ZTS: Optimize KSM on Linux and remove it for FreeBSD

Don't use KSM on the FreeBSD VMs and optimize KSM settings for
Linux to have faster run times.

Reviewed-by: Alexander Motin <[email protected]>
Reviewed-by: Tony Hutter <[email protected]>
Signed-off-by: Tino Reichardt <[email protected]>
Closes openzfs#17247

Author: mcmilk

.github/workflows/scripts/generate-ci-type.py

@@ -29,6 +29,7 @@
 Patterns of files that are considered to trigger full CI.
 """
 FULL_RUN_REGEX = list(map(re.compile, [
+    r'\.github/workflows/scripts/.*',
     r'cmd.*',
     r'configs/.*',
     r'META',

.github/workflows/scripts/qemu-1-setup.sh

@@ -10,36 +10,12 @@ set -eu
 export DEBIAN_FRONTEND="noninteractive"
 sudo apt-get -y update
 sudo apt-get install -y axel cloud-image-utils daemonize guestfs-tools \
-  ksmtuned virt-manager linux-modules-extra-$(uname -r) zfsutils-linux
+  virt-manager linux-modules-extra-$(uname -r) zfsutils-linux
 
 # generate ssh keys
 rm -f ~/.ssh/id_ed25519
 ssh-keygen -t ed25519 -f ~/.ssh/id_ed25519 -q -N ""
 
-# we expect RAM shortage
-cat << EOF | sudo tee /etc/ksmtuned.conf > /dev/null
-# /etc/ksmtuned.conf - Configuration file for ksmtuned
-# https://docs.redhat.com/en/documentation/red_hat_enterprise_linux/7/html/virtualization_tuning_and_optimization_guide/chap-ksm
-KSM_MONITOR_INTERVAL=60
-
-# Millisecond sleep between ksm scans for 16Gb server.
-# Smaller servers sleep more, bigger sleep less.
-KSM_SLEEP_MSEC=30
-
-KSM_NPAGES_BOOST=0
-KSM_NPAGES_DECAY=0
-KSM_NPAGES_MIN=1000
-KSM_NPAGES_MAX=25000
-
-KSM_THRES_COEF=80
-KSM_THRES_CONST=8192
-
-LOGFILE=/var/log/ksmtuned.log
-DEBUG=1
-EOF
-sudo systemctl restart ksm
-sudo systemctl restart ksmtuned
-
 # not needed
 sudo systemctl stop docker.socket
 sudo systemctl stop multipathd.socket
@@ -65,16 +41,14 @@ $DISK
 sync
 sleep 1
 
-# swap with same size as RAM
+# swap with same size as RAM (16GiB)
 sudo mkswap $DISK-part1
 sudo swapon $DISK-part1
 
-# 60GB data disk
+# JBOD 2xdisk for OpenZFS storage (test vm's)
 SSD1="$DISK-part2"
-
-# 10GB data disk on ext4
-sudo fallocate -l 10G /test.ssd1
-SSD2=$(sudo losetup -b 4096 -f /test.ssd1 --show)
+sudo fallocate -l 12G /test.ssd2
+SSD2=$(sudo losetup -b 4096 -f /test.ssd2 --show)
 
 # adjust zfs module parameter and create pool
 exec 1>/dev/null
@@ -83,11 +57,11 @@ ARC_MAX=$((1024*1024*512))
 echo $ARC_MIN | sudo tee /sys/module/zfs/parameters/zfs_arc_min
 echo $ARC_MAX | sudo tee /sys/module/zfs/parameters/zfs_arc_max
 echo 1 | sudo tee /sys/module/zfs/parameters/zvol_use_blk_mq
-sudo zpool create -f -o ashift=12 zpool $SSD1 $SSD2 \
-  -O relatime=off -O atime=off -O xattr=sa -O compression=lz4 \
-  -O mountpoint=/mnt/tests
+sudo zpool create -f -o ashift=12 zpool $SSD1 $SSD2 -O relatime=off \
+  -O atime=off -O xattr=sa -O compression=lz4 -O sync=disabled \
+  -O redundant_metadata=none -O mountpoint=/mnt/tests
 
 # no need for some scheduler
 for i in /sys/block/s*/queue/scheduler; do
-  echo "none" | sudo tee $i > /dev/null
+  echo "none" | sudo tee $i
 done

.github/workflows/scripts/qemu-2-start.sh

@@ -150,7 +150,7 @@ echo "ENV=$ENV" >> $ENV
 # result path
 echo 'RESPATH="/var/tmp/test_results"' >> $ENV
 
