cloverleaf-larry/lib/nc-status.sh

#!/usr/bin/env bash
# nc-status.sh — Cloverleaf engine runtime status. Native v3 wrapper around
# the shipped status/tstat binaries — modelled on v1 `status`, `tstats`,
# `tstat`, `thread_status`, `procstatus`, `dstatus`, `connstatus`, `not_up`.
#
# Subcommands:
#   sites       site-level status (daemon + processes + threads)
#   threads     thread-level status (port, state, host)
#   not-up      only threads not in 'up' state
#   connections raw connection state (from connstatus binary if available)
#   queued      threads with messages queued
#   raw         pass-through to the underlying tstat binary
#
# Flags (where applicable):
#   --site SITE         single site; default = all sites under $HCIROOT
#   --filter REGEX      match on thread name
#   --format text|tsv   default text
#
# This is V1-modeled but native-bash. The exact binary names depend on the
# Cloverleaf version; we auto-discover under $HCIROOT/bin (and server/bin).
set -o pipefail

NC_SELF="$0"

die() { printf 'nc-status: %s\n' "$*" >&2; exit 1; }
warn() { printf 'nc-status: %s\n' "$*" >&2; }

resolve_binary() {
  local name="$1"
  if command -v "$name" >/dev/null 2>&1; then command -v "$name"; return; fi
  for d in "${HCIROOT:-}/bin" "${HCIROOT:-}/server/bin"; do
    [ -x "$d/$name" ] && { echo "$d/$name"; return; }
  done
  return 1
}

list_sites() {
  local root="${HCIROOT:-}"
  [ -d "$root" ] || die "no \$HCIROOT or it doesn't exist"
  find "$root" -mindepth 1 -maxdepth 2 -name NetConfig -type f 2>/dev/null \
    | xargs -I{} dirname {} 2>/dev/null \
    | xargs -I{} basename {} 2>/dev/null \
    | sort -u
}

cmd_sites() {
  local target_site=""
  local format="text"
  while [ $# -gt 0 ]; do
    case "$1" in
      --site)   shift; target_site="$1" ;;
      --format) shift; format="$1" ;;
      *)        die "unknown flag: $1" ;;
    esac
    shift
  done
  local hcienginestat; hcienginestat=$(resolve_binary hcienginestat 2>/dev/null || true)
  local sites_to_check
  if [ -n "$target_site" ]; then sites_to_check="$target_site"; else sites_to_check=$(list_sites); fi
  [ "$format" = "tsv" ] && printf 'site\tstate\tdetail\n'
  while IFS= read -r s; do
    [ -z "$s" ] && continue
    local detail="" state="?"
    if [ -n "$hcienginestat" ]; then
      detail=$(HCISITE="$s" "$hcienginestat" 2>&1 || true)
      if echo "$detail" | grep -qiE 'running|active|up'; then state="up"
      elif echo "$detail" | grep -qiE 'stopped|down'; then state="down"
      else state="unknown"; fi
    else
      # Fallback: check for lock file / pid file as a poor-man's check
      if [ -e "${HCIROOT:-}/$s/lock" ] || [ -e "${HCIROOT:-}/$s/exec/processes" ]; then
        state="config-present"
      fi
    fi
    if [ "$format" = "tsv" ]; then
      printf '%s\t%s\t%s\n' "$s" "$state" "$(echo "$detail" | head -1)"
    else
      printf '== %-20s [%s] ==\n%s\n' "$s" "$state" "$(echo "$detail" | head -10)"
    fi
  done <<< "$sites_to_check"
}

cmd_threads() {
  local target_site="${HCISITE:-}"
  local filter=""
  local format="text"
  while [ $# -gt 0 ]; do
    case "$1" in
      --site)   shift; target_site="$1" ;;
      --filter) shift; filter="$1" ;;
      --format) shift; format="$1" ;;
      *)        die "unknown flag: $1" ;;
    esac
    shift
  done
  local tstat; tstat=$(resolve_binary tstat) || die "tstat binary not found under \$HCIROOT/bin or PATH"
  local args=()
  [ -n "$target_site" ] && args+=("-s" "$target_site")
  local output; output=$(HCISITE="$target_site" "$tstat" "${args[@]}" 2>&1 || true)
  if [ -n "$filter" ]; then
    output=$(printf '%s' "$output" | grep -E -- "$filter")
  fi
  if [ "$format" = "tsv" ]; then
    # Best-effort: parse the tstat output into TSV
    printf '%s\n' "$output" | awk 'NR>1 && NF>0 { gsub(/  +/, "\t"); print }'
  else
    printf '%s\n' "$output"
  fi
}

cmd_not_up() {
  local site="${HCISITE:-}"
  local filter=""
  local format="text"
  while [ $# -gt 0 ]; do
    case "$1" in
      --site)   shift; site="$1" ;;
      --filter) shift; filter="$1" ;;
      # Accept --format and forward it to cmd_threads. The larry.sh dispatch
      # ALWAYS appends --format, so without this not-up died "unknown flag:
      # --format" and was unusable via the wired tool.
      --format) shift; format="$1" ;;
      *)        die "unknown flag: $1" ;;
    esac
    shift
  done
  cmd_threads --site "$site" --format "$format" ${filter:+--filter "$filter"} \
    | awk 'NR==1 || tolower($0) !~ /(^|[^a-z])up([^a-z]|$)/'
}

cmd_connections() {
  local connstatus; connstatus=$(resolve_binary connstatus 2>/dev/null) || die "connstatus binary not found"
  "$connstatus" "$@"
}

cmd_queued() {
  local site="${HCISITE:-}"
  cmd_threads --site "$site" "$@" \
    | awk 'NR==1 || tolower($0) ~ /queued|obq|ibq/'
}

cmd_raw() {
  local tstat; tstat=$(resolve_binary tstat) || die "tstat binary not found"
  "$tstat" "$@"
}

SUB="${1:-sites}"
case "$SUB" in
  sites)       shift; cmd_sites "$@" ;;
  threads)     shift; cmd_threads "$@" ;;
  not-up)      shift; cmd_not_up "$@" ;;
  connections) shift; cmd_connections "$@" ;;
  queued)      shift; cmd_queued "$@" ;;
  raw)         shift; cmd_raw "$@" ;;
  help|-h|--help) sed -n '2,20p' "$NC_SELF" ;;
  *)           die "unknown subcommand: $SUB" ;;
esac