|
| 1 | +#!/bin/bash |
| 2 | +# ScyllaDB Host Setup Script |
| 3 | +# This script runs in a privileged container to configure host system for ScyllaDB |
| 4 | +# Values are calculated dynamically based on system resources per ScyllaDB recommendations |
| 5 | +# With --persist flag, also creates /etc/sysctl.d/99-scylladb.conf for permanent settings |
| 6 | + |
| 7 | +# Don't exit on errors - we want to configure what we can |
| 8 | +set +e |
| 9 | + |
| 10 | +# Check for --persist flag |
| 11 | +PERSIST_MODE=false |
| 12 | +if [ "$1" = "--persist" ]; then |
| 13 | + PERSIST_MODE=true |
| 14 | + echo "=== Running in PERSIST mode - will create /etc/sysctl.d/99-scylladb.conf ===" |
| 15 | +fi |
| 16 | + |
| 17 | +echo "=== ScyllaDB Host System Setup ===" |
| 18 | +if [ "$PERSIST_MODE" = true ]; then |
| 19 | + echo "Mode: Persistent (creating sysctl.d configuration)" |
| 20 | +else |
| 21 | + echo "Mode: Temporary (runtime only)" |
| 22 | +fi |
| 23 | +echo "" |
| 24 | + |
| 25 | +# Check if running with sufficient privileges |
| 26 | +if [ ! -d /host/proc ]; then |
| 27 | + echo "ERROR: Host /proc not mounted. Please ensure volumes are configured:" |
| 28 | + echo " volumes:" |
| 29 | + echo " - /proc:/host/proc" |
| 30 | + echo " - /sys:/host/sys" |
| 31 | + if [ "$PERSIST_MODE" = true ]; then |
| 32 | + echo " - /etc/sysctl.d:/host/sysctl.d" |
| 33 | + fi |
| 34 | + exit 1 |
| 35 | +fi |
| 36 | + |
| 37 | +# Check sysctl.d mount if in persist mode |
| 38 | +if [ "$PERSIST_MODE" = true ] && [ ! -d /host/sysctl.d ]; then |
| 39 | + echo "ERROR: Host /etc/sysctl.d not mounted. Please ensure volume is configured:" |
| 40 | + echo " volumes:" |
| 41 | + echo " - /etc/sysctl.d:/host/sysctl.d" |
| 42 | + exit 1 |
| 43 | +fi |
| 44 | + |
| 45 | +# Test if we have write access to critical parameters |
| 46 | +CAN_WRITE=true |
| 47 | +if [ -f /host/proc/sys/fs/aio-max-nr ]; then |
| 48 | + CURRENT_AIO=$(cat /host/proc/sys/fs/aio-max-nr 2>/dev/null) |
| 49 | + if ! echo "$CURRENT_AIO" >/host/proc/sys/fs/aio-max-nr 2>/dev/null; then |
| 50 | + echo "WARNING: Cannot write to /host/proc/sys - some settings may not be applied" |
| 51 | + echo "Container may need privileged mode for full configuration" |
| 52 | + CAN_WRITE=false |
| 53 | + fi |
| 54 | +else |
| 55 | + echo "WARNING: Cannot access /host/proc/sys/fs/aio-max-nr" |
| 56 | + CAN_WRITE=false |
| 57 | +fi |
| 58 | + |
| 59 | +# Get system information |
| 60 | +echo "=== System Information ===" |
| 61 | +CPU_CORES=$(nproc) |
| 62 | +MEMORY_KB=$(grep MemTotal /host/proc/meminfo | awk '{print $2}') |
| 63 | +MEMORY_GB=$((MEMORY_KB / 1024 / 1024)) |
| 64 | +echo "CPU Cores: ${CPU_CORES}" |
| 65 | +echo "Total Memory: ${MEMORY_GB} GB" |
| 66 | +echo "" |
| 67 | + |
| 68 | +# Calculate dynamic values based on system resources |