-# FreeBSD 13 has problems with: e1000+virtio
+# FreeBSD 13 has problems with: e1000 and virtio
 echo "NIC=$NIC" >> $ENV
 
 # freebsd15 -> used in zfs-qemu.yml
@@ -162,6 +162,14 @@ echo "OSv=\"$OSv\"" >> $ENV
 # FreeBSD 15 (Current) -> used for summary
 echo "OSNAME=\"$OSNAME\"" >> $ENV
 
+# default vm count for testings
+VMs=2
+echo "VMs=\"$VMs\"" >> $ENV
+
+# default cpu count for testing vm's
+CPU=2
+echo "CPU=\"$CPU\"" >> $ENV
+
 sudo mkdir -p "/mnt/tests"
 sudo chown -R $(whoami) /mnt/tests
 
@@ -223,12 +231,19 @@ sudo virt-install \
   --disk $DISK,bus=virtio,cache=none,format=$FORMAT,driver.discard=unmap \
   --import --noautoconsole >/dev/null
 
+# enable KSM on Linux
+if [ ${OS:0:7} != "freebsd" ]; then
+  sudo virsh dommemstat --domain "openzfs" --period 5
+  sudo virsh node-memory-tune 100 50 1
+  echo 1 | sudo tee /sys/kernel/mm/ksm/run > /dev/null
+fi
+
 # Give the VMs hostnames so we don't have to refer to them with
 # hardcoded IP addresses.
 #
 # vm0:          Initial VM we install dependencies and build ZFS on.
 # vm1..2        Testing VMs
-for i in {0..9} ; do
+for ((i=0; i<=VMs; i++)); do
   echo "192.168.122.1$i vm$i" | sudo tee -a /etc/hosts
 done
 

.github/workflows/scripts/qemu-5-setup.sh

@@ -14,39 +14,33 @@ PID=$(pidof /usr/bin/qemu-system-x86_64)
 tail --pid=$PID -f /dev/null
 sudo virsh undefine openzfs
 
-# default values per test vm:
-VMs=2
-CPU=2
-
 # cpu pinning
 CPUSET=("0,1" "2,3")
 
 case "$OS" in
   freebsd*)
-    # FreeBSD can't be optimized via ksmtuned
+    # FreeBSD needs only 6GiB
     RAM=6
     ;;
   *)
-    # Linux can be optimized via ksmtuned
+    # Linux needs more memory, but can be optimized to share it via KSM
     RAM=8
     ;;
 esac
 
-# this can be different for each distro
-echo "VMs=$VMs" >> $ENV
-
 # create snapshot we can clone later
 sudo zfs snapshot zpool/openzfs@now
 
 # setup the testing vm's
 PUBKEY=$(cat ~/.ssh/id_ed25519.pub)
-for i in $(seq 1 $VMs); do
 
+# start testing VMs
+for ((i=1; i<=VMs; i++)); do
   echo "Creating disk for vm$i..."
   DISK="/dev/zvol/zpool/vm$i"
   FORMAT="raw"
-  sudo zfs clone zpool/openzfs@now zpool/vm$i
-  sudo zfs create -ps -b 64k -V 80g zpool/vm$i-2
+  sudo zfs clone zpool/openzfs@now zpool/vm$i-system
+  sudo zfs create -ps -b 64k -V 64g zpool/vm$i-tests
 
   cat <<EOF > /tmp/user-data
 #cloud-config
@@ -83,23 +77,21 @@ EOF
     --graphics none \
     --cloud-init user-data=/tmp/user-data \
     --network bridge=virbr0,model=$NIC,mac="52:54:00:83:79:0$i" \
-    --disk $DISK,bus=virtio,cache=none,format=$FORMAT,driver.discard=unmap \
-    --disk $DISK-2,bus=virtio,cache=none,format=$FORMAT,driver.discard=unmap \
+    --disk $DISK-system,bus=virtio,cache=none,format=$FORMAT,driver.discard=unmap \
+    --disk $DISK-tests,bus=virtio,cache=none,format=$FORMAT,driver.discard=unmap \
     --import --noautoconsole >/dev/null
 done
 
-# check the memory state from time to time
+# generate some memory stats
 cat <<EOF > cronjob.sh
-# $OS
 exec 1>>/var/tmp/stats.txt
 exec 2>&1
-echo "*******************************************************"
-date
+echo "********************************************************************************"
 uptime
 free -m
-df -h /mnt/tests
 zfs list
 EOF
+
 sudo chmod +x cronjob.sh
 sudo mv -f cronjob.sh /root/cronjob.sh
 echo '*/5 * * * *  /root/cronjob.sh' > crontab.txt
@@ -108,15 +100,15 @@ rm crontab.txt
 
 # check if the machines are okay
 echo "Waiting for vm's to come up...  (${VMs}x CPU=$CPU RAM=$RAM)"
-for i in $(seq 1 $VMs); do
+for ((i=1; i<=VMs; i++)); do
   .github/workflows/scripts/qemu-wait-for-vm.sh vm$i
 done
 echo "All $VMs VMs are up now."
 
