Can't kill background process started in setup_file

[evadeflow]

After reading through issue #205 and issue-205.bats, I'm able to start a nodeJS-based process in the background from setup_file() and it does not hang my tests. However, I can't figure out how to kill this process in teardown_file.

To Reproduce
Steps to reproduce the behavior:

It's probably easiest to see what I'm trying to do by applying this diff:

diff --git a/test/fixtures/bats/issue-205.bats b/test/fixtures/bats/issue-205.bats
index ddc826a..8382ad8 100644
--- a/test/fixtures/bats/issue-205.bats
+++ b/test/fixtures/bats/issue-205.bats
@@ -6,7 +6,7 @@ function bgfunc {
   close_non_std_fds
   get_open_fds
   echo "${FUNCNAME[1]} fds after: (${open_fds[*]})" >>"${LOG_FILE}"
-  sleep 10
+  sleep 2222
   echo "bgfunc done"
   return 0
 }
@@ -92,6 +92,11 @@ function otherfunc {

 setup_file() { # see issue #530
   bgfunc &
+  SFPID=$!
+}
+
+teardown_file() {
+  kill ${SFPID}
 }

 @test "min bg" {

and then running ./bin/bats test/bats.bats. When the test suite completes, run ps -ef | grep "sleep 2222" | grep -v grep.

Expected behavior
I expected to see at most one instance of sleep 2222 running, but there are two:

$ ps -ef | grep "sleep 2222" | grep -v grep
bosch     5883    31  0 15:04 pts/3    00:00:00 sleep 2222
bosch     5887  5872  0 15:04 pts/3    00:00:00 sleep 2222

Environment (please complete the following information):

• Bats version 1.11.0 (master@a751f3d)
• operating system: WSL2 Linux, Ubuntu 20.04)
• bash --version: [e.g. 5.1]: GNU bash, version 5.0.17(1)-release (x86_64-pc-linux-gnu)
• Install method: git submodule

Additional context
I'm pretty unfamiliar with BATS, so I may be doing something stupid/silly. Attempting to store the PID in setup_file and then killing it in teardown_file seemed reasonable, but it doesn't work, so... how does one start a service in background and then kill it once all the tests have completed? I just want to spin up my app, send it some requests using curl, and verify that the responses are what's expected.

[martin-schulze-vireso]

Probably FD3 staying open again. Maybe the same issue as in #530