1888815 - Firefox keeps trying to resolve cname-only https rr

Status: RESOLVED FIXED

(Core :: Networking: DNS, defect, P2)

Description

[lilydjwg]

User Agent: Mozilla/5.0 (X11; Linux x86_64; rv:126.0) Gecko/20100101 Firefox/126.0

Steps to reproduce:

open a matrix client and login into a mozilla.org account so firefox keeps polling the matrix for incoming messages.

Actual results:

Firefox keeps trying to resolve the https DNS record for mozilla.modular.im very frequently, maybe one for each HTTP request.

Expected results:

Shouldn't Firefox cache the result and calm down a bit?

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

The Bugbug bot thinks this bug should belong to the 'Core::Networking: DNS' component, and is moving the bug to that component. Please correct in case you think the bot is wrong.

Comment 2

[Sunil Mayya]

Thank you for reporting the issue.
I would expect DNS cache to be used.
Could you please share the output of about:networking#dns

Comment 3

[lilydjwg]

Attachment: network-dns.png

Here it is. After I opened my matrix tab (app.schildi.chat), I did a "clear cache" first, observed repeated HTTPS record requests (for mozilla.modular.im which is the matrix server) via Wireshark, and then refreshed and took the screenshot.

Firefox does have the A record cached (I don't see repeated requests for it in Wireshark also), but not for the HTTPS record.

Comment 4

[Valentin Gosu [:valentin] (he/him)]

Attachment: Bug 1888815 - Add test for HTTPS record caching r=#necko

Comment 5

[Valentin Gosu [:valentin] (he/him)]

The test confirms that the DNS code works correctly. But I've reproduced the issue locally, and tracked down the problem to:
https://searchfox.org/mozilla-central/rev/c09764753ea40725eb50decad2c51edecbd33308/netwerk/protocol/http/nsHttpChannel.cpp#6613-6617

if (!LoadIsTRRServiceChannel() &&
    (mLoadFlags & VALIDATE_ALWAYS ||
     BYPASS_LOCAL_CACHE(mLoadFlags, LoadPreferCacheLoadOverBypass()))) {
  mCaps |= NS_HTTP_REFRESH_DNS;
}

This means the host resolver is instructed to bypass the cache.
Considering how often this happens, this could actually be a perf issue.
We need to:

1. Figure out if these flags are correct & if we should always refresh the cache when VALIDATE_ALWAYS is set
2. Consider not refreshing the DNS cache if the negative cache entry is recent.

Comment 6

[Valentin Gosu [:valentin] (he/him)]

Attachment: Bug 1888815 - Do not set the NS_HTTP_REFRESH_DNS flag unless force-reloading the page r=#necko

This flag was previously being set for every page reload.
Only setting it for force-reload makes it so that we don't needlessly refresh
DNS cache entries and improves page load performance.

Comment 7

[Valentin Gosu [:valentin] (he/him)]

Not sure if the change will break anything, but let's see.
https://treeherder.mozilla.org/jobs?repo=try&revision=bae8b2349219526ac8bc03b72ae3f1005fdfad86

Comment 8

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/2329d6565203
Do not set the NS_HTTP_REFRESH_DNS flag unless force-reloading the page r=necko-reviewers,jesup

Comment 9

[Aron Cseh]

https://hg.mozilla.org/mozilla-central/rev/2329d6565203