Split View: Linux 커널 파라미터 튜닝 가이드: sysctl + Boot Params, 안전한 변경과 롤백

#!/usr/bin/env bash
# pre-tuning-check.sh - Back up state before tuning

BACKUP_DIR="/root/kernel-tuning-backup/$(date +%Y%m%d-%H%M%S)"
mkdir -p "$BACKUP_DIR"

# 1. Full sysctl dump
sysctl -a > "$BACKUP_DIR/sysctl-before.txt" 2>/dev/null

# 2. Current boot parameters
cat /proc/cmdline > "$BACKUP_DIR/cmdline-before.txt"

# 3. GRUB config backup
cp /etc/default/grub "$BACKUP_DIR/grub-before"
[[ -d /etc/sysctl.d ]] && cp -r /etc/sysctl.d "$BACKUP_DIR/sysctl.d-before"

# 4. System state snapshot
free -h > "$BACKUP_DIR/memory-before.txt"
ss -s > "$BACKUP_DIR/socket-stats-before.txt"
vmstat 1 5 > "$BACKUP_DIR/vmstat-before.txt"
cat /proc/net/sockstat > "$BACKUP_DIR/sockstat-before.txt"

echo "Backup complete: $BACKUP_DIR"

1. Test in staging environment
2. Apply to 1 canary server -> Monitor (minimum 24 hours)
3. If no issues, rolling apply by group
4. Compare metrics after application (before vs after)

# sysctl rollback - Restore specific parameter from backup
PARAM="net.core.somaxconn"
OLD_VALUE=$(grep "^${PARAM}" /root/kernel-tuning-backup/latest/sysctl-before.txt | awk '{print $3}')
sysctl -w "${PARAM}=${OLD_VALUE}"

# Full sysctl rollback
while IFS='= ' read -r key value; do
  sysctl -w "${key}=${value}" 2>/dev/null
done < /root/kernel-tuning-backup/latest/sysctl-before.txt

# Boot parameter rollback - Restore previous GRUB config
cp /root/kernel-tuning-backup/latest/grub-before /etc/default/grub
grub2-mkconfig -o /boot/grub2/grub.cfg  # RHEL
# update-grub                            # Ubuntu

# /etc/sysctl.d/10-network.conf

# TCP backlog - Essential for high-traffic servers
net.core.somaxconn = 65535
net.ipv4.tcp_max_syn_backlog = 65535

# TCP socket buffers (bytes)
# min / default / max
net.core.rmem_default = 262144
net.core.rmem_max = 16777216
net.core.wmem_default = 262144
net.core.wmem_max = 16777216
net.ipv4.tcp_rmem = 4096 262144 16777216
net.ipv4.tcp_wmem = 4096 262144 16777216

# TCP congestion control
net.ipv4.tcp_congestion_control = bbr
net.core.default_qdisc = fq

# TIME_WAIT management
net.ipv4.tcp_tw_reuse = 1
net.ipv4.tcp_fin_timeout = 15
net.ipv4.tcp_max_tw_buckets = 2000000

# Keepalive (servers behind load balancers)
net.ipv4.tcp_keepalive_time = 60
net.ipv4.tcp_keepalive_intvl = 10
net.ipv4.tcp_keepalive_probes = 6

# Load BBR kernel module
modprobe tcp_bbr
echo "tcp_bbr" >> /etc/modules-load.d/bbr.conf

# Apply sysctl
sysctl -w net.core.default_qdisc=fq
sysctl -w net.ipv4.tcp_congestion_control=bbr

# Verify
sysctl net.ipv4.tcp_congestion_control
# net.ipv4.tcp_congestion_control = bbr

# /etc/sysctl.d/20-memory.conf

# Swap tendency (0=minimal, 100=aggressive)
# DB servers: 1-10, Web servers: 10-30
vm.swappiness = 10

# Dirty page ratio - disk write delay
vm.dirty_ratio = 40              # Max dirty page ratio relative to total memory
vm.dirty_background_ratio = 10   # Background flush start ratio

