SIGPIPE handler causing infinite loop and ~100% CPU usage on journald restart

[whage]

in core/uwsgi.c a handler for a SIGPIPE signal is registered:

if (!uwsgi.ignore_sigpipe) {
    signal(SIGPIPE, (void *) &warn_pipe);
}

We have a systemd-based uWSGI setup which means that uWSGI is started by systemd and the default behavior in this case is to pipe its stdout and stderr to systemd-journald.

When journald is restarted it raises a SIGPIPE. A similar issue can be found here: cybozu-go/well#13

Since uWSGI tries to handle a SIGPIPE by logging it, the log message ends up in a broken pipe which in turn raises a SIGPIPE and we are stuck in an infinite loop.

The ignore-sigpipe config option (as seen in the if statement above) can be used to avoid this but it might be better to just ignore the SIGPIPE signal by default.

[unbit]

Hi, thanks a lot for reporting (eventually feel free to make a pull request in systemd-related uwsgi-docs). Unfortunately not handling it could lead to unreported errors in other plugins that do not have a solid i/o error checking