Skip to content

andrewhamon/pcmd

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

pcmd

pcmd is a small utility to make SSH ProxyCommand more powerful. It might be useful any time you have a ProxyCommand that involves a lengthy or expensive set-up or tear-down procedure.

Rationale

The primary use case of pcmd is to wrap a ProxyCommand that might need to perform non-trivial set-up or tear-down operations. For example, one might have a ProxyCommand that provisions and deletes a cloud VPS for disposable development environments, accessible via ssh my.temporary.host. For such a use case, it is essential to a) not duplicate work and b) ensure that resources get cleaned up. Without pcmd it can be difficult to reliably meet those requirements.

Here are some of the challenges I faced when implementing on-demand VPS provisioning in a ProxyCommand:

  1. Tear-down. Some versions and configurations of SSH, such as on macOS, send a hangup signal as soon as the SSH connection is closed. This prevents any cleanup from occurring. That is bad because then you get billed for a VPS that you wanted to be disposable!
  2. Logging. It's nice to be able to see logs to stderr while waiting for the VPS to come up. Once the SSH connection is closed, though, it's very annoying if logs to stderr continue to get dumped into your terminal. That is what would happen if you resolved issue 1) by trapping and ignoring signals.
  3. Connection sharing and concurrency control. If you initiate multiple SSH connections, you probably want to re-use an existing VPS when possible. You might think: doesn't SSH let you configure this? You bet it does! You can set ControlMaster auto to make SSH opportunistically re-uses existing connections, skipping the ProxyCommand on subsequent SSH invocations. There's a small problem, though: what if the first connection is still in the "set-up" phase and hasn't established a connection? It turns out that SSH doesn't do any locking, and, in that case, it executes the ProxyCommand twice. One will win and become the master connection, the other will log a warning that the master already exists, and you will be paying for double the compute you thought you were.

pcmd solves all of those problems:

  1. pcmd shields the underlying ProxyCommand from interrupt and hangup signals. When a signal is received, pcmd closes stdio and starts a grace period timer (default 5 minutes). During this grace period, the ProxyCommand can perform any tear-down it needs. If the command hasn't exited by the end of the grace period, pcmd sends a kill signal.
  2. pcmd copies stderr of the ProxyCommand to a log file as well as to the terminal. That way, while waiting for the connection, you can see any relevant logs. Once the connection is closed, pcmd continues copying stderr of the command to the log file but stops copying stderr to the terminal. That way, you don't get annoying logs dumped into the terminal when you move on to something else.
  3. If instructed with the -lock flag, pcmd ensures only one copy of your ProxyCommand runs at once. If are sharing connections using ControlMaster, you can additionally specify -wait-for-master. If pcmd doesn't have a lock, it blocks waiting for the ControlMaster to come up. As a nicety, while waiting, it also tails the log file mentioned in 2) to stderr so you can monitor the progress of a lengthy set-up.

Compatibility and dependencies

pcmd is cross-compiled to many targets, but has only been tested on macOS and Linux (Ubuntu 18.04). Additionally, pcmd requires the tail and, depending on configuration, ssh commands to be available. These are likely already installed on your system.

Installation

Pre-compiled binaries

pcmd is available for download from the releases page. You can also use the following snippet to install the latest release of pcmd, adjusting the values accordingly for your environment:

# Possible options: darwin,linux,freebsd,openbsd,netbsd
# darwin == macOS
OS=darwin

# Possible options: amd64,386,arm,arm64(linux only)
ARCH=amd64

TARGET=pcmd-$OS-$ARCH

# Download and unzip
curl -OL https://github.com/andrewhamon/pcmd/releases/latest/download/$TARGET.zip
unzip $TARGET.zip

# Copy to somewhere on your $PATH
# replace ~/bin with something appropriate for your environment
cp $TARGET/pcmd ~/bin

Go get

If you have Go installed, you can also install pcmd using go get:

go get github.com/andrewhamon/pcmd

Examples

The easiest way to get started is to wrap your original ProxyCommand with pcmd in your SSH config. For example, if your SSH config looks like this:

Host some-host
ProxyCommand original-proxy-command --original-arg foobar

You can prefix original-proxy-command and its arguments with pcmd:

Host some-host
ProxyCommand pcmd original-proxy-command --original-arg foobar

This will continue to proxy as before but also ensures original-proxy-command has adequate time after the connection closes to perform cleanup.

Complete example: on-demand VPS with DigitalOcean

Included in this repo is an example bash script that uses doctl to create a VPS on the fly and proxy to it. The following is is the resulting SSH config, configured for connection sharing with ControlMaster. To use, you will need to ensure that the on-demand-proxy script is downloaded and available in your $PATH and that you have installed doctl.

Host ondemand.dev
User root
ProxyCommand pcmd --wait-for-master -r %r -h %h -p %p on-demand-proxy %h
ControlMaster auto
ControlPath ~/.ssh/%r@%h:%p.sock

# Keep the connection alive for 5 minutes after the last connection is closed.
# This lets you quickly re-connect. This is not the same as the -grace-period
# flag.
ControlPersist 300

# Even when reviving a snapshot, DigitalOcean seems to generate a new key, which
# will then cause scary warnings.
StrictHostKeyChecking no
UserKnownHostsFile=/dev/null

pcmd in action

Configurable grace period

You can configure the grace period (default 5 minutes) with the -grace-period flag. For example, to set the grace period to 10 minutes:

pcmd -grace-period 600 original-proxy-command --original-arg foobar

The grace period begins only once the SSH connection is closed.

Preventing concurrent ProxyCommand invocations

If you add the -lock flag, pcmd can ensure that only one copy of original-proxy-command runs at a time. To do so, pcmd needs to know the SSH remote user and host, which it uses to form a unique key for locking. For example:

pcmd -lock -r %r -h %h original-proxy-command --original-arg foobar

%r and %h are expanded by SSH automatically. See TOKENS in SSH_CONFIG(5) for more details.

Connection sharing

SSH provides a native mechanism for connection sharing via the ControlMaster and ControlPath configuration options, which allows you to share a single connection for multiple SSH sessions (see SSH_CONFIG(5) for more details). One issue with the native mechanism, however, is that SSH doesn't do any concurrency control. That means if you have a lengthy set-up in your proxy command, and then try to establish two concurrent connections, SSH does not block one connection waiting for a master. Instead, it runs both ProxyCommands at the same time.

pcmd can help with this, by doing two things:

  • Establishing a lock (using the flock system call) to ensure only one version of your ProxyCommand is running at a time.
  • Waits for the ControlMaster to come up, if the lock can not be acquired. pcmd checks the status of the control master using ssh user@host -O check.

To set up connection sharing, add the -wait-for-master flag, along with the -r and -h flags. For example:

Host some-host
ProxyCommand pcmd -wait-for-master -r %r -h %h original-proxy-command --original-arg foobar
ControlMaster auto
ControlPath ~/.ssh/ssh-%r@%h:%p

# Setting ControlPersist to a non-zero value ensures that SSH keeps the master
# connection running in the background, rather than blocking the first SSH
# invocation until all child connections also complete. If you want to keep your
# connection alive in the background for longer, set this to a higher value (in
# number of seconds)
ControlPersist 1

The above config will allow only a single master connection, and make any subsequent connections wait for the master to come up before continuing.