# OOM-related
vm.overcommit_memory = 0         # 0=default(heuristic), 1=always allow, 2=restrict
vm.panic_on_oom = 0              # Whether to panic on OOM (0=run OOM Killer)

# Max memory map areas (Elasticsearch, MongoDB, etc.)
vm.max_map_count = 262144

# Filesystem cache release (emergency only)
# echo 3 > /proc/sys/vm/drop_caches  # 1=pagecache, 2=dentries+inodes, 3=all

# Transparent Huge Pages (THP) - Recommended to disable for DB
# Set via boot parameter
# Add to GRUB_CMDLINE_LINUX:
# transparent_hugepage=never

# Runtime check/change
cat /sys/kernel/mm/transparent_hugepage/enabled
echo never > /sys/kernel/mm/transparent_hugepage/enabled

# Static Huge Pages (Oracle DB, DPDK, etc.)
# /etc/sysctl.d/20-memory.conf
vm.nr_hugepages = 1024  # 2MB * 1024 = 2GB

# Verify
grep -i huge /proc/meminfo

# /etc/sysctl.d/30-fs.conf

# Max open files (system-wide)
fs.file-max = 2097152

# inotify watch limits (IDE, file watch services)
fs.inotify.max_user_watches = 524288
fs.inotify.max_user_instances = 8192

# AIO (Asynchronous I/O) max request count
fs.aio-max-nr = 1048576

# /etc/security/limits.d/99-app.conf
# Must be configured together with sysctl's fs.file-max to take effect

*    soft    nofile    1048576
*    hard    nofile    1048576
*    soft    nproc     65535
*    hard    nproc     65535
*    soft    memlock   unlimited
*    hard    memlock   unlimited

# Check current scheduler
cat /sys/block/sda/queue/scheduler

# SSD: none or mq-deadline recommended
echo mq-deadline > /sys/block/sda/queue/scheduler

# Persistent config (udev rule)
# /etc/udev/rules.d/60-scheduler.rules
# ACTION=="add|change", KERNEL=="sd*", ATTR{queue/rotational}=="0", ATTR{queue/scheduler}="mq-deadline"
# ACTION=="add|change", KERNEL=="sd*", ATTR{queue/rotational}=="1", ATTR{queue/scheduler}="bfq"

# /etc/sysctl.d/40-security.conf

# ASLR (Address Space Layout Randomization)
kernel.randomize_va_space = 2  # 0=off, 1=partial, 2=full

# SysRq restriction (allow emergency recovery only)
kernel.sysrq = 176  # Bitmask: sync + remount-ro + reboot

# Core dump restriction
kernel.core_pattern = |/bin/false
fs.suid_dumpable = 0

# dmesg access restriction
kernel.dmesg_restrict = 1

# Kernel pointer hiding
kernel.kptr_restrict = 2

# BPF restriction (unprivileged users)
kernel.unprivileged_bpf_disabled = 1

# Network security
net.ipv4.conf.all.rp_filter = 1           # Reverse Path Filtering
net.ipv4.conf.all.accept_redirects = 0     # Reject ICMP redirects
net.ipv4.conf.all.send_redirects = 0
net.ipv6.conf.all.accept_redirects = 0
net.ipv4.conf.all.accept_source_route = 0  # Reject Source Routing
net.ipv4.conf.all.log_martians = 1         # Log suspicious packets
net.ipv4.icmp_echo_ignore_broadcasts = 1   # Smurf attack defense

# IP forwarding (disable unless router/container host)
net.ipv4.ip_forward = 0
# For container hosts:
# net.ipv4.ip_forward = 1

