test/vhost: live migration test case 3

To be run only manually as test pool does not
have DUTs connected in pairs.

Change-Id: If3ab3e671156b583adb35d5b23ee90003a6de732
Signed-off-by: Karol Latecki <karol.latecki@intel.com>
Reviewed-on: https://review.gerrithub.io/400988
Reviewed-by: Pawel Wodkowski <pawelx.wodkowski@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Tested-by: SPDK Automated Test System <sys_sgsw@intel.com>
Reviewed-by: Daniel Verkamp <daniel.verkamp@intel.com>
This commit is contained in:
Karol Latecki 2018-02-21 15:09:16 +01:00 committed by Daniel Verkamp
parent 0d1c3aefc3
commit 8ebaed2eb5
6 changed files with 341 additions and 10 deletions

View File

@ -220,8 +220,9 @@ function spdk_vhost_kill()
rm $vhost_pid_file
rc=1
else
#check vhost return code, activate trap on error
wait $vhost_pid
while kill -0 $vhost_pid; do
echo "."
done
fi
elif /bin/kill -0 $vhost_pid; then
error "vhost NOT killed - you need to kill it manually"
@ -294,9 +295,12 @@ function vm_create_ssh_config()
echo " UserKnownHostsFile=/dev/null"
echo " StrictHostKeyChecking=no"
echo " User root"
echo " ControlPath=$VM_BASE_DIR/%r@%h:%p.ssh"
echo " ControlPath=/tmp/%r@%h:%p.ssh"
echo ""
) > $ssh_config
# Control path created at /tmp because of live migration test case 3.
# In case of using sshfs share for the test - control path cannot be
# on share because remote server will fail on ssh commands.
fi
}

View File

@ -0,0 +1,14 @@
vhost_0_reactor_mask=0x1
vhost_0_master_core=0
vhost_1_reactor_mask=0x1
vhost_1_master_core=0
VM_0_qemu_mask=0x1
VM_0_qemu_numa_node=0
VM_1_qemu_mask=0x1
VM_1_qemu_numa_node=0
VM_2_qemu_mask=0x1
VM_2_qemu_numa_node=0

View File

@ -0,0 +1,20 @@
[global]
blocksize=4k-512k
iodepth=128
ioengine=libaio
filename=
group_reporting
thread
numjobs=1
direct=1
do_verify=1
verify=md5
verify_fatal=1
verify_dump=1
verify_backlog=8
[randwrite]
rw=randwrite
runtime=15
time_based
stonewall

