1492466 - TVHeadend broken by TRR

Status: RESOLVED WONTFIX

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

Description

[Jan Alexander Steffens [:heftig]]

Setup:
Tvheadend (from git master) running on an intranet host http://angel:9981/
Firefox Nightly

With network.trr.mode=0, 2, 4 or 5, everything is fine.
With network.trr.mode=3, connection fails, as expected.

With network.trr.mode=1, loading the page succeeds but most of the interface does not load. Its log says:

> There seems to be a problem with the live update feed from Tvheadend. Trying to reconnect...

Followed by several "Reconnected to Tvheadend".

These seem to be messages from Tvheadend's "comet" WebSocket client, which repeatedly attempts to connect to ws://angel:9981/comet/ws .

As soon as devtools is opened, the next websocket attempt succeeds. If devtools is open when the page is loaded, it works as normal.

Comment 1

[Jan Alexander Steffens [:heftig]]

Attachment: trr-2.log

nsHostResolver:5 log for trr.mode=2, which works.

Comment 2

[Jan Alexander Steffens [:heftig]]

Attachment: trr-1.log

nsHostResolver:5 log for trr.mode=1, which fails.

Comment 3

[Jan Alexander Steffens [:heftig]]

Attachment: trr-1d.log

nsHostResolver:5 log for trr.mode=1 started with --devtools, which works.

Comment 4

[Daniel Stenberg [:bagder]]

Mode 1 is truly a race between a TRR resolve and a native resolve and the first result will be used. Positive or negative.

It makes mode 1 very ineffective and annoying for this kind of internal host names since whenever TRR happens to respond faster, it will return a negative result, while all the times the native response is faster you get positive result an one or more IP addresses to use.

Negative responses are also cached for a while (60 seconds by default) by Firefox so even resolves done immediately following a failed resolve will return a negative response.

So, I'm afraid this is actually working as intended. I wrong strongly recommend using mode 2 (trr-first) instead since that will properly fallback and do a native resolve in case TRR fails.

Comment 5

[Daniel Stenberg [:bagder]]

The race mode really shouldn't be used if fallback behavior is wanted. This is also why we recommend mode 2, the trr-first that falls back to native when it detects problems.