 # Save the VM's serial output (ttyS0) to /var/tmp/console.txt
 # - ttyS0 on the VM corresponds to a local /dev/pty/N entry
 # - use 'virsh ttyconsole' to lookup the /dev/pty/N entry
-for i in $(seq 1 $VMs); do
+for ((i=1; i<=VMs; i++)); do
   mkdir -p $RESPATH/vm$i
   read "pty" <<< $(sudo virsh ttyconsole vm$i)
   sudo nohup bash -c "cat $pty > $RESPATH/vm$i/console.txt" &

.github/workflows/scripts/qemu-6-tests.sh

@@ -45,7 +45,7 @@ if [ -z ${1:-} ]; then
   echo 0 > /tmp/ctr
   date "+%s" > /tmp/tsstart
 
-  for i in $(seq 1 $VMs); do
+  for ((i=1; i<=VMs; i++)); do
     IP="192.168.122.1$i"
     daemonize -c /var/tmp -p vm${i}.pid -o vm${i}log.txt -- \
       $SSH zfs@$IP $TESTS $OS $i $VMs $CI_TYPE
@@ -58,7 +58,7 @@ if [ -z ${1:-} ]; then
   done
 
   # wait for all vm's to finish
-  for i in $(seq 1 $VMs); do
+  for ((i=1; i<=VMs; i++)); do
     tail --pid=$(cat vm${i}.pid) -f /dev/null
     pid=$(cat vm${i}log.pid)
     rm -f vm${i}log.pid
@@ -72,19 +72,24 @@ fi
 export PATH="$PATH:/bin:/sbin:/usr/bin:/usr/sbin:/usr/local/sbin:/usr/local/bin"
 case "$1" in
   freebsd*)
+    TDIR="/usr/local/share/zfs"
     sudo kldstat -n zfs 2>/dev/null && sudo kldunload zfs
     sudo -E ./zfs/scripts/zfs.sh
-    TDIR="/usr/local/share/zfs"
+    sudo mv -f /var/tmp/*.txt /tmp
+    sudo newfs -U -t -L tmp /dev/vtbd1 >/dev/null
+    sudo mount -o noatime /dev/vtbd1 /var/tmp
+    sudo chmod 1777 /var/tmp
+    sudo mv -f /tmp/*.txt /var/tmp
     ;;
   *)
     # use xfs @ /var/tmp for all distros
+    TDIR="/usr/share/zfs"
+    sudo -E modprobe zfs
     sudo mv -f /var/tmp/*.txt /tmp
     sudo mkfs.xfs -fq /dev/vdb
     sudo mount -o noatime /dev/vdb /var/tmp
     sudo chmod 1777 /var/tmp
     sudo mv -f /tmp/*.txt /var/tmp
-    sudo -E modprobe zfs
-    TDIR="/usr/share/zfs"
     ;;
 esac
 

.github/workflows/scripts/qemu-7-prepare.sh

@@ -28,7 +28,7 @@ BASE="$HOME/work/zfs/zfs"
 MERGE="$BASE/.github/workflows/scripts/merge_summary.awk"
 
 # catch result files of testings (vm's should be there)
-for i in $(seq 1 $VMs); do
+for ((i=1; i<=VMs; i++)); do
   rsync -arL zfs@vm$i:$RESPATH/current $RESPATH/vm$i || true
   scp zfs@vm$i:"/var/tmp/*.txt" $RESPATH/vm$i || true
   scp zfs@vm$i:"/var/tmp/*.rpm" $RESPATH/vm$i || true
@@ -37,7 +37,7 @@ cp -f /var/tmp/*.txt $RESPATH || true
 cd $RESPATH
 
 # prepare result files for summary
-for i in $(seq 1 $VMs); do
+for ((i=1; i<=VMs; i++)); do
   file="vm$i/build-stderr.txt"
   test -s $file && mv -f $file build-stderr.txt
 

.github/workflows/scripts/qemu-8-summary.sh

@@ -45,7 +45,7 @@ fi
 
 echo -e "\nFull logs for download:\n    $1\n"
 
-for i in $(seq 1 $VMs); do
+for ((i=1; i<=VMs; i++)); do
   rv=$(cat vm$i/tests-exitcode.txt)
 
   if [ $rv = 0 ]; then