| 69 | +# Based on ScyllaDB documentation and best practices |
| 70 | + |
| 71 | +# AIO: ScyllaDB recommends 65536 * number of shards |
| 72 | +# ScyllaDB typically uses 1 shard per core, but we'll add buffer |
| 73 | +SHARDS_PER_CORE=1 |
| 74 | +ESTIMATED_SHARDS=$((CPU_CORES * SHARDS_PER_CORE)) |
| 75 | +AIO_PER_SHARD=65536 |
| 76 | +# Add 50% buffer for safety |
| 77 | +AIO_REQUIRED=$((ESTIMATED_SHARDS * AIO_PER_SHARD * 3 / 2)) |
| 78 | +# Ensure minimum of 1048576 as recommended |
| 79 | +if [ "$AIO_REQUIRED" -lt 1048576 ]; then |
| 80 | + AIO_REQUIRED=1048576 |
| 81 | +fi |
| 82 | + |
| 83 | +# Network settings based on cores |
| 84 | +# ScyllaDB handles high connection counts |
| 85 | +SOMAXCONN=$((CPU_CORES * 1024)) |
| 86 | +if [ "$SOMAXCONN" -lt 4096 ]; then |
| 87 | + SOMAXCONN=4096 |
| 88 | +elif [ "$SOMAXCONN" -gt 65535 ]; then |
| 89 | + SOMAXCONN=65535 |
| 90 | +fi |
| 91 | + |
| 92 | +TCP_MAX_SYN_BACKLOG=$((CPU_CORES * 512)) |
| 93 | +if [ "$TCP_MAX_SYN_BACKLOG" -lt 4096 ]; then |
| 94 | + TCP_MAX_SYN_BACKLOG=4096 |
| 95 | +elif [ "$TCP_MAX_SYN_BACKLOG" -gt 65535 ]; then |
| 96 | + TCP_MAX_SYN_BACKLOG=65535 |
| 97 | +fi |
| 98 | + |
| 99 | +# Memory map count: ScyllaDB recommends high values |
| 100 | +# Base calculation: 65530 per GB of RAM |
| 101 | +VM_MAX_MAP_COUNT=$((MEMORY_GB * 65530)) |
| 102 | +if [ "$VM_MAX_MAP_COUNT" -lt 1048575 ]; then |
| 103 | + # ScyllaDB minimum recommendation |
| 104 | + VM_MAX_MAP_COUNT=1048575 |
| 105 | +fi |
| 106 | + |
| 107 | +# Network buffer sizes: 1% of RAM but capped |
| 108 | +NET_MEM_BYTES=$((MEMORY_KB * 1024 / 100)) |
| 109 | +if [ "$NET_MEM_BYTES" -gt 134217728 ]; then |
| 110 | + # Cap at 128MB |
| 111 | + NET_MEM_BYTES=134217728 |
| 112 | +elif [ "$NET_MEM_BYTES" -lt 16777216 ]; then |
| 113 | + # Minimum 16MB for good performance |
| 114 | + NET_MEM_BYTES=16777216 |
| 115 | +fi |
| 116 | + |
| 117 | +# File descriptors: ScyllaDB needs many |
| 118 | +FD_LIMIT=$((CPU_CORES * 200000)) |
| 119 | +if [ "$FD_LIMIT" -lt 1000000 ]; then |
| 120 | + FD_LIMIT=1000000 |
| 121 | +fi |
| 122 | + |
| 123 | +# Function to safely set sysctl values |
| 124 | +set_sysctl() { |
| 125 | + local param=$1 |
| 126 | + local value=$2 |
| 127 | + local description=$3 |
| 128 | + local current_value |
| 129 | + |
| 130 | + current_value=$(cat /host/proc/sys/${param//\.//} 2>/dev/null || echo "0") |
| 131 | + |
| 132 | + if [ "$current_value" -lt "$value" ]; then |
| 133 | + echo "Setting ${param} from ${current_value} to ${value} (${description})" |
| 134 | + if echo "$value" >/host/proc/sys/${param//\.//} 2>/dev/null; then |
| 135 | + echo "✓ ${param} = ${value}" |
| 136 | + else |
| 137 | + echo "⚠ Could not set ${param} (may need host reboot)" |
| 138 | + fi |
| 139 | + else |
