Skip to contentSkip to conversation

nvme: add nvmf reset/disconnect during traffic test

PATCH In review Sagi Grimberg 4 participants 6 messages Jun 30, 2022 – Jul 12, 2022 NVMe

lore
Share & tools
lore

Sagi Grimberg

author maintainer
#0 lore raw

PATCH v2 1/1 nvme: add nvmf reset/disconnect during traffic test

Test traffic controller reset and disconnect during traffic.

---
Changes from v1:
- fix copyright
- change subsys name
- fix test cleanup warnings
- don't log device names
- fix shellcheck

 tests/nvme/040     | 59 ++++++++++++++++++++++++++++++++++++++++++++++
 tests/nvme/040.out |  6 +++++
 tests/nvme/rc      |  8 +++++++
 3 files changed, 73 insertions(+)
 create mode 100755 tests/nvme/040
 create mode 100644 tests/nvme/040.out
tests/nvme/040 new file +59 -0
new file mode 100755
index 000000000000..d259784698d9
@@ -0,0 +1,59 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-3.0+
+# Copyright (C) 2022 Sagi Grimberg <[email protected]>
+#
+# Test nvme fabrics controller reset/disconnect/reconnect operation during I/O
+# This test is somewhat similar to test 032 but for fabrics controllers.
+
+. tests/nvme/rc
+
+DESCRIPTION="test nvme fabrics controller reset/disconnect operation during I/O"
+
+requires() {
+	_nvme_requires
+	_have_loop
+	_have_fio
+	_require_nvme_trtype_is_fabrics
+}
+
+test() {
+	local subsys="blktests-subsystem-1"
+	local port
+	local loop_dev
+	local nvmedev
+
+	echo "Running ${TEST_NAME}"
+
+	_setup_nvmet
+	truncate -s 1G "$TMPDIR/img"
+	loop_dev="$(losetup -f --show "$TMPDIR/img")"
+
+	port="$(_create_nvmet_port "${nvme_trtype}")"
+	_create_nvmet_subsystem "${subsys}" "${loop_dev}"
+	_add_nvmet_subsys_to_port "${port}" "${subsys}"
+	_nvme_connect_subsys "${nvme_trtype}" "${subsys}"
+	udevadm settle
+	nvmedev=$(_find_nvme_dev "${subsys}")
+
+	# start fio job
+	echo "starting background fio"
+	_run_fio_rand_io --filename="${nvmedev}n1" --size=1g \
+		--group_reporting --ramp_time=5  &> /dev/null &
+	sleep 5
+
+	# do reset/remove operation
+	echo "resetting controller"
+	_nvme_reset_ctrl ${nvmedev}
+	sleep 1
+	echo "deleting controller"
+	_nvme_delete_ctrl ${nvmedev}
+
+	echo "stopping background fio"
+	{ kill $!; wait; } &> /dev/null
+
+	_remove_nvmet_subsystem_from_port "${port}" "${subsys}"
+	_remove_nvmet_subsystem "${subsys}"
+	_remove_nvmet_port "${port}"
+
+	echo "Test complete"
+}
tests/nvme/040.out new file +6 -0
new file mode 100644
index 000000000000..50d02d4e3cd0
@@ -0,0 +1,6 @@
+Running nvme/040
+starting background fio
+resetting controller
+deleting controller
+stopping background fio
+Test complete
tests/nvme/rc +8 -0
index 4bebbc762cbb..dfea7f3a0da4 100644
@@ -447,3 +447,11 @@ _nvme_disable_err_inject()
         echo 0 > /sys/kernel/debug/"$1"/fault_inject/probability
         echo 0 > /sys/kernel/debug/"$1"/fault_inject/times
 }
+
+_nvme_reset_ctrl() {
+	echo 1 > /sys/class/nvme/"$1"/reset_controller
+}
+
+_nvme_delete_ctrl() {
+	echo 1 > /sys/class/nvme/"$1"/delete_controller
+}
Signature · 1 line
2.34.1
Reviewed-by on v2 1/1
#1 lore raw

Looks good. I’ve tested it on my config.

Quoted 117 lines · Sagi Grimberg

Test traffic controller reset and disconnect during traffic.

