On Mon, 2008-03-17 at 01:37 +0000, Martin Pool wrote:
> As an overview of all these bugs:
...
> To check this, I ran 'ssh localhost sleep 1h', and kill-9'd the client.
> This left the sleep process running and as far as I can tell with
> strace, it did not get a sighup. So it looks like just killing the
> client is not enough to be sure the server will shut down.
what state was the socket left in?
> One approach would be for the server to just try the lock once, rather
> than blocking on it, and then come back to the client to say if it was
> busy. This fixes a few things: the client gets told that it's waiting
> and can potentially show progress, and the server should rarely take a
> lock when there's no longer a client wanting it.
It will reduce the window, but it won't remove it entirely (networks for
the win). On the other hand, it will cause more ssh server branch-open
operations at the far end, and will likely skew lock obtaining to folk
with faster links.
On Mon, 2008-03-17 at 01:37 +0000, Martin Pool wrote:
> As an overview of all these bugs:
...
> To check this, I ran 'ssh localhost sleep 1h', and kill-9'd the client.
> This left the sleep process running and as far as I can tell with
> strace, it did not get a sighup. So it looks like just killing the
> client is not enough to be sure the server will shut down.
what state was the socket left in?
> One approach would be for the server to just try the lock once, rather
> than blocking on it, and then come back to the client to say if it was
> busy. This fixes a few things: the client gets told that it's waiting
> and can potentially show progress, and the server should rarely take a
> lock when there's no longer a client wanting it.
It will reduce the window, but it won't remove it entirely (networks for
the win). On the other hand, it will cause more ssh server branch-open
operations at the far end, and will likely skew lock obtaining to folk
with faster links.
-Rob www.robertcolli ns.net/ keys.txt>.
--
GPG key available at: <http://