Issues communicating with the agent: accept: Too many connections

[fholzer]

Recently we (my colleagues and I) started seeing issues with the agent when starting lots of ssh connections in a small period of time, e.g. when running Ansible on lots of servers. This manifests and authentication errors shown in the Ansible output. We use Pageant on our Windows machines, then ssh-pageant in cygwin. Drilling deeper into this I started ssh-pageant in one shell in debug mode, and set the env variable SSH_AUTH_SOCKET to the path of this agent in another shell. In this other shell I connect to our automation server (with agent forwarding) and from there I ran Ansible. I observed the same issues, but this time I saw that ssh-pageant running in debug mode outputs a lot of errors:

ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections
ssh-pageant: accept: Too many connections

Searching for this I found this: https://github.com/cuviper/ssh-pageant/blob/master/main.c#L246
But I am not sure what we can do to remedy this situation.

Edit: Just to clarify, even with this debug setup where I am running ssh-pageant in a different shell than where I am actually using the socket from... some connection from Ansible are successful, some are not.

[fholzer]

Wrt. the actual content of the error message "too many connections" In ansible we use the "-f" for parameter, so it will connect to multiple hosts concurrently. We usually use a fork value in the range of 8 (default) and 16. This doesn't seem too high to us. Also in the past we've never had issues like this. It is also worth noting that we experience this issue not only when using Ansible, but also when using ssh directly.

We are not sure what exactly changed on our system to suddenly trigger this behavior. Could be any number of things:

• windows update
• some antivirus/DPA agent that corporate potentially rolled out
• update of packages (ssh, others?) on our automation server
• ssh package of cygwin

[cuviper]

You could add a debug statement to print the new fd number, and again on close to see that they're not hanging open for some reason. I also wonder if some system/cygwin change is not recycling "low" fd numbers for some reason...