---
Changes from v1:
- fix copyright
- change subsys name
- fix test cleanup warnings
- don't log device names
- fix shellcheck

tests/nvme/040     | 59 ++++++++++++++++++++++++++++++++++++++++++++++
tests/nvme/040.out |  6 +++++
tests/nvme/rc      |  8 +++++++
3 files changed, 73 insertions(+)

create mode 100755 tests/nvme/040
create mode 100644 tests/nvme/040.out

tests/nvme/040 new file +59 -0
new file mode 100755
index 000000000000..d259784698d9
@@ -0,0 +1,59 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-3.0+
+# Copyright (C) 2022 Sagi Grimberg <[email protected]>
+#
+# Test nvme fabrics controller reset/disconnect/reconnect operation during I/O
+# This test is somewhat similar to test 032 but for fabrics controllers.
+
+. tests/nvme/rc
+
+DESCRIPTION="test nvme fabrics controller reset/disconnect operation during I/O"
+
+requires() {
+	_nvme_requires
+	_have_loop
+	_have_fio
+	_require_nvme_trtype_is_fabrics
+}
+
+test() {
+	local subsys="blktests-subsystem-1"
+	local port
+	local loop_dev
+	local nvmedev
+
+	echo "Running ${TEST_NAME}"
+
+	_setup_nvmet
+	truncate -s 1G "$TMPDIR/img"
+	loop_dev="$(losetup -f --show "$TMPDIR/img")"
+
+	port="$(_create_nvmet_port "${nvme_trtype}")"
+	_create_nvmet_subsystem "${subsys}" "${loop_dev}"
+	_add_nvmet_subsys_to_port "${port}" "${subsys}"
+	_nvme_connect_subsys "${nvme_trtype}" "${subsys}"
+	udevadm settle
+	nvmedev=$(_find_nvme_dev "${subsys}")
+
+	# start fio job
+	echo "starting background fio"
+	_run_fio_rand_io --filename="${nvmedev}n1" --size=1g \
+		--group_reporting --ramp_time=5  &> /dev/null &
+	sleep 5
+
+	# do reset/remove operation
+	echo "resetting controller"
+	_nvme_reset_ctrl ${nvmedev}
+	sleep 1
+	echo "deleting controller"
+	_nvme_delete_ctrl ${nvmedev}
+
+	echo "stopping background fio"
+	{ kill $!; wait; } &> /dev/null
+
+	_remove_nvmet_subsystem_from_port "${port}" "${subsys}"
+	_remove_nvmet_subsystem "${subsys}"
+	_remove_nvmet_port "${port}"
+
+	echo "Test complete"
+}
tests/nvme/040.out new file +6 -0
new file mode 100644
index 000000000000..50d02d4e3cd0
@@ -0,0 +1,6 @@
+Running nvme/040
+starting background fio
+resetting controller
+deleting controller
+stopping background fio
+Test complete
tests/nvme/rc +8 -0
index 4bebbc762cbb..dfea7f3a0da4 100644
@@ -447,3 +447,11 @@ _nvme_disable_err_inject()
        echo 0 > /sys/kernel/debug/"$1"/fault_inject/probability
        echo 0 > /sys/kernel/debug/"$1"/fault_inject/times
 }
+
+_nvme_reset_ctrl() {
+	echo 1 > /sys/class/nvme/"$1"/reset_controller
+}
+
+_nvme_delete_ctrl() {
+	echo 1 > /sys/class/nvme/"$1"/delete_controller
+}
Signature · 1 line
2.34.1
#2 lore raw
In reply to Sagi Grimberg

O

Review

Quoted 10 lines
(quoted diff) +10 -0
+
+	# do reset/remove operation
+	echo "resetting controller"
+	_nvme_reset_ctrl ${nvmedev}
+	sleep 1
+	echo "deleting controller"
+	_nvme_delete_ctrl ${nvmedev}
+
+	echo "stopping background fio"
+	{ kill $!; wait; } &> /dev/null

do we really need to kill the fio process explicitly ? I think graceful termination of the traffic application should be a part of this test without explicitly killing the process.

I ran the test with commenting above line it runs fine :-

blktests (master) # git diff

tests/nvme/040 +1 -1
index d259784..0dce128 100755
@@ -49,7 +49,7 @@ test() {
         _nvme_delete_ctrl ${nvmedev}
 
         echo "stopping background fio"
-       { kill $!; wait; } &> /dev/null
+#{ kill $!; wait; } &> /dev/null
 
         _remove_nvmet_subsystem_from_port "${port}" "${subsys}"
         _remove_nvmet_subsystem "${subsys}"

blktests (master) # ./check nvme/040
nvme/040 (test nvme fabrics controller reset/disconnect operation during I/O) [passed]

     runtime  8.119s  ...  8.084s

blktests (master) #

-ck

#3 lore raw
Quoted 35 lines · Chaitanya Kulkarni

O

Review

Quoted 10 lines
(quoted diff) +10 -0
+
+	# do reset/remove operation
+	echo "resetting controller"
+	_nvme_reset_ctrl ${nvmedev}
+	sleep 1
+	echo "deleting controller"
+	_nvme_delete_ctrl ${nvmedev}
+
+	echo "stopping background fio"
+	{ kill $!; wait; } &> /dev/null

do we really need to kill the fio process explicitly ? I think graceful termination of the traffic application should be a part of this test without explicitly killing the process.

