#!/usr/bin/env bash
set -euo pipefail

# 2026-05-21 Codex: protect the shared 110 host runner from overlapping
# host-side frontend production builds launched by other repositories.
# 2026-06-27 Codex: make the gate enforce real host pressure too. 110 is both a
# production host and a CI host, so CD must not start a new Docker/Next build
# while load, BuildKit, Gitea Actions, or headless smoke pressure is already high.
# This gate never kills, renices, or rewrites another repo's process tree.
# 2026-06-28 Codex: 110 runner pressure is incident-grade; CD remains fail-hard
# on host pressure until runner work is moved or hard-limited.
# 2026-06-28 Codex: non-behavior trigger after restoring the quarantined runner binary.
# 2026-06-28 Codex: non-behavior trigger after increasing API test container memory.
# 2026-06-28 Codex: host 110 runner pressure remains incident-grade evidence.
# Controlled CD keeps the readback and blocks until non-110 readiness is proved.
# 2026-06-28 Codex: cancel-stale-cd trigger for the pre-guard CD run queue.
# 2026-06-28 Codex: controlled-runtime CD trigger after API test OOM 137.
# 2026-06-28 Codex: old fail-closed pressure guard remains fail-hard in CD.
# 2026-06-28 Codex: controlled-runtime diff detection now uses event payload.
# 2026-06-28 Codex: controlled CD retry after opening 110 systemd guard.
# 2026-06-28 Codex: retry after disabling canonical failclosed enforcer.
# 2026-06-28 Codex: retry after disabling failclosed authority timer.
# 2026-06-28 Codex: retry after restoring controlled drain readback on 110.
# 2026-06-28 Codex: retry after disabling failclosed authority cron source.

ATTEMPTS="${HOST_WEB_BUILD_PRESSURE_ATTEMPTS:-${HOST_WEB_BUILD_PRESSURE_MAX_ATTEMPTS:-60}}"
SLEEP_SECONDS="${HOST_WEB_BUILD_PRESSURE_SLEEP_SECONDS:-${HOST_WEB_BUILD_PRESSURE_INTERVAL:-10}}"
WARN_ONLY="${HOST_WEB_BUILD_PRESSURE_WARN_ONLY:-0}"
MAX_LOAD5_PER_CORE="${HOST_WEB_BUILD_PRESSURE_MAX_LOAD5_PER_CORE:-0.85}"
MAX_CI_CPU_PERCENT="${HOST_WEB_BUILD_PRESSURE_MAX_CI_CPU_PERCENT:-250}"
# One Gitea Actions task container/process group is the current job itself.
# Block only when there is additional CI/BuildKit pressure unless explicitly
# tightened by the runner environment.
MAX_ACTIVE_CI_PROCESS_GROUPS="${HOST_WEB_BUILD_PRESSURE_MAX_ACTIVE_CI_PROCESS_GROUPS:-1}"
MAX_ACTIVE_CI_CONTAINERS="${HOST_WEB_BUILD_PRESSURE_MAX_ACTIVE_CI_CONTAINERS:-1}"
MAX_ORPHAN_BROWSER_GROUPS="${HOST_WEB_BUILD_PRESSURE_MAX_ORPHAN_BROWSER_GROUPS:-0}"
MAX_POSTGRES_CPU_CORES="${HOST_WEB_BUILD_PRESSURE_MAX_POSTGRES_CPU_CORES:-2.0}"
POSTGRES_CONTAINER_NAME="${HOST_WEB_BUILD_PRESSURE_POSTGRES_CONTAINER:-k3s-postgres-recovery}"
METRICS_FILE="${HOST_RUNAWAY_PROCESS_METRICS_FILE:-${HOST_WEB_BUILD_PRESSURE_METRICS_FILE:-/home/wooo/node_exporter_textfiles/host_runaway_process.prom}}"
DEFAULT_DOCKER_METRICS_FILE="/home/$(id -un)/node_exporter_textfiles/docker_stats.prom"
DOCKER_METRICS_FILE="${HOST_WEB_BUILD_PRESSURE_DOCKER_METRICS_FILE:-$DEFAULT_DOCKER_METRICS_FILE}"
EXPORTER="${HOST_RUNAWAY_PROCESS_EXPORTER:-/home/wooo/scripts/host-runaway-process-exporter.py}"

default_ps_command() {
  if ps -eo pid=,ppid=,pcpu=,pmem=,args= --sort=-pcpu >/dev/null 2>&1; then
    printf '%s\n' "ps -eo pid=,ppid=,pcpu=,pmem=,args= --sort=-pcpu"
    return
  fi
  printf '%s\n' "ps -axo pid=,ppid=,pcpu=,pmem=,command="
}

