====== Shell executor ======

Known issue with job cancellation.
Shell executor is using ''bash --login user'' and for every ''script:'' section another shell (bash) is spawned.
In result proces tree looks like:

{{:git:gitlab:runner:pasted:20220105-175808.png}}

Default bash behavior is to ignore signals when waiting for child.

  If bash is waiting for a command to complete and receives a signal for which a trap has been set, 
  the trap will not be executed until the command completes. When bash is waiting for an asynchronous
  command via the wait builtin, the reception of a signal for which a trap has been set will cause
  the wait builtin to return immediately with an exit status greater than 128, immediately after which 
  the trap is executed. 

The same issue described here:
  * [[https://gitlab.com/gitlab-org/gitlab-runner/-/issues/3101|Canceling jobs in shell executor doesnt work]]

Some references:
  * [[https://unix.stackexchange.com/questions/146756/forward-sigterm-to-child-in-bash|Forward SIGTERM to child in Bash]]
  * [[https://gitlab.com/gitlab-org/gitlab-runner/-/issues/3031|Gracefully halt cancelled jobs]]

Solution:
  * Using bash wrapper script plus replace current shell with new process using ''exec''
<file yaml .gitlab-ci.yml>
build:
  stage: build
  script:
    - exec tools/ci/launch_job.sh make all
    - exec tools/ci/launch_job.sh make deploy
</file>

<file bash tools/ci/launch_job.sh>
#!/bin/bash
#_term() {
#    echo "$0 caught SIGTERM signal!"
#    trap - SIGTERM
#    kill -TERM $child $$
#}

launch_job() {
    PARENT=$1
    shift
    "$@" &
    CHILD=$!
    trap "trap - SIGTERM; kill -TERM $CHILD $$" SIGTERM

    while sleep 1; do
	if [ ! -e /proc/$PARENT ]; then 
	    echo "Parent PID $PARENT disappears, terminating child $CHILD"
	    kill -TERM $CHILD
	    exit
	fi

	if [ ! -e /proc/$CHILD ]; then
	    wait $CHILD
	    trap - SIGTERM
	    exit $?
	fi
    done
}

#trap _term SIGTERM
#launch_job $$ "$@" &
#echo "$0 Waiting for $! to finish"
#wait $!

launch_job $PPID "$@"
</file>