stack/scripts/test-federation-acceptance.sh

#!/usr/bin/env bash
# FED R7 / #1802. Scratch acceptance with labelled ancestry stand-in (sage-3625).
# No systemd units. All children run in owned process groups, removed on exit.
set -euo pipefail
R7_INTERRUPT=0
case "${1:-}" in
  '') ;;
  --kill-after-start) R7_INTERRUPT=1 ;;
  *) echo 'usage: scripts/test-federation-acceptance.sh [--kill-after-start]' >&2; exit 2 ;;
esac
[ "$#" -le 1 ] || exit 2
for R7_TOOL in ss rg timeout node ps awk; do
  command -v "$R7_TOOL" >/dev/null || { echo "R7 requires $R7_TOOL" >&2; exit 1; }
done
cd "$(dirname "${BASH_SOURCE[0]}")/.."
R7_ROOT="$(mktemp -d /mnt/storage/scratch/tmp/r7.XXXXXXXX)"
mkdir -p "$R7_ROOT/home" "$R7_ROOT/tmp"
touch "$R7_ROOT/groups" "$R7_ROOT/ports" "$R7_ROOT/sockets"

cleanup() {
  local result=$? pid start raw port socket listeners cleanup_failed=0 attempt
  local -a fields
  trap - EXIT INT TERM
  if [ -n "${R7_DRIVER:-}" ]; then
    kill -TERM "$R7_DRIVER" 2>/dev/null || true
    wait "$R7_DRIVER" 2>/dev/null || true
  fi
  # A detached child is its group's leader. Confirm pid/start time/group before
  # signalling only groups in this run's ledger; never a live stack process.
  while read -r pid start; do
    [ -n "$pid" ] || continue
    if [ -r "/proc/$pid/stat" ]; then
      raw="$(cat "/proc/$pid/stat")"
      read -r -a fields <<< "${raw##*) }"
      if [ "${fields[19]}" != "$start" ] || [ "${fields[2]}" != "$pid" ]; then
        echo "R7 cleanup failed: group identity changed: $pid" >&2
        result=1; cleanup_failed=1
        continue
      fi
      kill -TERM -- "-$pid" 2>/dev/null || true
      for ((attempt=0; attempt<800; attempt++)); do
        ps -eo pgid=,stat= | awk -v g="$pid" '$1==g && $2 !~ /^Z/ {found=1} END {exit !found}' || break
        sleep 0.05
      done
      kill -KILL -- "-$pid" 2>/dev/null || true
      for ((attempt=0; attempt<40; attempt++)); do
        ps -eo pgid=,stat= | awk -v g="$pid" '$1==g && $2 !~ /^Z/ {found=1} END {exit !found}' || break
        sleep 0.05
      done
    fi
    # Zombies hold no tools/listeners and are reaped by their parent. The test
    # requires no live member of an owned group, including orphan descendants.
    if ps -eo pgid=,stat= | awk -v g="$pid" '$1==g && $2 !~ /^Z/ {found=1} END {exit !found}'; then
      echo "R7 cleanup failed: group $pid still has a live process" >&2
      result=1; cleanup_failed=1
    fi
  done < "$R7_ROOT/groups"
  while IFS= read -r port; do
    [ -n "$port" ] || continue
    if ! listeners="$(ss -H -ltn "sport = :$port")"; then
      echo "R7 cleanup failed: cannot inspect port $port" >&2
      result=1; cleanup_failed=1
    elif [ -n "$listeners" ]; then
      echo "R7 cleanup failed: port $port still listens" >&2
      result=1; cleanup_failed=1
    fi
  done < "$R7_ROOT/ports"
  while IFS= read -r socket; do
    [ -n "$socket" ] || continue
    if ! listeners="$(ss -H -lx)"; then
      echo "R7 cleanup failed: cannot inspect socket $socket" >&2
      result=1; cleanup_failed=1
    elif printf '%s\n' "$listeners" | rg -F -- "$socket" >/dev/null; then
      echo "R7 cleanup failed: socket $socket still listens" >&2
      result=1; cleanup_failed=1
    fi
  done < "$R7_ROOT/sockets"
  if listeners="$(ss -H -lx)"; then
    if printf '%s\n' "$listeners" | rg -F -- "$R7_ROOT/" >/dev/null; then
      echo 'R7 cleanup failed: another scratch Unix listener remains' >&2
      result=1; cleanup_failed=1
    fi
  else
    result=1; cleanup_failed=1
  fi
  if [ "$result" -ne 0 ]; then
    for raw in "$R7_ROOT"/*.log; do
      [ -f "$raw" ] || continue
      echo "R7 diagnostic: ${raw##*/}" >&2
      tail -40 "$raw" >&2
    done
  fi
  rm -rf -- "$R7_ROOT"
  if [ -e "$R7_ROOT" ]; then
    echo 'R7 cleanup failed: scratch directory remains' >&2
    result=1; cleanup_failed=1
  fi
  if [ "$cleanup_failed" -eq 0 ]; then echo 'R7 cleanup: no child groups, ports or sockets remain; no units created'; fi
  exit "$result"
}
trap cleanup EXIT
trap 'exit 130' INT
trap 'exit 143' TERM
# No host configuration, preload hooks, credentials or agent env in fixtures.
# The ancestry stand-in lives only in federation-host.mjs, per sage-3625.
timeout -k 10 210 env -i PATH="$PATH" HOME="$R7_ROOT/home" TMPDIR="$R7_ROOT/tmp" \
  MOSAIC_R7_ROOT="$R7_ROOT" MOSAIC_R7_INTERRUPT="$R7_INTERRUPT" \
  node --test --test-concurrency=1 packages/cli/tests/federation-acceptance.test.mjs &
R7_DRIVER=$!
wait "$R7_DRIVER"
R7_DRIVER=