PS_COMMAND="${HOST_WEB_BUILD_PRESSURE_PS_COMMAND:-$(default_ps_command)}"

list_foreign_web_builds() {
  bash -c "$PS_COMMAND" | awk '
    BEGIN { IGNORECASE = 1 }
    function is_diagnostic_command(line) {
      return line ~ /(^|[[:space:]])(grep|rg|awk|sed|perl|find|pgrep|pkill)([[:space:]]|$)/ \
        || line ~ /bash -c .*grep/ \
        || line ~ /bash -c .*rg/ \
        || line ~ /bash -c .*awk/
    }
    /[n]ext[\/[:alnum:]._-]*[[:space:]]+build|[t]urbo[[:space:]]+build|[v]ite[[:space:]]+build/ {
      if (is_diagnostic_command($0)) next
      if ($0 ~ /\/workspace\/wooo\/awoooi/) next
      if ($0 ~ /\/Users\/ogt\/awoooi/) next
      if ($0 ~ /\/private\/tmp\/awoooi/) next
      if ($0 ~ /\/app\/apps\/web/) next
      if ($0 ~ /scripts\/ci\/wait-host-web-build-pressure\.sh/) next
      print
    }
  '
}

refresh_metrics() {
  if [ -x "$EXPORTER" ]; then
    AIOPS_HOST_LABEL="${AIOPS_HOST_LABEL:-110}" \
      NODE_EXPORTER_TEXTFILE_DIR="${NODE_EXPORTER_TEXTFILE_DIR:-/home/wooo/node_exporter_textfiles}" \
      AIOPS_RUNAWAY_PROCESS_MIN_AGE_SECONDS="${AIOPS_RUNAWAY_PROCESS_MIN_AGE_SECONDS:-1800}" \
      AIOPS_RUNAWAY_PROCESS_MIN_CPU_PERCENT="${AIOPS_RUNAWAY_PROCESS_MIN_CPU_PERCENT:-50}" \
      "$EXPORTER" >/dev/null 2>&1 || true
  fi
}

metric_value() {
  local name="$1"
  if [ ! -r "$METRICS_FILE" ]; then
    return 1
  fi
  awk -v metric="$name" '
    $1 ~ ("^" metric "(\\{|$)") {
      value = $NF
    }
    END {
      if (value != "") print value
      else exit 1
    }
  ' "$METRICS_FILE"
}

metric_sum() {
  local name="$1"
  if [ ! -r "$METRICS_FILE" ]; then
    return 1
  fi
  awk -v metric="$name" '
    $1 ~ ("^" metric "(\\{|$)") {
      sum += $NF
      found = 1
    }
    END {
      if (found) print sum
      else exit 1
    }
  ' "$METRICS_FILE"
}

docker_metric_labeled_value() {
  local name="$1"
  local label_key="$2"
  local label_value="$3"
  if [ ! -r "$DOCKER_METRICS_FILE" ]; then
    return 1
  fi
  awk -v metric="$name" -v key="$label_key" -v val="$label_value" '
    $1 ~ ("^" metric "\\{") && $0 ~ (key "=\"" val "\"") {
      value = $NF
    }
    END {
      if (value != "") print value
      else exit 1
    }
  ' "$DOCKER_METRICS_FILE"
}

load5_per_core() {
  metric_value "awoooi_host_load5_per_core" 2>/dev/null || awk '
    BEGIN {
      cores = 0
      while ((getline line < "/proc/cpuinfo") > 0) {
        if (line ~ /^processor[[:space:]]*:/) cores += 1
      }
      close("/proc/cpuinfo")
      if (cores < 1) cores = 1
      if ((getline loadline < "/proc/loadavg") <= 0) exit 1
      split(loadline, parts, " ")
      printf "%.6f\n", parts[2] / cores
    }
  '
}

greater_than() {
  awk -v left="$1" -v right="$2" 'BEGIN { exit !(left > right) }'
}

list_headless_smoke_pressure() {
  bash -c "$PS_COMMAND" | awk '
    BEGIN { IGNORECASE = 1 }
    function is_diagnostic_command(line) {
      return line ~ /(^|[[:space:]])(grep|rg|awk|sed|perl|find|pgrep|pkill)([[:space:]]|$)/ \
        || line ~ /bash -c .*grep/ \
        || line ~ /bash -c .*rg/ \
        || line ~ /bash -c .*awk/
    }
    /[c]hrome.*\/tmp\/stockplatform|[s]tockplatform-[[:alnum:]_-]*smoke|[h]eadless=new/ {
      if (is_diagnostic_command($0)) next
      if ($0 ~ /scripts\/ci\/wait-host-web-build-pressure\.sh/) next
      print
    }
  '
}