| 140 | + echo "✓ ${param} = ${current_value} (already sufficient, recommended: ${value})" |
| 141 | + fi |
| 142 | +} |
| 143 | + |
| 144 | +echo "=== Configuring Host System Parameters ===" |
| 145 | +echo "" |
| 146 | + |
| 147 | +# 1. AIO - Most critical for ScyllaDB |
| 148 | +echo "1. Asynchronous I/O (AIO) Configuration:" |
| 149 | +echo " Calculated: ${AIO_REQUIRED} (${CPU_CORES} cores × ${SHARDS_PER_CORE} shard × ${AIO_PER_SHARD} + buffer)" |
| 150 | +set_sysctl "fs.aio-max-nr" "${AIO_REQUIRED}" "async I/O operations" |
| 151 | +echo "" |
| 152 | + |
| 153 | +# 2. Network settings |
| 154 | +echo "2. Network Settings (based on ${CPU_CORES} cores):" |
| 155 | +set_sysctl "net.core.somaxconn" "${SOMAXCONN}" "socket listen backlog" |
| 156 | +set_sysctl "net.ipv4.tcp_max_syn_backlog" "${TCP_MAX_SYN_BACKLOG}" "TCP SYN queue" |
| 157 | + |
| 158 | +# Disable SYN cookies (ScyllaDB recommendation) |
| 159 | +echo "0" >/host/proc/sys/net/ipv4/tcp_syncookies 2>/dev/null && |
| 160 | + echo "✓ Disabled TCP SYN cookies (for performance)" || |
| 161 | + echo "⚠ Could not disable TCP SYN cookies" |
| 162 | + |
| 163 | +# Network buffers |
| 164 | +set_sysctl "net.core.rmem_max" "${NET_MEM_BYTES}" "receive buffer max" |
| 165 | +set_sysctl "net.core.wmem_max" "${NET_MEM_BYTES}" "send buffer max" |
| 166 | +set_sysctl "net.core.netdev_max_backlog" "10000" "network device backlog" |
| 167 | + |
| 168 | +# TCP memory and buffers (space-separated values need special handling) |
| 169 | +echo "Setting TCP buffer sizes..." |
| 170 | +echo "4096 87380 ${NET_MEM_BYTES}" >/host/proc/sys/net/ipv4/tcp_rmem 2>/dev/null || |
| 171 | + echo "⚠ Could not set TCP receive buffers" |
| 172 | +echo "4096 65536 ${NET_MEM_BYTES}" >/host/proc/sys/net/ipv4/tcp_wmem 2>/dev/null || |
| 173 | + echo "⚠ Could not set TCP send buffers" |
| 174 | + |
| 175 | +# TCP tuning for low latency |
| 176 | +set_sysctl "net.ipv4.tcp_timestamps" "1" "TCP timestamps" |
| 177 | +set_sysctl "net.ipv4.tcp_sack" "1" "TCP selective ack" |
| 178 | +set_sysctl "net.ipv4.tcp_window_scaling" "1" "TCP window scaling" |
| 179 | +echo "" |
| 180 | + |
| 181 | +# 3. Memory settings |
| 182 | +echo "3. Memory Settings (based on ${MEMORY_GB} GB RAM):" |
| 183 | +set_sysctl "vm.max_map_count" "${VM_MAX_MAP_COUNT}" "memory map areas" |
| 184 | +set_sysctl "vm.swappiness" "0" "disable swap usage" |
| 185 | +set_sysctl "vm.dirty_ratio" "5" "dirty page ratio" |
| 186 | +set_sysctl "vm.dirty_background_ratio" "2" "background dirty ratio" |
| 187 | +echo "" |
| 188 | + |
| 189 | +# 4. File descriptors |
| 190 | +echo "4. File Descriptor Limits:" |
| 191 | +set_sysctl "fs.file-max" "${FD_LIMIT}" "max file descriptors" |
| 192 | +set_sysctl "fs.nr_open" "${FD_LIMIT}" "max open files per process" |
| 193 | +echo "" |
