Support dumping gevent stacks from uWSGI workers

To help debug cases where a worker process "hangs", add a SIGCONT signal handler
which dumps both the native and greenlet threads.

viahtml/wsgi.py

@@ -1,11 +1,24 @@
 """The application providing a WSGI entry-point."""
 
+import faulthandler
 import os
+import signal
+import sys
 
 # Our job here is to leave this `application` attribute laying around as
 # it's what uWSGI expects to find.
 from viahtml.app import Application
 
+
+def _dump_stacks(*args):  # pragma: no cover
+    # Print both Python thread and gevent "greenlet" stats and backtraces.
+    #
+    # See https://www.gevent.org/monitoring.html#visibility
+    import gevent
+
+    gevent.util.print_run_info()
+
+
 application = Application()
 
 if os.environ.get("SENTRY_DSN"):  # pragma: no cover
@@ -17,3 +30,11 @@
     # pylint: disable=redefined-variable-type
     sentry_sdk.init(dsn=os.environ["SENTRY_DSN"])
     application = SentryWsgiMiddleware(application)
+
+# Add a way to dump stacks so we can see what a uWSGI worker is doing if it
+# hangs.
+#
+# We use `SIGCONT` for this because this handler isn't used for anything
+# important, and uWSGI defines its own `SIGUSR1` and `SIGUSR2` handlers after
+# this code runs.
+signal.signal(signal.SIGCONT, _dump_stacks)