Commit 9ba80de0 authored by Michal Berger's avatar Michal Berger Committed by Ben Walker
Browse files

test/nvme: Don't wait for spdk_stub0 if stub dies prematurely 



If stub terminates right after execution, e.g. due to lack of system
resources (requested number of cpu cores to run on, right amount of
memory, etc.) start_stub() would end up blocking forever since there
wouldn't be any entity around that would mknod spdk_stub0 for it to
break the loop.

Avoid the above scenario by checking if $stubpid is still visible
under procfs and return if it goes missing.

To make sure kill_stub() is still called to clean up after start_stub()
declare proper trap prior the call to start_stub().

Additionally, avoid potential stderr noise in case kill|wait are told
to act on a PID that's not visible in the ns anymore.

Change-Id: Ief41200c57957f84b4f96a54baabc8da1f27dd43
Signed-off-by: default avatarMichal Berger <michallinuxstuff@gmail.com>
Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/482653


Tested-by: default avatarSPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: default avatarJim Harris <james.r.harris@intel.com>
Reviewed-by: default avatarBen Walker <benjamin.walker@intel.com>
parent 4b1dbc94

test/common/autotest_common.sh

+14 −3
Original line number Diff line number Diff line
@@ -575,7 +575,7 @@ function rbd_cleanup() { 
	fi

}



function start_stub() {

function _start_stub() {

	# Disable ASLR for multi-process testing.  SPDK does support using DPDK multi-process,

	# but ASLR can still be unreliable in some cases.

	# We will reenable it again after multi-process testing is complete in kill_stub().
@@ -586,14 +586,25 @@ function start_stub() { 
	stubpid=$!

	echo Waiting for stub to ready for secondary processes...

	while ! [ -e /var/run/spdk_stub0 ]; do

		# If stub dies while we wait, bail

		[[ -e /proc/$stubpid ]] || return 1

		sleep 1s

	done

	echo done.

}



function start_stub() {

	if ! _start_stub "$@"; then

		echo "stub failed" >&2

		return 1

	fi

}



function kill_stub() {

	if [[ -e /proc/$stubpid ]]; then

		kill $1 $stubpid

		wait $stubpid

	fi 2>/dev/null || :

	rm -f /var/run/spdk_stub0

	# Re-enable ASLR now that we are done with multi-process testing

	# Note: "1" enables ASLR w/o randomizing data segments, "2" adds data segment

test/nvme/nvme.sh

+1 −1
Original line number Diff line number Diff line
@@ -102,8 +102,8 @@ if [ $(uname) = Linux ]; then 
fi



if [ $(uname) = Linux ]; then

	start_stub "-s 4096 -i 0 -m 0xE"

	trap "kill_stub -9; exit 1" SIGINT SIGTERM EXIT

	start_stub "-s 4096 -i 0 -m 0xE"

fi



run_test "nvme_reset" $testdir/reset/reset -q 64 -w write -s 4096 -t 5

test/nvme/spdk_nvme_cli.sh

+1 −1
Original line number Diff line number Diff line
@@ -18,8 +18,8 @@ if [ ! -d $spdk_nvme_cli ]; then 
fi



if [ $(uname) = Linux ]; then

	start_stub "-s 2048 -i 0 -m 0xF"

	trap "kill_stub; exit 1" SIGINT SIGTERM EXIT

	start_stub "-s 2048 -i 0 -m 0xF"

fi



# Build against the version of SPDK under test