code/federation/side_cycle_watchdog.sh

#!/bin/bash
# Auto-restart the side_cycle if it ever exits (set -e crash, etc). The script
# itself loops internally; this catches hard exits so the distributed dispatcher
# never silently stays dead.
export AGILLM41_SAVE_DIR="${AGILLM41_SAVE_DIR:-/workspace/agillm4_4090_ckpts_active}"
export AGILLM41_MAINLINE="${AGILLM41_MAINLINE:-/workspace/agillm41-mainline}"
export AGILLM41_GETH_KEY="${AGILLM41_GETH_KEY:-/root/.ssh/agillm41_geth_ed25519}"
export AGILLM41_SIDE_CYCLE_SEC="${AGILLM41_SIDE_CYCLE_SEC:-900}" AGILLM41_SIDE_THREADS=8 AGILLM41_SMALL_NODE_THREADS=2 AGILLM41_SIDE_KEEP_ROUNDS=3
while true; do
  echo "{\"event\":\"watchdog_launch\",\"at\":\"$(date -u +%FT%TZ)\"}" >> /workspace/agillm41_side_cycle.log
  bash /workspace/agillm41_vast_side_cycle.sh >> /workspace/agillm41_side_cycle.log 2>&1
  echo "{\"event\":\"watchdog_side_cycle_exited_restarting_30s\",\"at\":\"$(date -u +%FT%TZ)\"}" >> /workspace/agillm41_side_cycle.log
  sleep 30
done