Firefox IPC has problems

[kworr]

Firefox was built in synth. When running it emits messages like:

[Child 7804, Chrome_ChildThread] WARNING: pipe error: Incorrect argument: file /construction/www/firefox/firefox-57.0/ipc/chromium/src/chrome/common/ipc_channel_posix.cc, line 709

There are a lot of those mesages. While they are displayed some tabs may become unresponsive, slow or even can fail completely. This can be easily fixed with:

browser.tabs.remote.autostart.2=false

[jbeich]

Do you still see IPC warnings as of Firefox 60? Can you reproduce with a fresh profile? Upstream continues to polish IPC each release. Maybe try Firefox 61 as well.

[kworr]

Well with 60 situation changed. browser.tabs.remote.autostart.2 is no more so I'm using browser.tabs.remote.autostart.

When starting firefox without the tweak (multiprocess mode) it starts up fine but crashes in a few minutes:

May 25 10:30:55 probe kernel: pid 1267 (firefox), uid 1001: exited on signal 11 (core dumped)
May 25 10:30:55 probe kernel: proc_unstop: lwp 1267/18 not sleeping
May 25 10:30:56 probe kernel: pid 1269 (firefox), uid 1001: exited on signal 4 (core dumped)
May 25 10:30:56 probe kernel: proc_unstop: lwp 1269/3 not sleeping
May 25 10:30:56 probe kernel: proc_unstop: lwp 1269/13 not sleeping

Tested a few times and result is the same. However if I disable multiprocess mode everything is stable and browser survives for more than a day.

This might not be IPC right now, just a coincidence. There's no more errors about IPC in the logs. Also the browser crashes the same way if I build it with clang, only with clang disabling multiprocessing dosn't change anything.

PS: I'm not writing this to urge someone to fix it but rather to share information in case it could be useful to anybody.

[tuxillo]

Thanks @kworr for the pointers. I'm trying to build firefox with the option DEBUG. If we can get it to build then maybe we can get meaningful backtraces.

[jbeich]

@tuxillo, maybe check if DragonFly is affected by FreeBSD bug 181741 (sample).

[tuxillo]

@jbeich here's the test results in latest master (HEAD) https://leaf.dragonflybsd.org/~tuxillo/archive/temp/t_issue813.log

[kworr]

For people just passing by: right now issue can be mitigated by:

export MOZ_FORCE_DISABLE_E10S = 1