# Check current boot parameters
cat /proc/cmdline

# RHEL / Rocky
vi /etc/default/grub
# GRUB_CMDLINE_LINUX="... parameters_to_add"
grub2-mkconfig -o /boot/grub2/grub.cfg

# Ubuntu
vi /etc/default/grub
# GRUB_CMDLINE_LINUX_DEFAULT="... parameters_to_add"
update-grub

# Check currently applied mitigations
grep -r . /sys/devices/system/cpu/vulnerabilities/ 2>/dev/null

# Disable mitigations (benchmark/isolated environments only!)
# Add to GRUB_CMDLINE_LINUX:
# mitigations=off

# Performance impact (varies by workload)
# mitigations=auto: 5~30% overhead on syscall-intensive workloads
# mitigations=off: Security risk - not recommended for production

# /etc/sysctl.d/99-web-server.conf

# Network
net.core.somaxconn = 65535
net.ipv4.tcp_max_syn_backlog = 65535
net.ipv4.ip_local_port_range = 1024 65535
net.ipv4.tcp_tw_reuse = 1
net.ipv4.tcp_fin_timeout = 15
net.ipv4.tcp_keepalive_time = 60
net.ipv4.tcp_keepalive_intvl = 10
net.ipv4.tcp_keepalive_probes = 6
net.ipv4.tcp_congestion_control = bbr
net.core.default_qdisc = fq
net.core.rmem_max = 16777216
net.core.wmem_max = 16777216

# Files
fs.file-max = 2097152

# Memory
vm.swappiness = 10

# /etc/sysctl.d/99-database.conf

# Memory
vm.swappiness = 1
vm.dirty_ratio = 40
vm.dirty_background_ratio = 10
vm.overcommit_memory = 0
vm.max_map_count = 262144

# Files
fs.file-max = 2097152
fs.aio-max-nr = 1048576

# Network (primarily internal communication)
net.core.somaxconn = 65535
net.ipv4.tcp_keepalive_time = 60

# Huge Pages (PostgreSQL, Oracle, etc.)
# vm.nr_hugepages calculation: shared_buffers / 2MB + some headroom
# Example: shared_buffers=8GB -> vm.nr_hugepages = 4200
vm.nr_hugepages = 4200

transparent_hugepage=never

# /etc/sysctl.d/99-container-host.conf

# IP forwarding required
net.ipv4.ip_forward = 1
net.bridge.bridge-nf-call-iptables = 1
net.bridge.bridge-nf-call-ip6tables = 1

# Network
net.core.somaxconn = 65535
net.ipv4.ip_local_port_range = 1024 65535
net.netfilter.nf_conntrack_max = 1048576

# inotify (when running many Pods)
fs.inotify.max_user_watches = 524288
fs.inotify.max_user_instances = 8192

# PID limit
kernel.pid_max = 4194304

# Files
fs.file-max = 2097152

# roles/sysctl/tasks/main.yml
- name: Apply sysctl parameters
  ansible.posix.sysctl:
    name: "{{ item.key }}"
    value: "{{ item.value }}"
    sysctl_file: /etc/sysctl.d/99-tuning.conf
    reload: true
    state: present
  loop: "{{ sysctl_params | dict2items }}"

# roles/sysctl/defaults/main.yml
sysctl_params:
  net.core.somaxconn: 65535
  net.ipv4.tcp_max_syn_backlog: 65535
  vm.swappiness: 10
  fs.file-max: 2097152

#!/usr/bin/env bash
# verify-tuning.sh - Verify tuning values

declare -A EXPECTED=(
  ["net.core.somaxconn"]="65535"
  ["net.ipv4.tcp_congestion_control"]="bbr"
  ["vm.swappiness"]="10"
  ["fs.file-max"]="2097152"
)

FAILED=0
for param in "${!EXPECTED[@]}"; do
  actual=$(sysctl -n "$param" 2>/dev/null)
  expected="${EXPECTED[$param]}"
  if [[ "$actual" != "$expected" ]]; then
    echo "FAIL: $param = $actual (expected: $expected)"
    (( FAILED++ ))
  else
    echo "OK:   $param = $actual"
  fi
done

echo "---"
if (( FAILED > 0 )); then
  echo "Verification failed: ${FAILED} item(s)"
  exit 1
else
  echo "All parameters verified successfully"
fi