When I use Tortoise, particularly the repo browser, with a remote
Linux-based server via svn+ssh, it's working fine but as I use it,
more and more sshd and svnserve processes accumulate and stay running.
My service has cut off my ssh access several times due to alarms
resulting from these ghost sshd and svnserve processes. They take
about a half hour to die on their own.
While the ghost processes are running at the server end, I can verify
that I absolutely positively have no TCP connections open to the
server, having finished whatever I was doing with Tortoise.
I have taken ethernet traces. I can't see anything in the payloads
(encrypted) but what I do see is that TortoisePlink talks back and
forth with the server a bunch then at some point it sends a 36 byte
message to the server and immediately follows that with a TCP reset
(RST) packet. This is an "unclean" close. The normal TCP connection
closure involves an exchange of FIN packets between the peers.
Presumably, the sshd and child svnserve processes on the other end are
sitting there waiting for something like a normal TCP connection
closure. To me, they should die when receiving a reset packet but they
I searched the subversion users and developers lists and found several
threads, all relating to Tortoise, where the same symptoms were
reported. In those cases the subversion gods blamed it on Tortoise.
The sysadmin at my hosting company also says the problem is unique to
Tortoise. I don't normally use anecdotal evidence like that, but in
this case, the winds all point to unclean closing of the connection
when a transaction between tortoise and svnserve (via SSH) is
For reference, using the command line svn tool with ssh config'd to
use TortoisePlink doesn't seem to leave the ghost processes behind,
but I need to test some more with that.
To unsubscribe from this discussion, e-mail: [users-unsubscribe_at_tortoisesvn.tigris.org].
Received on 2009-09-15 16:50:42 CEST