[LU-9938] unload_modules() should fail on remote node errors or memory leaks

Sat, 29 Jan 2022 10:47:42 +0000

The current memory leak detection scheme in TF is not very effective. Comparing single node runs with what I see in AT I think we are failing to fail when a memory leak occurs on a remote node.

unload_modules() {
        wait_exit_ST client # bug 12845                                                   

	$LUSTRE_RMMOD ldiskfs || return 2

	if $LOAD_MODULES_REMOTE; then
                local list=$(comma_list $(remote_nodes_list))
	        if [ -n "$list" ]; then
                        echo "unloading modules on: '$list'"
	        	do_rpc_nodes "$list" $LUSTRE_RMMOD ldiskfs
                        do_rpc_nodes "$list" check_mem_leak
                fi
        fi

        local sbin_mount=$(readlink -f /sbin)/mount.lustre
	if grep -qe "$sbin_mount " /proc/mounts; then
                umount $sbin_mount || true
                [ -s $sbin_mount ] && ! grep -q "STUB MARK" $sbin_mount ||
                        rm -f $sbin_mount
        fi

        check_mem_leak || return 254
        ...

Furthermore, cleanupall() does not check the return value of unload_modules() so it may be that we are missing memory leaks when we cleanup at the end of most test scripts.

Whamcloud Community JIRA

[LU-9938] unload_modules() should fail on remote node errors or memory leaks