4 ms·

> Which wasn't present in previous versions. So this is still a consequence of updates. Looks like HTTP/3 has been in Firefox for around a year

jacquesm 5y ago
Yes, but if it wasn't active then it might as well not have been. This is a great way to bypass continuity testing, I really pity all those people working the desks in hospitals right now using FireFox who are typically less savvy than your average HN'er in trying to get their work done. If you ship a browser with a time-bomb you are utterly irresponsible.
- Vinnl 5y ago
  At this time it doesn't appear to be a surreptitious update, but a load balancer issue: https://bugzilla.mozilla.org/show_bug.cgi?id=1749908#c17 https://bugzilla.mozilla.org/show_bug.cgi?id=1749908#c17
  - jacquesm 5y ago
    Sorry, but my traffic isn't supposed to go through anybody's centralized load balancer, least of all one operated by parties with whom I do not have a relationship on the basis of the delivery of some service. That's not how this works. HTTP/3 support is optional as far as I'm concerned and plenty of websites that I tested with do not support it and still failed due to this issue. Absolutely unacceptable.
    - Vinnl 5y ago
      I'd wait a bit for the postmortem; I don't think there's reason to conclude at this time that your traffic is going through someone's centralised load balancer. Another update says that the load balancer issue triggered a bug in Firefox, so presumably fixing the load balancer issue will prevent that bug from being triggered, and then the bug can be fixed separately after: https://bugzilla.mozilla.org/show_bug.cgi?id=1749908#c21 https://bugzilla.mozilla.org/show_bug.cgi?id=1749908#c21 But mostly: I have no real idea what's going on, so I'm awaiting the postmortem.
      - jacquesm 5y ago
        That part seems to be related to the telemetry issue, which, surprise has been re-enabled during one of the updates.
- kevingadd 5y ago
  This isn't a time bomb, it's a symptom of the way modern browsers tend to have centralized I/O which means disk and network traffic goes through a single chokepoint. HTTP3 traffic appears to be able to cause one of Firefox's socket threads to hang, and since everything goes through it, all your network traffic is now dead. Chromium uses a similar model of routing all I/O through specific places, so it's vulnerable to similar sorts of problems (though AFAIK it has never failed this badly in production). Basically, this could have happened at any point if a production service run by someone like Google, Facebook, Cloudflare etc managed to trigger a sufficiently bad bug in a modern browser's network stack. If not for the centralized I/O this would just hang a single browser process, which isn't as impactful since Firefox and Chrome both split content out into many processes. It would also make it more obvious which server(s) are responsible since only certain tabs would be dying. FWIW, as far as I know this centralized I/O model was popularized by Google, not Mozilla.
  - jacquesm 5y ago
    It is because: (1) I did not explicitly enable this (2) The telemetry setting seems to have re-enabled itself on some update (3) I don't want any services from Mozilla, I want a browser (4) It worked until it blew up revealing that in fact, I suddenly did have service dependencies And finally, the reason I use FireFox is exactly your last sentence, so to see that they are slipping this in under the radar is a pretty good reason to drop FF altogether, it looks as if they fail to understand the difference between shipping software and getting me hooked on some service that I am not even aware of existing. And on top of that re-enabling their telemetry when it was explicitly disabled. That really takes the cake.
    - kevingadd 5y ago
      This is caused by HTTP3. The only way for you to have never experienced this issue would be if Mozilla never integrated HTTP3, which isn't going to happen because the major service providers like Google and Cloudflare are adopting it. Telemetry and updates happening to use HTTP3 merely pushed the issue to the foreground, it would have happened eventually and possibly in a way that got fixed more slowly because it was happening intermittently and only to visitors of obscure websites. Switching to a browser without telemetry and automatic updates won't protect you from this kind of network stack bug.

jacquesm 5y ago

Yes, but if it wasn't active then it might as well not have been. This is a great way to bypass continuity testing, I really pity all those people working the desks in hospitals right now using FireFox who are typically less savvy than your average HN'er in trying to get their work done. If you ship a browser with a time-bomb you are utterly irresponsible.

Vinnl 5y ago
At this time it doesn't appear to be a surreptitious update, but a load balancer issue: https://bugzilla.mozilla.org/show_bug.cgi?id=1749908#c17 https://bugzilla.mozilla.org/show_bug.cgi?id=1749908#c17
- jacquesm 5y ago
  Sorry, but my traffic isn't supposed to go through anybody's centralized load balancer, least of all one operated by parties with whom I do not have a relationship on the basis of the delivery of some service. That's not how this works. HTTP/3 support is optional as far as I'm concerned and plenty of websites that I tested with do not support it and still failed due to this issue. Absolutely unacceptable.
  - Vinnl 5y ago
    I'd wait a bit for the postmortem; I don't think there's reason to conclude at this time that your traffic is going through someone's centralised load balancer. Another update says that the load balancer issue triggered a bug in Firefox, so presumably fixing the load balancer issue will prevent that bug from being triggered, and then the bug can be fixed separately after: https://bugzilla.mozilla.org/show_bug.cgi?id=1749908#c21 https://bugzilla.mozilla.org/show_bug.cgi?id=1749908#c21 But mostly: I have no real idea what's going on, so I'm awaiting the postmortem.
    - jacquesm 5y ago
      That part seems to be related to the telemetry issue, which, surprise has been re-enabled during one of the updates.
kevingadd 5y ago
This isn't a time bomb, it's a symptom of the way modern browsers tend to have centralized I/O which means disk and network traffic goes through a single chokepoint. HTTP3 traffic appears to be able to cause one of Firefox's socket threads to hang, and since everything goes through it, all your network traffic is now dead. Chromium uses a similar model of routing all I/O through specific places, so it's vulnerable to similar sorts of problems (though AFAIK it has never failed this badly in production). Basically, this could have happened at any point if a production service run by someone like Google, Facebook, Cloudflare etc managed to trigger a sufficiently bad bug in a modern browser's network stack. If not for the centralized I/O this would just hang a single browser process, which isn't as impactful since Firefox and Chrome both split content out into many processes. It would also make it more obvious which server(s) are responsible since only certain tabs would be dying. FWIW, as far as I know this centralized I/O model was popularized by Google, not Mozilla.
- jacquesm 5y ago
  It is because: (1) I did not explicitly enable this (2) The telemetry setting seems to have re-enabled itself on some update (3) I don't want any services from Mozilla, I want a browser (4) It worked until it blew up revealing that in fact, I suddenly did have service dependencies And finally, the reason I use FireFox is exactly your last sentence, so to see that they are slipping this in under the radar is a pretty good reason to drop FF altogether, it looks as if they fail to understand the difference between shipping software and getting me hooked on some service that I am not even aware of existing. And on top of that re-enabling their telemetry when it was explicitly disabled. That really takes the cake.
  - kevingadd 5y ago
    This is caused by HTTP3. The only way for you to have never experienced this issue would be if Mozilla never integrated HTTP3, which isn't going to happen because the major service providers like Google and Cloudflare are adopting it. Telemetry and updates happening to use HTTP3 merely pushed the issue to the foreground, it would have happened eventually and possibly in a way that got fixed more slowly because it was happening intermittently and only to visitors of obscure websites. Switching to a browser without telemetry and automatic updates won't protect you from this kind of network stack bug.