Testing HOWTO
This HOWTO is intended to demonstrate the basics of configuring and running a small subset of tests on a multi-node configuration. There are many tests available in the suite, and the principles demonstrated here will apply to them all, but this HOWTO will focus on the sanity .
Note that there is a configuration distributed with the Lustre test suite, local.sh, that easily enables you to test Lustre on a single node using lookback devices with no additional configuration needed. This works well, and tests the Lustre software, but the purpose of this HOWTO is to demonstrate using multiple servers and clients to test more Lustre features in an environment representative of a real install.
While these examples do use virtual machines, they are merely examples and the specifics should be easy to apply to real hardware with the prerequisites setup.
System Configuration
This HOWTO uses a cluster of six virtual machines running CentOS 7.1 with a recent lustre-master to run the Lustre tests. Two clients, two MDS, two OSS. This enables testing of a wide variety of Lustre features.
- node01 - MGS and MDS - 192.168.56.201
- 512MB MGT - /dev/sdb
- 1GB MDT - /dev/sdc
- node02 - MDS - 192.168.56.202
- 1GB MDT - /dev/sdb
- node03 - OSS - 192.168.56.203
- Four 16GB OST - /dev/sdb, /dev/sdc, /dev/sdd, /dev/sde
- node04 - OSS - 192.168.56.204
- Four 16GB OST - /dev/sdb, /dev/sdc, /dev/sdd, /dev/sde
- node05 - client - 192.168.56.205
- 16GB shared directory for tests and results
- node06 - client - 192.168.56.206
System Setup
- Install the Lustre clients on two machines, and the Lustre server on the other four, using the same version of Lustre
- Follow this guide to install Lustre RPMs - Walk-thru - Deploying Lustre pre-built RPMs
- Disable SELINUX
- Set SELINUX=disabled in /etc/sysconfig/selinux
- Disable the firewall
- service firewalld stop && systemctl disable firewalld.service
- Generate passwordless ssh keys for hosts and exchange identities across all nodes, and also accept the host keys
- Install PDSH and ensure you can execute commands across the cluster
- Install the epel-release package to enable the EPEL repo
- Install the net-tools package for netstat
- Add user 'runas' with UID 500 and GID 500 to all the nodes
- Create an NFS share that is mounted on all the nodes
- A small number of tests will make use of a shared storage location
- Configure hostnames and populate /etc/hosts
127.0.0.1 localhost localhost.localdomain localhost4 localhost4.localdomain4 ::1 localhost localhost.localdomain localhost6 localhost6.localdomain6 192.168.56.201 node01 node01.unix.localdomain 192.168.56.202 node02 node02.unix.localdomain 192.168.56.203 node03 node03.unix.localdomain 192.168.56.204 node04 node04.unix.localdomain 192.168.56.205 node05 node05.unix.localdomain 192.168.56.206 node06 node06.unix.localdomain
Installing Additional Applications for Testing
- Install dbench - optional, used to gather baseline metrics of disks used in testing
- Available from the EPEL repo, or from the Intel HPDD toolkit builds https://build.hpdd.intel.com/job/toolkit/arch=x86_64,distro=el7/lastSuccessfulBuild/artifact/_topdir/RPMS/x86_64/
- Install iozone - optional, used to gather baseline metrics of filesystem performance
- Available from Intel HPDD EL7 builds of the toolkit - https://build.hpdd.intel.com/job/toolkit/arch=x86_64,distro=el7/lastSuccessfulBuild/artifact/_topdir/RPMS/x86_64/
- Install Parallel IO Simulator (PIOS) - optional, used to simulate shared file and file per process IO
- Available from the Intel HPDD EL6 builds of the toolkit - https://build.hpdd.intel.com/job/toolkit/arch=x86_64,distro=el6/lastSuccessfulBuild/artifact/_topdir/RPMS/x86_64/
Test Configuration
Install this configuration file in /usr/lib64/lustre/tests/cfg/multinode.sh on all of the nodes.
