#!/usr/bin/env bash
# 010-tune-postgres - Memory & stability tuning for the 1GB e2-micro host
# (B-tier of the 2026-08 tuning review). Settings land as a conf.d drop-in;
# postgresql.conf itself stays pristine (include_dir = 'conf.d' is stock).
#
#   max_connections        100 -> 30     each backend costs MBs; real
#                                        concurrency comes from PgBouncer
#                                        (013). 30 = pool 20 + admin headroom
#   effective_cache_size    4GB -> 768MB  planner hint; box has 1GB, the 4GB
#                                        default overrates cache -> bad plans
#   work_mem                4MB (explicit, equals default; kept for determinism)
#   statement_timeout       0 -> 30s      runaway-query guard; long migrations:
#                                        SET statement_timeout = 0; in-session
#   idle_in_transaction_session_timeout 0 -> 60s   leaked transactions block
#                                        vacuum and hold locks; kill them
#   temp_file_limit         -1 -> 2GB     per-backend sort-spill cap so one bad
#                                        query cannot fill the 20G data disk
# up:    write drop-in; restart only if max_connections changes, else reload
# down:  remove drop-in; restart (restoring max_connections needs it)
set -euo pipefail

PGBIN=/etc/postgresql/18/main
CONF=$PGBIN/conf.d/010-tune-postgres.conf
CLUSTER=$(pg_lsclusters --no-header | awk 'NR==1{print $1"/"$2}')

# desired state as "guc value" pairs; SHOW renders exactly these strings
WANT=(
  'max_connections 30'
  'effective_cache_size 768MB'
  'work_mem 4MB'
  'statement_timeout 30s'
  'idle_in_transaction_session_timeout 60s'
  'temp_file_limit 2GB'
)

pg() { runuser -u postgres -- psql -d postgres -v ON_ERROR_STOP=1 -tAc "$1"; }

write_conf() {
  install -d -m 755 "$PGBIN/conf.d"
  cat > "$CONF" <<'EOF'
# managed by sysmig 010-tune-postgres - memory & stability for 1GB e2-micro
max_connections = 30
effective_cache_size = 768MB
work_mem = 4MB
statement_timeout = 30s
idle_in_transaction_session_timeout = 60s
temp_file_limit = 2GB
EOF
  chown root:postgres "$CONF"
  chmod 644 "$CONF"
}

case "${1:-}" in
  up)
    write_conf
    if [[ $(pg 'SHOW max_connections') != 30 ]]; then
      pg_ctlcluster "$CLUSTER" restart
      pg_isready -q
      echo "    cluster restarted (max_connections requires it)"
    else
      pg_ctlcluster "$CLUSTER" reload
      echo "    cluster reloaded"
    fi
    for kv in "${WANT[@]}"; do
      read -r guc want <<<"$kv"
      got=$(pg "SHOW $guc")
      [[ $got == "$want" ]] || { echo "FATAL: $guc is '$got', expected '$want'" >&2; exit 1; }
    done
    echo "    ${#WANT[@]} memory/stability settings active"
    ;;
  status)
    if [[ $EUID -eq 0 ]]; then
      q() { runuser -u postgres -- psql -d postgres -v ON_ERROR_STOP=1 -tAc "$1"; }
    elif psql -d postgres -tAc 'SELECT 1' >/dev/null 2>&1; then
      q() { psql -d postgres -v ON_ERROR_STOP=1 -tAc "$1"; }
    else
      echo "? needs root or peer DB access to verify"
      exit 2
    fi
    [[ -f $CONF ]] || { echo "DRIFT drop-in $CONF missing"; exit 1; }
    bad=()
    for kv in "${WANT[@]}"; do
      read -r guc want <<<"$kv"
      got=$(q "SHOW $guc" 2>/dev/null)
      [[ $got == "$want" ]] || bad+=("$guc=$got(want $want)")
    done
    if ((${#bad[@]})); then echo "DRIFT ${bad[*]}"; exit 1; fi
    echo "OK ${#WANT[@]} settings active (max_connections=30, timeouts, temp_file_limit)"
    ;;
  down)
    rm -f "$CONF"
    if [[ $(pg 'SHOW max_connections') == 30 ]]; then
      pg_ctlcluster "$CLUSTER" restart
      echo "    drop-in removed, cluster restarted (defaults restored)"
    else
      pg_ctlcluster "$CLUSTER" reload
      echo "    drop-in removed, cluster reloaded"
    fi
    ;;
esac