Fix race condition on deployment creation (#979) (#982)

* Run the provisioner scripts as the base AMI user

* Stop processes before removing directory

Author: agarciamontoro

deployment/terraform/agent.go

@@ -63,7 +63,10 @@ func (t *Terraform) configureAndRunAgents(extAgent *ssh.ExtAgent) error {
 	}
 
 	commands := []string{
-		"rm -rf mattermost-load-test-ng*",
+		// Stop and remove are joined with ; so the rm still runs on the first deploy when the services don't exist yet.
+		// The services must be stopped before removing the directory to avoid a race condition with ltbrowserapi
+		// writing to browser/node_modules, which causes rm to fail with "Directory not empty".
+		"sudo systemctl stop ltapi ltbrowserapi 2>/dev/null; sudo rm -rf mattermost-load-test-ng*",
 		"tar xzf tmp.tar.gz",
 		"mv mattermost-load-test-ng* mattermost-load-test-ng",
 		"rm tmp.tar.gz",

deployment/terraform/assets/user_data.sh.tpl

@@ -12,14 +12,11 @@ PROVEOF
 chmod +x /tmp/provisioner.sh
 
 # Cloud-init runs user_data as root with a minimal environment.
-# Set HOME to the AMI user's home directory since provisioner scripts
-# (e.g. nvm) install to $HOME and other code expects files there.
-export HOME="/home/${ami_user}"
-export USER="${ami_user}"
-
+# Use runuser -l to execute the provisioner as the AMI user with a full
+# login shell so that tools like nvm (sourced via ~/.bashrc) are available.
 cd /tmp
 rc=0
-/tmp/provisioner.sh || rc=$?
+runuser -l "${ami_user}" -c /tmp/provisioner.sh || rc=$?
 if [ $rc -eq 0 ]; then
   touch /var/lib/cloud/instance/provisioning-done
 else