bitbake: cooker/process: Fix signal handling lockups

If a parser process is terminated while holding a write lock, then it
will lead to a deadlock (see
https://docs.python.org/3/library/multiprocessing.html#multiprocessing.Process.terminate).

With SIGTERM, we don't want to terminate holding the lock. We also don't
want a SIGINT to cause a partial write to the event stream.

I tried using signal masks to avoid this but it doesn't work, see
https://bugs.python.org/issue47139

Instead, add a signal handler and catch the calls around the critical section.
We also need a thread lock to ensure other threads in the same process don't
handle the signal until all the threads are not in the lock.

(Bitbake rev: a40efaa5556a188dfe46c8d060adde37dc400dcd)

Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This commit is contained in:
Richard Purdie
2022-03-29 15:17:10 +01:00
parent c0ff6c75ee
commit 928bcb10a4
2 changed files with 46 additions and 4 deletions

View File

@@ -2017,6 +2017,22 @@ class Parser(multiprocessing.Process):
self.context = bb.utils.get_context().copy()
self.handlers = bb.event.get_class_handlers().copy()
self.profile = profile
self.queue_signals = False
self.signal_received = []
self.signal_threadlock = threading.Lock()
def catch_sig(self, signum, frame):
if self.queue_signals:
self.signal_received.append(signum)
else:
self.handle_sig(signum, frame)
def handle_sig(self, signum, frame):
if signum == signal.SIGTERM:
signal.signal(signal.SIGTERM, signal.SIG_DFL)
os.kill(os.getpid(), signal.SIGTERM)
elif signum == signal.SIGINT:
signal.default_int_handler(signum, frame)
def run(self):
@@ -2036,9 +2052,17 @@ class Parser(multiprocessing.Process):
prof.dump_stats(logfile)
def realrun(self):
signal.signal(signal.SIGTERM, signal.SIG_DFL)
# Signal handling here is hard. We must not terminate any process or thread holding the write
# lock for the event stream as it will not be released, ever, and things will hang.
# Python handles signals in the main thread/process but they can be raised from any thread and
# we want to defer processing of any SIGTERM/SIGINT signal until we're outside the critical section
# and don't hold the lock (see server/process.py). We therefore always catch the signals (so any
# new thread should also do so) and we defer handling but we handle with the local thread lock
# held (a threading lock, not a multiprocessing one) so that no other thread in the process
# can be in the critical section.
signal.signal(signal.SIGTERM, self.catch_sig)
signal.signal(signal.SIGHUP, signal.SIG_DFL)
signal.signal(signal.SIGINT, signal.SIG_IGN)
signal.signal(signal.SIGINT, self.catch_sig)
bb.utils.set_process_name(multiprocessing.current_process().name)
multiprocessing.util.Finalize(None, bb.codeparser.parser_cache_save, exitpriority=1)
multiprocessing.util.Finalize(None, bb.fetch.fetcher_parse_save, exitpriority=1)

View File

@@ -20,6 +20,7 @@ import os
import sys
import time
import select
import signal
import socket
import subprocess
import errno
@@ -737,11 +738,28 @@ class ConnectionWriter(object):
# Why bb.event needs this I have no idea
self.event = self
def send(self, obj):
obj = multiprocessing.reduction.ForkingPickler.dumps(obj)
def _send(self, obj):
with self.wlock:
self.writer.send_bytes(obj)
def send(self, obj):
obj = multiprocessing.reduction.ForkingPickler.dumps(obj)
# See notes/code in CookerParser
# We must not terminate holding this lock else processes will hang.
# For SIGTERM, raising afterwards avoids this.
# For SIGINT, we don't want to have written partial data to the pipe.
# pthread_sigmask block/unblock would be nice but doesn't work, https://bugs.python.org/issue47139
process = multiprocessing.current_process()
if process and hasattr(process, "queue_signals"):
with process.signal_threadlock:
process.queue_signals = True
self._send(obj)
process.queue_signals = False
for sig in process.signal_received.pop():
process.handle_sig(sig, None)
else:
self._send(obj)
def fileno(self):
return self.writer.fileno()