caches for reloads, but given the referrer is different for two documents in the same origins, we fail the check in bug 1599160.

I'm not sure about different referrer policies, but should we be able to cache stylesheets for different referrers? That'd be a pretty massive performance optimization.

Comment 3

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/582660fa0f7c
Exclude referrer policy from the preload hash key, r=ckerschb

Comment 4

[Dorel Luca [:dluca]]

https://hg.mozilla.org/mozilla-central/rev/582660fa0f7c

Comment 5

[Florin Strugariu [:Bebe]]

== Change summary for alert #26163 (as of Tue, 09 Jun 2020 10:20:28 GMT) ==

Regressions:

6% raptor-tp6-facebook-redesign-firefox-cold not-replayed windows10-64-shippable opt 319.88 -> 337.83
3% raptor-tp6-facebook-redesign-firefox-cold confidence macosx1014-64-shippable opt 89.17 -> 86.50

Improvements:

18% raptor-tp6-facebook-redesign-firefox-cold fcp linux64-shippable opt 1,026.00 -> 844.17
17% raptor-tp6-facebook-redesign-firefox-cold fcp windows10-64-shippable-qr opt 956.00 -> 790.25
17% raptor-tp6-facebook-redesign-firefox-cold windows7-32-shippable opt 971.75 -> 805.53
17% raptor-tp6-facebook-redesign-firefox-cold windows10-64-shippable-qr opt 920.06 -> 763.73
17% raptor-tp6-facebook-redesign-firefox-cold windows10-64-shippable opt 986.99 -> 820.19
16% raptor-tp6-facebook-redesign-firefox-cold fcp linux64-shippable-qr opt 1,078.25 -> 902.42
16% raptor-tp6-facebook-redesign-firefox-cold fcp windows10-64-shippable opt 1,066.75 -> 896.08
16% raptor-tp6-facebook-redesign-firefox-cold fcp windows7-32-shippable opt 1,069.42 -> 898.75
14% raptor-tp6-facebook-redesign-firefox-cold windows10-64-shippable-qr opt 894.87 -> 766.01
13% raptor-tp6-facebook-redesign-firefox-cold fcp macosx1014-64-shippable opt 1,506.25 -> 1,304.42
13% raptor-tp6-facebook-redesign-firefox-cold linux64-shippable opt 921.44 -> 798.83
13% raptor-tp6-facebook-redesign-firefox-cold linux64-shippable-qr opt 953.73 -> 830.24
10% raptor-tp6-facebook-redesign-firefox-cold macosx1014-64-shippable opt 1,378.67 -> 1,234.60
8% raptor-tp6-yahoo-mail-firefox-cold not-replayed linux64-shippable opt 350.08 -> 323.17
7% raptor-tp6-yahoo-mail-firefox-cold not-replayed linux64-shippable-qr opt 351.50 -> 325.67
7% raptor-tp6-yahoo-mail-firefox-cold not-replayed macosx1014-64-shippable opt 342.08 -> 318.92
3% raptor-tp6-yahoo-mail-firefox-cold confidence linux64-shippable opt 72.58 -> 74.75

For up to date results, see: https://treeherder.mozilla.org/perf.html#/alerts?id=26163

Comment 6

[Emilio Cobos Álvarez (:emilio)]

(In reply to Florin Strugariu [:Bebe] (needinfo me) from comment #5)

== Change summary for alert #26163 (as of Tue, 09 Jun 2020 10:20:28 GMT) ==

Regressions:

6% raptor-tp6-facebook-redesign-firefox-cold not-replayed windows10-64-shippable opt 319.88 -> 337.83
3% raptor-tp6-facebook-redesign-firefox-cold confidence macosx1014-64-shippable opt 89.17 -> 86.50

Per bug 1640984, these are not performance regressions.

Comment 7

[Ryan VanderMeulen [:RyanVM]]

Does this need a Beta approval request to fix bug 1641532 for 78 as well?

Comment 8

[Honza Bambas (:mayhemer)]

(In reply to Ryan VanderMeulen [:RyanVM] from comment #7)

Does this need a Beta approval request to fix bug 1641532 for 78 as well?

No. Preload is enabled on early beta, which does no longer apply to 78. Let it ride.