View File

@ -0,0 +1,206 @@
source $SPDK_BUILD_DIR/test/nvmf/common.sh
source $BASE_DIR/autotest.config
MGMT_TARGET_IP="10.102.17.181"
MGMT_INITIATOR_IP="10.102.17.180"
RDMA_TARGET_IP="10.0.0.1"
RDMA_INITIATOR_IP="10.0.0.2"
incoming_vm=1
target_vm=2
incoming_vm_ctrlr=naa.VhostScsi0.$incoming_vm
target_vm_ctrlr=naa.VhostScsi0.$target_vm
share_dir=$TEST_DIR/share
job_file=$BASE_DIR/migration-tc3.job
function ssh_remote()
{
local ssh_cmd="ssh -i $SPDK_VHOST_SSH_KEY_FILE \
-o UserKnownHostsFile=/dev/null \
-o StrictHostKeyChecking=no -o ControlMaster=auto \
root@$1"
shift
$ssh_cmd "$@"
}
function wait_for_remote()
{
local timeout=40
set +x
while [[ ! -f $share_dir/DONE ]]; do
echo -n "."
if (( timeout-- == 0 )); then
error "timeout while waiting for FIO!"
fi
sleep 1
done
set -x
rm -f $share_dir/DONE
}
function check_rdma_connection()
{
local nic_name=$(ip -4 -o addr show to $RDMA_TARGET_IP up | cut -d' ' -f2)
if [[ -z $nic_name ]]; then
error "There is no NIC with IP address $RDMA_TARGET_IP configured"
fi
if ! ls /sys/class/infiniband/*/device/net/$nic_name &> /dev/null; then
error "$nic_name with IP $RDMA_TARGET_IP is not a RDMA capable NIC"
fi
}
function host1_cleanup_nvmf()
{
notice "Shutting down nvmf_tgt on local server"
if [[ ! -z "$1" ]]; then
pkill --signal $1 -F $nvmf_dir/nvmf_tgt.pid
else
pkill -F $nvmf_dir/nvmf_tgt.pid
fi
rm -f $nvmf_dir/nvmf_tgt.pid
}
function host1_cleanup_vhost()
{
trap 'host1_cleanup_nvmf SIGKILL; error_exit "${FUNCNAME}" "${LINENO}"' INT ERR EXIT
notice "Shutting down VM $incoming_vm"
vm_kill $incoming_vm
notice "Removing bdev & controller from vhost on local server"
$rpc_0 delete_bdev Nvme0n1
$rpc_0 remove_vhost_controller $incoming_vm_ctrlr
notice "Shutting down vhost app"
spdk_vhost_kill 0
host1_cleanup_nvmf
}
function host1_start_nvmf()
{
nvmf_dir="$TEST_DIR/nvmf_tgt"
rpc_nvmf="python $SPDK_BUILD_DIR/scripts/rpc.py -s $nvmf_dir/nvmf_rpc.sock"
notice "Starting nvmf_tgt instance on local server"
mkdir -p $nvmf_dir
rm -rf $nvmf_dir/*
cp $SPDK_BUILD_DIR/test/nvmf/nvmf.conf $nvmf_dir/nvmf.conf
$SPDK_BUILD_DIR/scripts/gen_nvme.sh >> $nvmf_dir/nvmf.conf
trap 'host1_cleanup_nvmf SIGKILL; error_exit "${FUNCNAME}" "${LINENO}"' INT ERR EXIT
$SPDK_BUILD_DIR/app/nvmf_tgt/nvmf_tgt -s 512 -c $nvmf_dir/nvmf.conf -r $nvmf_dir/nvmf_rpc.sock &
nvmf_tgt_pid=$!
echo $nvmf_tgt_pid > $nvmf_dir/nvmf_tgt.pid
waitforlisten "$nvmf_tgt_pid" "$nvmf_dir/nvmf_rpc.sock"
$rpc_nvmf construct_nvmf_subsystem nqn.2018-02.io.spdk:cnode1 \
"trtype:RDMA traddr:$RDMA_TARGET_IP trsvcid:4420" "" -a -s SPDK01 -n Nvme0n1
}
function host1_start_vhost()
{
rpc_0="python $SPDK_BUILD_DIR/scripts/rpc.py -s $(get_vhost_dir 0)/rpc.sock"
notice "Starting vhost0 instance on local server"
trap 'host1_cleanup_vhost; error_exit "${FUNCNAME}" "${LINENO}"' INT ERR EXIT
spdk_vhost_run --conf-path=$BASE_DIR --vhost-num=0
$rpc_0 construct_nvme_bdev -b Nvme0 -t rdma -f ipv4 -a $RDMA_TARGET_IP -s 4420 -n "nqn.2018-02.io.spdk:cnode1"
$rpc_0 construct_vhost_scsi_controller $incoming_vm_ctrlr
$rpc_0 add_vhost_scsi_lun $incoming_vm_ctrlr 0 Nvme0n1
