Required lots of rework since the GOBChannel appeared to be reading
ahead of the data it actually needed. Now using more low-level IO
to send the clientId over to the agent instead.
Was a concurrency issue int he Write method that retained the
slice p that was passed in, making concurrency issues much more likely
with unbuffered channels.
Wrote ChannelReadWriter that simulates a connection inmemory.
This is used by the agentserver test for testing the initialization. The
first test is already working.
Alternative solution would be to run all initialization code in go routines to make it independent of initialization order but having a defined initialization order is much cleaner.
The server then sends it the protocol version and the client connection info describing whether an agent was found or not. This improves usability for users.
With the --raw option it bypasses this query parameter and wsproxy then works in the old way as a simple stdio-websocket connector. It then still works with converge server but can also be used for simple websocket troubleshooting.
* removed the Async utility
* now using Ping message to webclient for keep alive instaed of actual content
* added remote shell to AgentInfo
* retry of connections to the agent
* better logging for SynchronizeStreams
the agent if PPROF_PORT is set.
Fixed issue with converge server not cleaning up goroutines because of blocking channel. Made sure to create channels with > 1 size everywhere it can be done. The blocking behavior of a default channel size is mostly in the way.
Known issue: Killing the SSH client will lead to the server side process not being terminated and some goroutines still running in the agent. This would require additional investigation to solve. The remote processes are still being cleaned up ok (at least on linux) when the agent exits.
This should not be an issue at all since the agent is a short-lived process and when running in a containerized environment with containers running on demand the cleanup will definitely work.
Protocol version is now checked when the agent connects to the converge server.
Next up: sending connection metadata and username password from server to agent and sending environment information back to the server. This means then that the side channel will only be used for expiry time messages and session type with the client id passed in so the converge server can than correlate the results back to the correct channel.
Now hijacking the ssh connection setup in the listener to exchange some information before passing the connection on to the SSH server.
Next step is to do the full exchange of required information and to make it easy some simple Read and Write methods with timeouts are needed that use gob.
