Mercurial > hg
changeset 30891:b1b36c6499c2
commandserver: handle backlog before exiting
Previously, when a chg server is exiting, it does not handle connected
clients so clients may get ECONNRESET and crash:
1. client connect() # success
2. server shouldexit = True and exit
3. client recv() # ECONNRESET
d7875bfbfccb makes this race condition easier to reproduce if a lot of short
chg commands are started in parallel.
This patch fixes the above issue by unlinking the socket path to stop
queuing new connections and processing all pending connections before exit.
| author | Jun Wu <quark@fb.com> |
|---|---|
| date | Wed, 08 Feb 2017 14:45:30 -0800 |
| parents | a95fc01aaffe |
| children | 441705506d24 |
| files | mercurial/commandserver.py |
| diffstat | 1 files changed, 13 insertions(+), 1 deletions(-) [+] |
line wrap: on
line diff
--- a/mercurial/commandserver.py Wed Feb 08 14:37:38 2017 -0800 +++ b/mercurial/commandserver.py Wed Feb 08 14:45:30 2017 -0800 @@ -477,11 +477,23 @@ self._cleanup() def _mainloop(self): + exiting = False h = self._servicehandler - while not h.shouldexit(): + while True: + if not exiting and h.shouldexit(): + # clients can no longer connect() to the domain socket, so + # we stop queuing new requests. + # for requests that are queued (connect()-ed, but haven't been + # accept()-ed), handle them before exit. otherwise, clients + # waiting for recv() will receive ECONNRESET. + self._unlinksocket() + exiting = True try: ready = select.select([self._sock], [], [], h.pollinterval)[0] if not ready: + # only exit if we completed all queued requests + if exiting: + break continue conn, _addr = self._sock.accept() except (select.error, socket.error) as inst: