kickstart/lib/kickstart-deploy

#!/bin/bash -e

[[ "$1" == "--sudo" ]] && prefix="sudo " && shift
[[ "$1" == "--password" ]] && shift && prefix="echo \"$1\" | sudo -S true && sudo " && shift

target=$1
[ "$target" ] || echo "Please, provide an ssh target"
shift

[ "$DEBUG" ] && debug="-x"

kickstart compile "$@"

copy_command=$(
cat <<CMD
{
  deploy_folder=\$(mktemp -d -t kickstart.XXXXX)
  mkdir -p "\$deploy_folder"
  cd "\$deploy_folder"
  tar xz
  echo "\$deploy_folder"
}
CMD
)
deploy_folder=$(tar chz -C compile . | ssh $target "bash -c '$copy_command'")

execute_command=$(
cat <<CMD
{
  cd $deploy_folder
  exit_status=0
  $prefix bash $debug install.sh || exit_status=\$?
  rm -rf "$deploy_folder"
  exit \$exit_status
}
CMD
)
ssh -t $target "bash -c '$execute_command'"
Introduce kickstart deploy 2014-02-20 16:19:43 +00:00			`#!/bin/bash -e`

Support running with sudo 2014-02-21 18:59:46 +00:00			`[[ "$1" == "--sudo" ]] && prefix="sudo " && shift`
Fix prompt when executing the deployment with the password `sudo -S` reads the password from stdin. The way it was it was executing the deploy script with the password as the stdin of the pipe. Now, when password is passed as an argument we aquire a sudo session executing a command that always work and then prefixing the installation with sudo. 2015-09-27 02:35:05 +00:00			`[[ "$1" == "--password" ]] && shift && prefix="echo \"$1\" \| sudo -S true && sudo " && shift`
Support running with sudo 2014-02-21 18:59:46 +00:00
Introduce kickstart deploy 2014-02-20 16:19:43 +00:00			`target=$1`
			`[ "$target" ] \|\| echo "Please, provide an ssh target"`
			`shift`

Add support to debug 2014-02-22 00:14:39 +00:00			`[ "$DEBUG" ] && debug="-x"`
Support running with sudo 2014-02-21 18:59:46 +00:00
Allows deployment scripts to be interactive Previously, because we were piping the compiled script blob over ssh to decompress over there without creating an intermediated bundled file, we were allocating the stdin of the ssh session to have the read from the pipe. This blocked the allocation of the /dev/tty over the session and prevented users to interact with the script. We would like some of the interactions to work, like prompting for a confirmation or typing a password. To fix this, the deployment process was split in two steps. The first step still receives the bundled file over a pipe, decompresses it on a temprary folder, and outputs the location of the deployment folder. We grab this output to feed on the second step, which has a script to go to the deployment folder and start the execution. Now, roles can make use of read, as well as sudo with password. 2015-09-27 01:30:55 +00:00			`kickstart compile "$@"`

			`copy_command=$(`
Introduce kickstart deploy 2014-02-20 16:19:43 +00:00			`cat <<CMD`
Wraps deploy script execution on to prevent partial execution if it is terminated before finished transfering 2015-09-26 03:51:19 +00:00			`{`
Renames variable on deployment script to not confuse with deployment host target 2015-09-26 03:31:38 +00:00			`deploy_folder=\$(mktemp -d -t kickstart.XXXXX)`
			`mkdir -p "\$deploy_folder"`
			`cd "\$deploy_folder"`
Introduce kickstart deploy 2014-02-20 16:19:43 +00:00			`tar xz`
Allows deployment scripts to be interactive Previously, because we were piping the compiled script blob over ssh to decompress over there without creating an intermediated bundled file, we were allocating the stdin of the ssh session to have the read from the pipe. This blocked the allocation of the /dev/tty over the session and prevented users to interact with the script. We would like some of the interactions to work, like prompting for a confirmation or typing a password. To fix this, the deployment process was split in two steps. The first step still receives the bundled file over a pipe, decompresses it on a temprary folder, and outputs the location of the deployment folder. We grab this output to feed on the second step, which has a script to go to the deployment folder and start the execution. Now, roles can make use of read, as well as sudo with password. 2015-09-27 01:30:55 +00:00			`echo "\$deploy_folder"`
			`}`
			`CMD`
			`)`
			`deploy_folder=$(tar chz -C compile . \| ssh $target "bash -c '$copy_command'")`

			`execute_command=$(`
			`cat <<CMD`
			`{`
			`cd $deploy_folder`
Reports the deployment execution status 2015-06-26 15:13:41 +00:00			`exit_status=0`
			`$prefix bash $debug install.sh \|\| exit_status=\$?`
Allows deployment scripts to be interactive Previously, because we were piping the compiled script blob over ssh to decompress over there without creating an intermediated bundled file, we were allocating the stdin of the ssh session to have the read from the pipe. This blocked the allocation of the /dev/tty over the session and prevented users to interact with the script. We would like some of the interactions to work, like prompting for a confirmation or typing a password. To fix this, the deployment process was split in two steps. The first step still receives the bundled file over a pipe, decompresses it on a temprary folder, and outputs the location of the deployment folder. We grab this output to feed on the second step, which has a script to go to the deployment folder and start the execution. Now, roles can make use of read, as well as sudo with password. 2015-09-27 01:30:55 +00:00			`rm -rf "$deploy_folder"`
Reports the deployment execution status 2015-06-26 15:13:41 +00:00			`exit \$exit_status`
Wraps deploy script execution on to prevent partial execution if it is terminated before finished transfering 2015-09-26 03:51:19 +00:00			`}`
Introduce kickstart deploy 2014-02-20 16:19:43 +00:00			`CMD`
			`)`
Allocates a tty when deploying to allow full fledge interaction To have some interactions on the script we need to allocate a tty. This adds a pseudo-tty to the ssh session. This enables `read -s` on deployment scripts. 2015-09-27 02:44:25 +00:00			`ssh -t $target "bash -c '$execute_command'"`