#!/bin/bash
# deploy-guest-perfpack.sh — guest-side perf pack (ticket #737)
# Run INSIDE a guest VM: bash tests/remote.sh vm-file proxmox/perf/scripts/deploy-guest-perfpack.sh
# Idempotent. Applies: sysctl tuning (swappiness/dirty/net buffers/slow-start),
# fq + bbr (when the kernel module exists), THP->madvise (skipped when tuned
# manages THP, e.g. k8s cnodes on network-latency), fstrim.timer.
# Pairs with the host-side qm flags (iothread/ssd=1/discard=on).
set -u

log() { echo "[perfpack] $*"; }

# --- 1. sysctl profile ---
bbr=0
if modprobe tcp_bbr 2>/dev/null && grep -qw bbr /proc/sys/net/ipv4/tcp_available_congestion_control 2>/dev/null; then
    bbr=1
fi
conf=/etc/sysctl.d/99-perfopt.conf
{
    echo "# perf pack #737 (idempotent redeploy overwrites)"
    echo "vm.swappiness = 10"
    echo "vm.dirty_ratio = 10"
    echo "vm.dirty_background_ratio = 3"
    echo "net.core.rmem_max = 4194304"
    echo "net.core.wmem_max = 4194304"
    echo "net.core.rmem_default = 262144"
    echo "net.core.wmem_default = 262144"
    echo "net.ipv4.tcp_slow_start_after_idle = 0"
    echo "net.core.default_qdisc = fq"
    if [ "$bbr" = 1 ]; then
        echo "net.ipv4.tcp_congestion_control = bbr"
    fi
} > "$conf"
sysctl --system >/dev/null 2>&1
log "sysctl applied ($conf): swappiness=$(sysctl -n vm.swappiness 2>/dev/null) cc=$(sysctl -n net.ipv4.tcp_congestion_control 2>/dev/null) qdisc=$(sysctl -n net.core.default_qdisc 2>/dev/null) bbr=$bbr"

# --- 2. THP -> madvise (live now + persisted via oneshot unit) ---
tuned_active=0
if command -v tuned-adm >/dev/null 2>&1; then
    tuned-adm active 2>/dev/null | grep -qvi 'no current\|inactive\|none' && tuned_active=1
fi
if [ "$tuned_active" = 1 ]; then
    log "THP: tuned profile active ($(tuned-adm active 2>/dev/null | awk '{print $NF}')) — leaving THP to tuned"
else
    thp=/sys/kernel/mm/transparent_hugepage/enabled
    cur=$(tr '[]' '  ' < "$thp" 2>/dev/null | awk '{print $NF}')
    if [ "$cur" != "madvise" ]; then
        if echo madvise > "$thp" 2>/dev/null; then
            log "THP: ${cur:-unknown} -> madvise (live)"
        else
            log "THP: WARNING live set failed (check $thp permissions)"
        fi
    else
        log "THP: already madvise"
    fi
    unit=/etc/systemd/system/thp-madvise.service
    if [ ! -f "$unit" ]; then
        {
            echo "[Unit]"
            echo "Description=Set transparent hugepages to madvise (perf pack #737)"
            echo "DefaultDependencies=no"
            echo "After=local-fs.target"
            echo "Before=sysinit.target"
            echo
            echo "[Service]"
            echo "Type=oneshot"
            echo "ExecStart=/bin/sh -c 'echo madvise > /sys/kernel/mm/transparent_hugepage/enabled'"
            echo
            echo "[Install]"
            echo "WantedBy=sysinit.target"
        } > "$unit"
        systemctl daemon-reload
        systemctl enable thp-madvise.service >/dev/null 2>&1
        log "THP: persistence unit installed + enabled"
    fi
fi

# --- 3. fstrim (pairs with host-side discard=on) ---
if systemctl enable --now fstrim.timer >/dev/null 2>&1; then
    log "fstrim.timer enabled"
else
    log "fstrim.timer: not available on this guest"
fi

log "done"