You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Uses of SocketServer.BaseServer.shutdown have a race #46555
Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.
Show more details
GitHub fields:
assignee=Noneclosed_at=<Date2010-04-25.22:28:59.672>created_at=<Date2008-03-16.15:15:55.330>labels= ['type-bug', 'library']
title='Uses of SocketServer.BaseServer.shutdown have a race'updated_at=<Date2010-04-25.22:28:59.671>user='https://bugs.python.org/jyasskin'
With the code as it stands, calls to shutdown that happen before
serve_forever enters its loop will deadlock, and there's no simple way
for the user to avoid this. The attached patch prevents the deadlock and
allows multiple serve_forever..shutdown cycles, but it's pretty
complicated. I could make it a lot simpler by making shutdown permanent:
any later serve_forever calls would return immediately.
A third choice would be to add a .serve_in_thread function that returns
a token that can be used to shut down exactly that loop, instead of
putting .shutdown() on the server. Any opinions?
With the code as it stands, calls to shutdown that happen before
serve_forever enters its loop will deadlock, and there's no simple way
for the user to avoid this. The attached patch prevents the deadlock and
allows multiple serve_forever..shutdown cycles, but it's pretty
complicated. I could make it a lot simpler by making shutdown permanent:
any later serve_forever calls would return immediately.
Never thought of using the SocketServer taht way, wouldn't the person
doing this bunch of shutdown()s and serve_forever()s be better off using
handle_request() on a loop instead ?
A third choice would be to add a .serve_in_thread function that returns
a token that can be used to shut down exactly that loop, instead of
putting .shutdown() on the server. Any opinions?
I don't think I understand this part, what loop do you refer to ?
After a bit of investigation, this issue seems to be exactly why test_httpservers sometimes hangs.
The patch looks complicated to me, though; I don't think we really have to support the multiple shutdowns case.
Here is a simpler patch. It also fixes the wrong use of a lock instead of an event in test_httpservers. With this patch, test_httpservers runs forever without freezing.
By the way, getting rid of poll_interval for a file descriptor is easy under Unix, but wouldn't work under Windows (where select() only takes sockets, not arbitrary file descriptors).
Activity
jyasskin commentedon Mar 16, 2008
With the code as it stands, calls to shutdown that happen before
serve_forever enters its loop will deadlock, and there's no simple way
for the user to avoid this. The attached patch prevents the deadlock and
allows multiple serve_forever..shutdown cycles, but it's pretty
complicated. I could make it a lot simpler by making shutdown permanent:
any later serve_forever calls would return immediately.
A third choice would be to add a .serve_in_thread function that returns
a token that can be used to shut down exactly that loop, instead of
putting .shutdown() on the server. Any opinions?
zanella commentedon May 7, 2008
Never thought of using the SocketServer taht way, wouldn't the person
doing this bunch of shutdown()s and serve_forever()s be better off using
handle_request() on a loop instead ?
I don't think I understand this part, what loop do you refer to ?
pitrou commentedon Apr 17, 2010
After a bit of investigation, this issue seems to be exactly why test_httpservers sometimes hangs.
The patch looks complicated to me, though; I don't think we really have to support the multiple shutdowns case.
pitrou commentedon Apr 17, 2010
Here is a simpler patch. It also fixes the wrong use of a lock instead of an event in test_httpservers. With this patch, test_httpservers runs forever without freezing.
pitrou commentedon Apr 19, 2010
By the way, getting rid of poll_interval for a file descriptor is easy under Unix, but wouldn't work under Windows (where select() only takes sockets, not arbitrary file descriptors).
pitrou commentedon Apr 25, 2010
Fixed with a test in r80484 (trunk), r80486 (2.6), r80487 (py3k), r80491 (3.1).