public-tools/EPICS2SMS.sh

#!/usr/bin/env bash

# --- configuration ---------------------------------------------------------

# Example of an enum PV (typically returned with quotes by caget, e.g. "\"ON\""):
# PV_NAME="X12SA-FE-VMMG-0010:PLC_RELAY-D"

PV_NAME="AGEBD-PARAMS:INJECTION-RATE" # EPICS PV to monitor
PHONE_NUMBER="0041793083005"          # phone number(s), space separated (may be empty)
EMAIL="andreas.menzel@psi.ch"         # email address(es), space separated (may be empty)
POLL_INTERVAL=5                       # seconds between polls while Running
ERROR_INTERVAL=60                     # seconds between polls while NOT Running

# Define what "Running" means for the PV:
#   CHECK_MODE="numeric": Running if numeric PV value > OK_THRESHOLD
#   CHECK_MODE="enum":    Running if PV string equals OK_STATE exactly
CHECK_MODE="numeric" # numeric | enum
OK_THRESHOLD=0       # used when CHECK_MODE=numeric
OK_STATE="\"ON\""    # used when CHECK_MODE=enum (match exact caget output)

# Notification delivery
SENDMAIL="/usr/sbin/sendmail"
SMS_GATEWAY_DOMAIN="sms.switch.ch"

# Set to 1 to print the email payload instead of sending it (useful for testing)
DRY_RUN=0

# --- helpers ---------------------------------------------------------------

log() {
  printf '[%s] %s\n' "$(date '+%F %T')" "$*"
}

require_command() {
  local cmd="$1"
  command -v "$cmd" >/dev/null 2>&1 || {
    log "ERROR: required command not found: $cmd"
    exit 2
  }
}

is_running() {
  # Decide whether a PV value should be treated as "Running".
  # Returns:
  #   0 => Running
  #   1 => Not Running
  #   2 => Configuration error
  local v="$1"

  case "$CHECK_MODE" in
  numeric)
    # IMPORTANT: pass PV value as data (-v v="...") rather than splicing it into awk code.
    # (v+0) forces numeric conversion so non-numeric values become 0.
    awk -v v="$v" -v t="$OK_THRESHOLD" 'BEGIN { exit !((v+0) > (t+0)) }'
    ;;
  enum)
    [[ "$v" == "$OK_STATE" ]]
    ;;
  *)
    log "ERROR: unknown CHECK_MODE='$CHECK_MODE' (expected 'numeric' or 'enum')"
    return 2
    ;;
  esac
}

send_notification() {
  # This sends to:
  #   - phone numbers via <number>@sms.switch.ch
  #   - email addresses directly
  #
  # Usage: send_notification "<phone numbers>" "<email addresses>" "<message>"
  local numbers="$1 $2" # combine PHONE_NUMBER and EMAIL
  local message="$3"

  local subject="[EPICS] $PV_NAME"
  local -a to_lines=()

  # Build To: header lines
  local -a recipients=()
  read -r -a recipients <<<"$numbers"

  local tok
  for tok in "${recipients[@]}"; do
    if [[ "$tok" == *"@"* ]]; then
      to_lines+=("To: $tok")
    else
      to_lines+=("To: ${tok}@${SMS_GATEWAY_DOMAIN}")
    fi
  done

  if ((${#to_lines[@]} == 0)); then
    log "ERROR: no recipients configured (PHONE_NUMBER and EMAIL are empty)"
    return 3
  fi

  # Append a timestamp
  local message_with_date
  message_with_date="$message"$'\n'"$(date)"

  # Choose send command (real vs dry-run)
  local -a send_cmd=("$SENDMAIL" -t)
  ((DRY_RUN)) && send_cmd=(cat)

  {
    printf '%s\n' "${to_lines[@]}"
    printf 'Subject: %s\n' "$subject"
    printf '\n'
    printf '%s\n' "$message_with_date"
  } | "${send_cmd[@]}"
}

# --- main loop ------------------------------------------------------------

require_command caget

# Basic startup sanity checks
if [[ -z "${PHONE_NUMBER// /}" && -z "${EMAIL// /}" ]]; then
  log "ERROR: at least one recipient must be configured (PHONE_NUMBER and/or EMAIL)."
  exit 2
fi

if ((!DRY_RUN)); then
  if [[ ! -x "$SENDMAIL" ]]; then
    log "ERROR: SENDMAIL='$SENDMAIL' is not executable (set DRY_RUN=1 to test without sendmail)."
    exit 2
  fi
fi

alert_sent=0 # 0 = no alert sent for current outage, 1 = already sent

log "Starting EPICS monitor for PV '$PV_NAME' (CHECK_MODE=$CHECK_MODE)..."

while true; do
  # Get PV value as a plain string; adjust flags if needed (-S for string PVs)
  value=$(caget -noname -nounit "$PV_NAME" 2>/dev/null)
  status=$?

  # If caget fails, treat that as "not Running" (e.g. IOC down)
  if [[ $status -ne 0 ]]; then
    log "WARNING: caget failed for '$PV_NAME' (exit $status)"
    value="UNAVAILABLE (caget exit $status)"
  fi

  if is_running "$value"; then
    # System is OK again
    if [[ $alert_sent -eq 1 ]]; then
      log "PV '$PV_NAME' back to Running (value='$value')."
    fi
    alert_sent=0
    sleep "$POLL_INTERVAL"
  else
    rc=$?
    if [[ $rc -eq 2 ]]; then
      log "ERROR: stopping due to configuration error in CHECK_MODE."
      exit 2
    fi

    # System not OK
    if [[ $alert_sent -eq 0 ]]; then
      msg="Alert: PV $PV_NAME is '$value'."
      if send_notification "$PHONE_NUMBER" "$EMAIL" "$msg"; then
        alert_sent=1
        log "Alert sent."
      else
        log "ERROR: failed to send alert; will retry on next check."
      fi
    fi
    sleep "$ERROR_INTERVAL"
  fi
done