vm_setup --os="$share_dir/migration.qcow2" --force=$incoming_vm --disk-type=spdk_vhost_scsi --disks=VhostScsi0 \
--migrate-to=$target_vm --memory=512 --queue_num=1
# TODO: Fix loop calculating cpu_num in common.sh
# We need -smp 1 and -queue_num 1 for this test to work, and this loop
# in some cases calculates wrong cpu_num.
sed -i "s#smp 2#smp 1#g" $VM_BASE_DIR/$incoming_vm/run.sh
vm_run $incoming_vm
vm_wait_for_boot 300 $incoming_vm
}
function cleanup_share()
{
set +e
notice "Cleaning up share directory on remote and local server"
ssh_remote $MGMT_INITIATOR_IP "umount $VM_BASE_DIR"
ssh_remote $MGMT_INITIATOR_IP "umount $share_dir; rm -f $share_dir/*"
rm -f $share_dir/migration.qcow2
rm -f $share_dir/spdk.tar.gz
set -e
}
function host_1_create_share()
{
notice "Creating share directory on local server to re-use on remote"
mkdir -p $share_dir
mkdir -p $VM_BASE_DIR # This dir would've been created later but we need it now
rm -rf $share_dir/spdk.tar.gz $share_dir/spdk || true
cp $os_image $share_dir/migration.qcow2
tar --exclude="*.o"--exclude="*.d" --exclude="*.git" -C $SPDK_BUILD_DIR -zcf $share_dir/spdk.tar.gz .
}
function host_2_create_share()
{
# Copy & compile the sources for later use on remote server.
ssh_remote $MGMT_INITIATOR_IP "uname -a"
ssh_remote $MGMT_INITIATOR_IP "mkdir -p $share_dir"
ssh_remote $MGMT_INITIATOR_IP "mkdir -p $VM_BASE_DIR"
ssh_remote $MGMT_INITIATOR_IP "sshfs -o ssh_command=\"ssh -i $SPDK_VHOST_SSH_KEY_FILE\" root@$MGMT_TARGET_IP:$VM_BASE_DIR $VM_BASE_DIR"
ssh_remote $MGMT_INITIATOR_IP "sshfs -o ssh_command=\"ssh -i $SPDK_VHOST_SSH_KEY_FILE\" root@$MGMT_TARGET_IP:$share_dir $share_dir"
ssh_remote $MGMT_INITIATOR_IP "mkdir -p $share_dir/spdk"
ssh_remote $MGMT_INITIATOR_IP "tar -zxf $share_dir/spdk.tar.gz -C $share_dir/spdk --strip-components=1"
ssh_remote $MGMT_INITIATOR_IP "cd $share_dir/spdk; make clean; ./configure --with-rdma --enable-debug; make -j40"
}
function host_2_start_vhost()
{
ssh_remote $MGMT_INITIATOR_IP "nohup $share_dir/spdk/test/vhost/migration/migration.sh --test-cases=3b --work-dir=$TEST_DIR --os=$share_dir/migration.qcow2 &>$share_dir/output.log &"
notice "Waiting for remote to be done with vhost & VM setup..."
wait_for_remote
}
function setup_share()
{
trap 'cleanup_share; error_exit "${FUNCNAME}" "${LINENO}"' INT ERR EXIT
host_1_create_share
host_2_create_share
}
function migration_tc3()
{
check_rdma_connection
setup_share
host1_start_nvmf
host1_start_vhost
host_2_start_vhost
# Do migration
notice "Starting fio on local VM"
vm_check_scsi_location $incoming_vm
run_fio $fio_bin --job-file="$job_file" --local --vm="${incoming_vm}$(printf ':/dev/%s' $SCSI_DISK)"
sleep 5
if ! is_fio_running $incoming_vm; then
vh_ssh $incoming_vm "cat /root/$(basename ${job_file}).out"
error "Fio not running on local VM before starting migration!"
fi
vm_migrate $incoming_vm $RDMA_INITIATOR_IP
sleep 1
# Verify migration on remote host and clean up vhost
ssh_remote $MGMT_INITIATOR_IP "pkill -CONT -F $TEST_DIR/tc3b.pid"
notice "Waiting for remote to finish FIO on VM and clean up..."
wait_for_remote
# Clean up local stuff
host1_cleanup_vhost
cleanup_share
}
migration_tc3