| 194 | + |
| 195 | +# 5. Transparent Huge Pages (THP) - ScyllaDB requires this disabled |
| 196 | +echo "5. Transparent Huge Pages (THP):" |
| 197 | +THP_STATUS=$(cat /host/sys/kernel/mm/transparent_hugepage/enabled 2>/dev/null | grep -o '\[.*\]' | tr -d '[]' || echo "unknown") |
| 198 | +if [ "$THP_STATUS" != "never" ]; then |
| 199 | + echo "Warning: THP is '${THP_STATUS}', ScyllaDB requires 'never'" |
| 200 | + if echo never >/host/sys/kernel/mm/transparent_hugepage/enabled 2>/dev/null; then |
| 201 | + echo "✓ THP disabled" |
| 202 | + else |
| 203 | + echo "⚠ Could not disable THP (may need kernel boot parameter)" |
| 204 | + fi |
| 205 | + |
| 206 | + # Also disable defrag |
| 207 | + echo never >/host/sys/kernel/mm/transparent_hugepage/defrag 2>/dev/null || true |
| 208 | +else |
| 209 | + echo "✓ THP is already disabled" |
| 210 | +fi |
| 211 | +echo "" |
| 212 | + |
| 213 | +# 6. CPU frequency scaling (for consistent performance) |
| 214 | +echo "6. CPU Performance Settings:" |
| 215 | +if [ -d /host/sys/devices/system/cpu/cpu0/cpufreq ]; then |
| 216 | + for gov in /host/sys/devices/system/cpu/cpu*/cpufreq/scaling_governor; do |
| 217 | + echo "performance" >"$gov" 2>/dev/null || true |
| 218 | + done |
| 219 | + echo "✓ Set CPU governor to performance mode" |
| 220 | +else |
| 221 | + echo "! CPU frequency scaling not available" |
| 222 | +fi |
| 223 | +echo "" |
| 224 | + |
| 225 | +# 7. NUMA settings if available |
| 226 | +echo "7. NUMA Settings:" |
| 227 | +if [ -f /host/proc/sys/kernel/numa_balancing ]; then |
| 228 | + echo "0" >/host/proc/sys/kernel/numa_balancing 2>/dev/null && |
| 229 | + echo "✓ Disabled NUMA balancing (ScyllaDB manages NUMA)" || |
| 230 | + echo "⚠ Could not disable NUMA balancing" |
| 231 | +else |
| 232 | + echo "! NUMA not available on this system" |
| 233 | +fi |
| 234 | +echo "" |
| 235 | + |
| 236 | +# Create persistent sysctl configuration if requested |
| 237 | +if [ "$PERSIST_MODE" = true ]; then |
| 238 | + echo "" |
| 239 | + echo "=== Creating Persistent Configuration ===" |
| 240 | + |
| 241 | + SYSCTL_CONFIG="/host/sysctl.d/99-scylladb.conf" |
| 242 | + |
| 243 | + # Check if config already exists |
| 244 | + if [ -f "$SYSCTL_CONFIG" ]; then |
| 245 | + echo "Found existing $SYSCTL_CONFIG" |
| 246 | + echo "Backing up to ${SYSCTL_CONFIG}.bak" |
| 247 | + cp "$SYSCTL_CONFIG" "${SYSCTL_CONFIG}.bak" |
| 248 | + fi |
| 249 | + |
| 250 | + cat >"$SYSCTL_CONFIG" <<EOF |
| 251 | +# ScyllaDB Performance Tuning |
| 252 | +# Generated on $(date) |
| 253 | +# System: ${CPU_CORES} cores, ${MEMORY_GB} GB RAM |
| 254 | +
|
| 255 | +# Asynchronous I/O |
| 256 | +fs.aio-max-nr = ${AIO_REQUIRED} |
| 257 | +
|
| 258 | +# Network settings |
| 259 | +net.core.somaxconn = ${SOMAXCONN} |