pressure_report() {
  local report=""
  local load_ratio active_ci_cpu active_ci_groups active_ci_containers orphan_groups postgres_cpu_cores
  load_ratio="$(load5_per_core 2>/dev/null || echo 0)"
  active_ci_cpu="$(metric_value "awoooi_host_gitea_actions_active_process_cpu_percent" 2>/dev/null || echo 0)"
  active_ci_groups="$(metric_value "awoooi_host_gitea_actions_active_process_group_count" 2>/dev/null || echo 0)"
  active_ci_containers="$(metric_value "awoooi_host_gitea_actions_active_container_count" 2>/dev/null || echo 0)"
  orphan_groups="$(
    metric_sum "awoooi_host_runaway_browser_orphan_group_count" 2>/dev/null \
      || metric_sum "awoooi_host_orphan_browser_group_count" 2>/dev/null \
      || echo 0
  )"
  postgres_cpu_cores="$(
    docker_metric_labeled_value "docker_container_cpu_cores" "container_name" "$POSTGRES_CONTAINER_NAME" 2>/dev/null \
      || echo 0
  )"

  if greater_than "$load_ratio" "$MAX_LOAD5_PER_CORE"; then
    report="${report}host load5/core ${load_ratio} > ${MAX_LOAD5_PER_CORE}"$'\n'
  fi
  if greater_than "$active_ci_cpu" "$MAX_CI_CPU_PERCENT"; then
    report="${report}active CI/BuildKit CPU ${active_ci_cpu}% > ${MAX_CI_CPU_PERCENT}%"$'\n'
  fi
  if greater_than "$active_ci_groups" "$MAX_ACTIVE_CI_PROCESS_GROUPS"; then
    report="${report}active CI/BuildKit process groups ${active_ci_groups} > ${MAX_ACTIVE_CI_PROCESS_GROUPS}"$'\n'
  fi
  if greater_than "$active_ci_containers" "$MAX_ACTIVE_CI_CONTAINERS"; then
    report="${report}active Gitea Actions containers ${active_ci_containers} > ${MAX_ACTIVE_CI_CONTAINERS}"$'\n'
  fi
  if greater_than "$orphan_groups" "$MAX_ORPHAN_BROWSER_GROUPS"; then
    report="${report}orphan browser/smoke groups ${orphan_groups} > ${MAX_ORPHAN_BROWSER_GROUPS}"$'\n'
  fi
  if greater_than "$postgres_cpu_cores" "$MAX_POSTGRES_CPU_CORES"; then
    report="${report}${POSTGRES_CONTAINER_NAME} CPU cores ${postgres_cpu_cores} > ${MAX_POSTGRES_CPU_CORES}"$'\n'
  fi

  local smoke_pressure
  smoke_pressure="$(list_headless_smoke_pressure || true)"
  if [ -n "$smoke_pressure" ]; then
    report="${report}active headless smoke pressure detected"$'\n'"$(printf '%s\n' "$smoke_pressure" | head -n 6)"$'\n'
  fi

  printf '%s' "$report"
}

for attempt in $(seq 1 "$ATTEMPTS"); do
  refresh_metrics
  active_builds="$(list_foreign_web_builds || true)"
  host_pressure="$(pressure_report || true)"
  if [ -z "$active_builds" ] && [ -z "$host_pressure" ]; then
    echo "✅ no host web/build/smoke pressure detected"
    exit 0
  fi

  echo "⏳ host web/build/smoke pressure detected (attempt ${attempt}/${ATTEMPTS}); waiting ${SLEEP_SECONDS}s"
  if [ -n "$host_pressure" ]; then
    printf '%s\n' "$host_pressure" | sed -n '1,12p'
  fi
  if [ -n "$active_builds" ]; then
    printf '%s\n' "$active_builds" | head -n 8
  fi
  if [ "$attempt" -lt "$ATTEMPTS" ]; then
    sleep "$SLEEP_SECONDS"
  fi
done

echo "⚠️ host web/build/smoke pressure still active after ${ATTEMPTS} checks"
if [ "$WARN_ONLY" = "1" ]; then
  echo "⚠️ continuing to avoid a stuck deploy; see ops/runner/README.md for the runner isolation plan"
  exit 0
fi

echo "❌ refusing to start AWOOI image build while host web/build/smoke pressure is still active"
exit 1