View File

@ -0,0 +1,80 @@
# Set -m option is needed to be able to use "suspend" command
# as we are usin non-interactive session to connect to remote.
# Without -m it would be not possible to suspend the process.
set -m
source $BASE_DIR/autotest.config
RDMA_TARGET_IP="10.0.0.1"
incoming_vm=1
target_vm=2
target_vm_ctrl=naa.VhostScsi0.$target_vm
rpc="python $SPDK_BUILD_DIR/scripts/rpc.py -s $(get_vhost_dir 1)/rpc.sock"
share_dir=$TEST_DIR/share
function host_2_cleanup_vhost()
{
notice "Shutting down VM $target_vm"
vm_kill $target_vm
notice "Removing bdev & controller from vhost 1 on remote server"
$rpc delete_bdev Nvme0n1
$rpc remove_vhost_controller $target_vm_ctrl
notice "Shutting down vhost app"
spdk_vhost_kill 1
sleep 1
}
function host_2_start_vhost()
{
echo "BASE DIR $TEST_DIR"
vhost_work_dir=$TEST_DIR/vhost1
mkdir -p $vhost_work_dir
rm -f $vhost_work_dir/*
notice "Starting vhost 1 instance on remote server"
trap 'host_2_cleanup_vhost; error_exit "${FUNCNAME}" "${LINENO}"' INT ERR EXIT
spdk_vhost_run --conf-path=$BASE_DIR --vhost-num=1
$rpc construct_nvme_bdev -b Nvme0 -t rdma -f ipv4 -a $RDMA_TARGET_IP -s 4420 -n "nqn.2018-02.io.spdk:cnode1"
$rpc construct_vhost_scsi_controller $target_vm_ctrl
$rpc add_vhost_scsi_lun $target_vm_ctrl 0 Nvme0n1
vm_setup --os="$os_image" --force=$target_vm --disk-type=spdk_vhost_scsi --disks=VhostScsi0 \
--memory=512 --vhost-num=1 --incoming=$incoming_vm
vm_run $target_vm
sleep 1
# Use this file as a flag to notify main script
# that setup on remote server is done
echo "DONE" > $share_dir/DONE
}
echo $$ > $TEST_DIR/tc3b.pid
host_2_start_vhost
suspend -f
if ! vm_os_booted $target_vm; then
fail "VM$target_vm is not running!"
fi
if ! is_fio_running $target_vm; then
vm_ssh $target_vm "cat /root/migration-tc3.job.out"
error "FIO is not running on remote server after migration!"
fi
notice "Waiting for FIO to finish on remote server VM"
timeout=40
while is_fio_running $target_vm; do
sleep 1
echo -n "."
if (( timeout-- == 0 )); then
error "timeout while waiting for FIO!"
fi
done
notice "FIO result after migration:"
vm_ssh $target_vm "cat /root/migration-tc3.job.out"
host_2_cleanup_vhost
echo "DONE" > $share_dir/DONE

View File

@ -70,6 +70,11 @@ function vm_migrate()
local target_vm_dir="$(readlink -e $from_vm_dir/vm_migrate_to)"
local target_vm="$(basename $target_vm_dir)"
local target_vm_migration_port="$(cat $target_vm_dir/migration_port)"
if [[ -n "$2" ]]; then
local target_ip=$2
else
local target_ip="127.0.0.1"
fi
# Sanity check if target VM (QEMU) is configured to accept source VM (QEMU) migration
if [[ "$(readlink -e ${target_vm_dir}/vm_incoming)" != "$(readlink -e ${from_vm_dir})" ]]; then
@ -80,7 +85,7 @@ function vm_migrate()
notice "Migrating VM $1 to VM "$(basename $target_vm_dir)
echo -e \
"migrate_set_speed 1g\n" \
"migrate tcp:127.0.0.1:$target_vm_migration_port\n" \
"migrate tcp:$target_ip:$target_vm_migration_port\n" \
"info migrate\n" \
"quit" | vm_monitor_send $1 "$from_vm_dir/migration_result"
@ -90,10 +95,15 @@ function vm_migrate()
fail "Migration failed:\n"
fi
# Don't perform the following check if target VM is on remote server
# as we won't have access to it.
# If you need this check then perform it on your own.
if [[ "$target_ip" == "127.0.0.1" ]]; then
if ! vm_os_booted $target_vm; then
fail "VM$target_vm is not running"
cat $target_vm $target_vm_dir/cont_result
fi
fi
notice "Migration complete"
timing_exit vm_migrate
@ -114,9 +124,6 @@ function is_fio_running()
return $ret
}
# FIXME: this shoul'd not be needed
vm_kill_all
for test_case in ${test_cases//,/ }; do
assert_number "$test_case"
notice "==============================="