#!/bin/bash
# Consolidated watcher for the 3 live C512 EP runs. nohup'd so it survives session events.
# Logs each run's latest val-CE line + alive/dead every 10 min to runs/watch_all.log.
R=/home/yurenh2/ept/ep_run/runs
LOG=$R/watch_all.log
while true; do
  TS=$(date '+%m-%d %H:%M')
  for f in ep_resreg_warm ep_jacreg ep_rr_ajr; do
    if pgrep -f "ckpt runs/$f.pt" >/dev/null; then a=ALIVE; else a=DEAD; fi
    line=$(grep -iE "val CE" "$R/$f.log" 2>/dev/null | tail -1)
    echo "$TS [$a] $f | $line" >> "$LOG"
  done
  echo "----" >> "$LOG"
  sleep 600
done