I ran the test with commenting above line it runs fine :-

blktests (master) # git diff

tests/nvme/040 +1 -1
index d259784..0dce128 100755
@@ -49,7 +49,7 @@ test() {
         _nvme_delete_ctrl ${nvmedev}
 
         echo "stopping background fio"
-       { kill $!; wait; } &> /dev/null
+#{ kill $!; wait; } &> /dev/null
 
         _remove_nvmet_subsystem_from_port "${port}" "${subsys}"
         _remove_nvmet_subsystem "${subsys}"

blktests (master) # ./check nvme/040
nvme/040 (test nvme fabrics controller reset/disconnect operation during I/O) [passed]

Quoted 4 lines
     runtime  8.119s  ...  8.084s

blktests (master) #

-ck

Sagi, what do you think on this comment?

To keep this test case consistent with nvme/032, it would be the better to keep the kill and wait. But we may not need to stick to it if it does better testing without the kill and wait.

Signature · 1 line
Shin'ichiro Kawasaki
author maintainer
#4 lore raw
Quoted 39 lines
Quoted 39 lines

O

Review

Quoted 10 lines
(quoted diff) +10 -0
+
+	# do reset/remove operation
+	echo "resetting controller"
+	_nvme_reset_ctrl ${nvmedev}
+	sleep 1
+	echo "deleting controller"
+	_nvme_delete_ctrl ${nvmedev}
+
+	echo "stopping background fio"
+	{ kill $!; wait; } &> /dev/null

do we really need to kill the fio process explicitly ? I think graceful termination of the traffic application should be a part of this test without explicitly killing the process.

I ran the test with commenting above line it runs fine :-

blktests (master) # git diff

tests/nvme/040 +1 -1
index d259784..0dce128 100755
@@ -49,7 +49,7 @@ test() {
          _nvme_delete_ctrl ${nvmedev}
 
          echo "stopping background fio"
-       { kill $!; wait; } &> /dev/null
+#{ kill $!; wait; } &> /dev/null
 
          _remove_nvmet_subsystem_from_port "${port}" "${subsys}"
          _remove_nvmet_subsystem "${subsys}"

blktests (master) # ./check nvme/040
nvme/040 (test nvme fabrics controller reset/disconnect operation during I/O) [passed]

      runtime  8.119s  ...  8.084s

blktests (master) #

Quoted 7 lines
Quoted 1 line

-ck

Sagi, what do you think on this comment?

To keep this test case consistent with nvme/032, it would be the better to keep the kill and wait. But we may not need to stick to it if it does better testing without the kill and wait.

I don't particularly care. Can you remove it when applying?

#5 lore raw
Quoted 43 lines · Sagi Grimberg
Quoted 43 lines
Quoted 36 lines

O

Review

Quoted 10 lines
(quoted diff) +10 -0
+
+	# do reset/remove operation
+	echo "resetting controller"
+	_nvme_reset_ctrl ${nvmedev}
+	sleep 1
+	echo "deleting controller"
+	_nvme_delete_ctrl ${nvmedev}
+
+	echo "stopping background fio"
+	{ kill $!; wait; } &> /dev/null

do we really need to kill the fio process explicitly ? I think graceful termination of the traffic application should be a part of this test without explicitly killing the process.

I ran the test with commenting above line it runs fine :-

blktests (master) # git diff

tests/nvme/040 +1 -1
index d259784..0dce128 100755
@@ -49,7 +49,7 @@ test() {
          _nvme_delete_ctrl ${nvmedev}
 
          echo "stopping background fio"
-       { kill $!; wait; } &> /dev/null
+#{ kill $!; wait; } &> /dev/null
 
          _remove_nvmet_subsystem_from_port "${port}" "${subsys}"
          _remove_nvmet_subsystem "${subsys}"

blktests (master) # ./check nvme/040
nvme/040 (test nvme fabrics controller reset/disconnect operation during I/O) [passed]

Quoted 4 lines
      runtime  8.119s  ...  8.084s

blktests (master) #

-ck

Sagi, what do you think on this comment?

Quoted 5 lines
Quoted 3 lines

To keep this test case consistent with nvme/032, it would be the better to keep the kill and wait. But we may not need to stick to it if it does better testing without the kill and wait.

I don't particularly care. Can you remove it when applying?

Sure, I've applied with the edit.

Signature · 1 line
Shin'ichiro Kawasaki