Hi.. I've found that under high load, the ssh-agent process dies. The high load environment: Home computer: Fedora 10, openssh-5.1p1 Gateway computer: Fedora 11, openssh-5.2p1 Cluster head node: MacOSX 10.6, openssh-5.2p1 6 Child nodes: MacOSX 10.4, 10.5, 10.6, various versions The program: On the cluster head node, run "ssh node### blastall -p blastp ..." hundreds of times (80 jobs simultaneously, using 'make -j 80'), where node### is a random child node. The crash: The ssh-agent process running on the home computer dies after a few minutes of this, causing the saved key to be lost, and new ssh connections require the passphrase for my private key. Running 'ssh-agent tcsh' and 'ssh-add' fixes the problem at least for the children of one terminal window. Logging out and back in restores full functionality. Observations: Running 40 jobs simultaneously works. I should probably be using an ssh-agent on the cluster head node, or some other method that doesn't use ssh-agent. I didn't see any rate-limiting code in ssh-agent for openssh-5.1p1, so this is probably a bug, not a feature. Thanks!
Are you sure that the ssh-agent running on the home computer (Fedora 10) is a real ssh-agent and not for example gnome-keyring-daemon emulating ssh-agent? echo $SSH_AUTH_SOCK will tell you more.
(In reply to comment #1) > Are you sure that the ssh-agent running on the home computer (Fedora > 10) is a real ssh-agent and not for example gnome-keyring-daemon > emulating ssh-agent? > > echo $SSH_AUTH_SOCK > will tell you more. Ah, I see; it is the gnome-keyring-daemon. I'll go file this bug with them. Thanks!
Apparently it's already been fixed in a later version. I'm using gnome-keyring-2.24.1-1.fc10.i386.rpm, and the bug was reported in https://bugzilla.gnome.org/show_bug.cgi?id=580068 , and apparently fixed around August 2009.
Mass move of bugs RESOLVED->CLOSED following the release of openssh-5.5p1