| 260 | +net.ipv4.tcp_max_syn_backlog = ${TCP_MAX_SYN_BACKLOG} |
| 261 | +net.ipv4.tcp_syncookies = 0 |
| 262 | +net.core.rmem_max = ${NET_MEM_BYTES} |
| 263 | +net.core.wmem_max = ${NET_MEM_BYTES} |
| 264 | +net.core.netdev_max_backlog = 10000 |
| 265 | +net.ipv4.tcp_timestamps = 1 |
| 266 | +net.ipv4.tcp_sack = 1 |
| 267 | +net.ipv4.tcp_window_scaling = 1 |
| 268 | +
|
| 269 | +# Memory settings |
| 270 | +vm.max_map_count = ${VM_MAX_MAP_COUNT} |
| 271 | +vm.swappiness = 0 |
| 272 | +vm.dirty_ratio = 5 |
| 273 | +vm.dirty_background_ratio = 2 |
| 274 | +
|
| 275 | +# File descriptors |
| 276 | +fs.file-max = ${FD_LIMIT} |
| 277 | +fs.nr_open = ${FD_LIMIT} |
| 278 | +
|
| 279 | +# NUMA (if available) |
| 280 | +kernel.numa_balancing = 0 |
| 281 | +
|
| 282 | +# TCP buffer sizes (space-separated values handled separately) |
| 283 | +# Apply with: sysctl -p /etc/sysctl.d/99-scylladb.conf |
| 284 | +# Then manually set: |
| 285 | +# echo "4096 87380 ${NET_MEM_BYTES}" > /proc/sys/net/ipv4/tcp_rmem |
| 286 | +# echo "4096 65536 ${NET_MEM_BYTES}" > /proc/sys/net/ipv4/tcp_wmem |
| 287 | +EOF |
| 288 | + |
| 289 | + if [ -f "$SYSCTL_CONFIG" ]; then |
| 290 | + echo "✓ Created $SYSCTL_CONFIG successfully" |
| 291 | + echo "" |
| 292 | + echo "To apply on host system, run:" |
| 293 | + echo " sudo sysctl --system" |
| 294 | + echo "" |
| 295 | + echo "Note: Some settings require additional steps:" |
| 296 | + echo " - Transparent Huge Pages: Add 'transparent_hugepage=never' to kernel boot parameters" |
| 297 | + echo " - TCP buffers: Apply manually as shown in config comments" |
| 298 | + echo " - CPU governor: Set via cpupower or similar tool" |
| 299 | + else |
| 300 | + echo "⚠ Failed to create $SYSCTL_CONFIG" |
| 301 | + fi |
| 302 | +fi |
| 303 | + |
| 304 | +# Create a flag file to indicate setup is complete |
| 305 | +touch /tmp/scylla-setup-complete |
| 306 | + |
| 307 | +echo "=== Configuration Summary ===" |
| 308 | +echo "System Resources:" |
| 309 | +echo " - CPU Cores: ${CPU_CORES}" |
| 310 | +echo " - Memory: ${MEMORY_GB} GB" |
| 311 | +echo "" |
| 312 | +echo "Applied Settings:" |
| 313 | +echo " - AIO max-nr: ${AIO_REQUIRED}" |
| 314 | +echo " - Socket backlog: ${SOMAXCONN}" |
| 315 | +echo " - TCP SYN backlog: ${TCP_MAX_SYN_BACKLOG}" |
| 316 | +echo " - VM max map count: ${VM_MAX_MAP_COUNT}" |
| 317 | +echo " - Network buffers: $((NET_MEM_BYTES / 1024 / 1024)) MB" |
| 318 | +echo " - File descriptors: ${FD_LIMIT}" |
| 319 | +echo "" |
| 320 | + |
| 321 | +echo "=== ScyllaDB Host Setup Complete ===" |
| 322 | +echo "" |
| 323 | +echo "All parameters have been configured based on system resources." |
| 324 | +echo "ScyllaDB containers can now start with optimal performance settings." |
| 325 | +echo "" |
| 326 | + |
| 327 | +exit 0 |
0 commit comments