# Enables verbose acc-sm output. VERBOSE=${VERBOSE:-"false"} # File system configuration FSNAME="testfs" FSTYPE=ldiskfs # Network configuration NETTYPE="tcp" # fact hosts mds_HOST=${mds_HOST:-node01} mgs_HOST=${mgs_HOST:-$mds_HOST} ost_HOST=${ost_HOST:-node03} # MDS and MDT configuration MDSCOUNT=2 SINGLEMDS=${SINGLEMDS:-mds1} MDSSIZE=8589000 MDS_FS_MKFS_OPTS=${MDS_FS_MKFS_OPTS:-} MDS_MOUNT_OPTS=${MDS_MOUNT_OPTS:-} MDSFSTYPE=ldiskfs mds1_HOST="node01" MDSDEV1="/dev/sdd" mds1_MOUNT="/mnt/testfs/mdt1" mds1_FSTYPE=ldiskfs mds2_HOST="node02" MDSDEV2="/dev/sdd" mds2_MOUNT="/mnt/testfs/mdt2" mds2_FSTYPE=ldiskfs # MGS and MGT configuration mgs_HOST=${mgs_HOST:-"$mds_HOST"} # combination mgs/mds MGSOPT=${MGSOPT:-} MGS_FS_MKFS_OPTS=${MGS_FS_MKFS_OPTS:-} MGS_MOUNT_OPTS=${MGS_MOUNT_OPTS:-} MGSFSTYPE=ldiskfs MGSDEV="/dev/sdb" MGSSIZE=536000 mgs_MOUNT="/mnt/testfs/mgt" MGSNID="192.168.56.201@tcp" mgs_FSTYPE=ldiskfs # OSS and OST configuration OSTCOUNT=${OSTCOUNT:-8} OSTSIZE=${OSTSIZE:-16777216} OSTFSTYPE=ldiskfs ost1_HOST="node03" OSTDEV1="/dev/sdb" ost1_MOUNT="/mnt/testfs/ost1" ost1_FSTYPE=ldiskfs ost2_HOST="node03" OSTDEV2="/dev/sdc" ost2_MOUNT="/mnt/testfs/ost2" ost2_FSTYPE=ldiskfs ost3_HOST="node03" OSTDEV3="/dev/sdd" ost3_MOUNT="/mnt/testfs/ost3" ost3_FSTYPE=ldiskfs ost4_HOST="node03" OSTDEV4="/dev/sde" ost4_MOUNT="/mnt/testfs/ost4" ost4_FSTYPE=ldiskfs ost5_HOST="node04" OSTDEV5="/dev/sdb" ost5_MOUNT="/mnt/testfs/ost5" ost5_FSTYPE=ldiskfs ost6_HOST="node04" OSTDEV6="/dev/sdc" ost6_MOUNT="/mnt/testfs/ost6" ost6_FSTYPE=ldiskfs ost7_HOST="node04" OSTDEV7="/dev/sdd" ost7_MOUNT="/mnt/testfs/ost7" ost7_FSTYPE=ldiskfs ost8_HOST="node04" OSTDEV8="/dev/sde" ost8_MOUNT="/mnt/testfs/ost8" ost8_FSTYPE=ldiskfs # OST striping configuration STRIPE_BYTES=${STRIPE_BYTES:-1048576} STRIPES_PER_OBJ=${STRIPES_PER_OBJ:-0} # Client configuration CLIENTCOUNT=2 CLIENTS="node05,node06" CLIENT1="node05" CLIENT2="node06" RCLIENTS="node06" MOUNT="/mnt/testfs" MOUNT1="/mnt/testfs" MOUNT2="/mnt/testfs2" DIR=${DIR:-$MOUNT} DIR1=${DIR:-$MOUNT1} DIR2=${DIR2:-$MOUNT2} # UID and GID configuration # Used by several tests to set the UID and GID if [ $UID -ne 0 ]; then log "running as non-root uid $UID" RUNAS_ID="$UID" RUNAS_GID=`id -g $USER` RUNAS="" else RUNAS_ID=${RUNAS_ID:-500} RUNAS_GID=${RUNAS_GID:-$RUNAS_ID} RUNAS=${RUNAS:-"runas -u $RUNAS_ID -g $RUNAS_GID"} fi # Software configuration PDSH="/usr/bin/pdsh -S -Rssh -w" FAILURE_MODE=${FAILURE_MODE:-SOFT} # or HARD POWER_DOWN=${POWER_DOWN:-"powerman --off"} POWER_UP=${POWER_UP:-"powerman --on"} SLOW=${SLOW:-no} FAIL_ON_ERROR=${FAIL_ON_ERROR:-true} # Debug configuration #PTLDEBUG=${PTLDEBUG:-"vfstrace rpctrace dlmtrace neterror ha config \ PTLDEBUG=${PTLDEBUG:-"vfstrace dlmtrace neterror ha config \ ioctl super lfsck"} SUBSYSTEM=${SUBSYSTEM:-"all -lnet -lnd -pinger"} # Lustre timeout TIMEOUT=${TIMEOUT:-"30"} # promise 2MB for every cpu if [ -f /sys/devices/system/cpu/possible ]; then _debug_mb=$((($(cut -d "-" -f 2 /sys/devices/system/cpu/possible)+1)*2)) else _debug_mb=$(($(getconf _NPROCESSORS_CONF)*2)) fi DEBUG_SIZE=${DEBUG_SIZE:-$_debug_mb} DEBUGFS=${DEBUGFS:-"/sbin/debugfs"} TMP=${TMP:-/tmp} SHARED_DIRECTORY=${SHARED_DIRECTORY:-/opt/testing/shared}
Running the Tests
Now you are ready to run the tests. I recommend by starting with one single subtest to check that you have everything configured correctly.
Run this command from the client1 node (node05 in this example) to launch the sanity test, running only test 0.
auster is the tool used to drive the tests. In this example we are passing the flags
- -f multinode - use the multinode.sh configuration in /usr/lib64/lustre/tests/cfg/ when running the tests
- -r - allow the tests to reformat the devices
- -s - run the SLOW tests, which are skipped if we don't pass this flag
- -v - provide verbose output
- sanity --only 0 - run the sanity tests, but only test 0a
[root@node05 tests]# pwd /usr/lib64/lustre/tests [root@node05 tests]# ./auster -f multinode -rsv -d /opt/results/ sanity --only 0a
You will see a lot of output while the tests format the targets, start the filesystem, and mount it on the clients.
You will see this output if the test successfully runs:
== sanity test 0a: touch; rm ========================================================================= 11:17:34 (1441210654) /mnt/testfs/f0a.sanity has type file OK /mnt/testfs/f0a.sanity: absent OK Resetting fail_loc on all nodes...done. PASS 0a (1s) == sanity test complete, duration 6 sec ============================================================== 11:17:35 (1441210655)
Now you're ready to run the full sanity test. Just remove the --only flag to auster:
[root@node05 tests]# pwd /usr/lib64/lustre/tests [root@node05 tests]# ./auster -f multinode -rsv -d /opt/results/ sanity
Running sanity with the -s flag takes about 1h32m in the virtual machine cluster on my computer.
Test Output
The -d flag directed auster to store the results in the base directory /opt/results/. auster will create
/opt/results/<date>/<time>
directories. Inside the directory corresponding to your test run, you'll find the test output.