Linux 5.5.x: IPC/file-system operations seems to get sporadically freezed

[vladimiry]

Summary: there is no reliable solution for the issue at the moment except for reverting to the 5.4 kernel.

I just tried to toggle the Start minimized to tray option on the "General" app settings block and got the Invocation timeout of calling "patchBaseConfig" method on "electron-mail:ipcMain-api" channel with 25000ms timeout error message which I expected to see since I visually saw that the form remains blocked which means the flag toggling process is still in progress.

The issue occurred after the computer got out from the suspend mode.

Toggling the Start minimized to tray flag simply sends the signal to the main process, then the main process updates the config.json and sends the updated version back.

So far the issue has been noticed only on Arch Linux based systems (see also #250 and #252). The app instability issues started to happen a few days ago (on Feb 8-12), so seems to be caused by recent system update since the app is a static thing (was not updated).

Toggling the Start minimized to tray flag is just one case I noticed but it looks like the inter-process communication is affected as a whole thing (see #250 and #252). The wild guess is that the file system access involvement is a required but not limited condition for an issue occurring (access to FS gets freezed and so IPC request doesn't respond within the timeout).

CC @lachlantula, @gourdcaptain, @srappan. Not sure if this issue is similar to issue you reported just recently, since I use @electron v8 based build but you were using released build which is @electron v7 based.

[vladimiry]

I also noticed that after the recent system updates the nm-applet gets segfault-ed quite often especially after the suspend mode with the error message:

systemd-coredump[26719]: Process 22518 (nm-applet) of user 1000 dumped core.
                                                 #14 0x0000563bf888d1d6 n/a (nm-applet)
                                                 #16 0x0000563bf888d2fe n/a (nm-applet)
kernel: nm-applet[22518]: segfault at 20 ip 00007fcc90d51ca0 sp 00007ffca0d56530 error 6 in libglib-2.0.so.0.6200.4[7fcc90cdd000+81000]

The 5.5.2-1 kernel was used. Maybe it's worth to revert to the LTS/5.4 kernel.

[vladimiry]

In all the 4 reported cases of weird app behavior on Arch-based systems the 5.5.x Linux kernel was used (5.5.2 and 5.5.3).

[vladimiry]

I was able to find in the syslog mentioning libraries like libpthread.so.0 and libc.so.6 related to the node/elecron/nm-applet processes segfaults stacktraces. So I guess the following issues might be similar to what some Linux/Arch users face using ElectronMail with kernel 5.5 (especially the app based on @electron v7, like v4.2.2 release):

• Analysis server hangs with Linux 5.5 and IntelliJ based IDEs dart-lang/sdk#40589 (log, analysis)
• Multiple issues running Flutter on Linux 5.5 flutter/flutter#49185
• Flutter Debug: Syncing files to device flutter/flutter#22229

That's a Dart stuff of curse, but another blind guess is that @nodejs / @electron might experience similar epoll-related issues. That would explain the guess posted above in the first message:

The wild guess is that the file system access involvement is a required but not limited condition for an issue occurring (access to FS gets freezed and so IPC request doesn't respond within the timeout).

[vladimiry]

I would be interested to see if someone who recently reported EM v4.2.2 use issues/freezes/timeouts on Arch + 5.5.x kernel could run journalctl -p 3 -r --since "2020-02-12 00:00:00" --until "2020-02-12 23:59:59"-like command and located there some electron / electron-mail / node-like processes segfaults (set the date range you faced the issue within).

[lachlanshoesmith]

I found the following:

Feb 13 22:43:37 okcomputer systemd-coredump[82098]: Process 82000 (electron-mail) of user 1000 dumped core.

There were several instances of crashes like that.

[vladimiry]

Might be related kernel issue https://bugzilla.kernel.org/show_bug.cgi?id=205933 (picked from flutter issues tracker).

[vladimiry]

Also possibly related task electron/electron#22291. It would explain why ElectronMail build based on Electron v8 works for @lachlantula, @gourdcaptain, @srappan. The lib https://www.archlinux.org/packages/core/x86_64/glibc/ landed to Arch repo on Feb 11 / Last Updated: 2020-02-11 09:07 UTC which is within above-mentioned time slot:

The app instability issues started to happen a few days ago (on Feb 8-12), so seems to be caused by recent system update since the app is a static thing (was not updated).

[vladimiry]

Electron v8 based release https://github.com/vladimiry/ElectronMail/releases/tag/v4.3.0 got published. So closing the issue for now as there has been no freezing issues reported if Electron v8 based build was used.

[vladimiry]

Referencing relevant messages:

• Desktop: Joplin Freezing During Syncing and Decrypting On Linux Kernel 5.5+ laurent22/joplin#2518 (comment)
• Electron apps freeze on linux 5.x kernel electron/electron#21415 (comment)

Thanks @bedwardly-down for such a thorough investigation.

[bedwardly-down]

You're quite welcome. The little info you gave me went a long way and to be fair, that investigation was a bit frustrating but fun. i think I'd like to contribute more in that regard. I also use Protonmail as my main email service, so I'll try your client out and see what i can do here. ☺️

[vladimiry]

Looks like 5.7 kernel doesn't experience the issue. So using either 5.4 LTS or 5.7 is a solution.

[bedwardly-down]

5.6.13 up seems to have fixed the issue for Joplin (the other Electron app that had this issue). There were two epoll related commits submitted by the maintainers that created the original problem commit. Hopefully, no one will mess with this further in the kernel.