#!/bin/bash
# Coordinator script: launch master on this node + worker on h100-b
set -euo pipefail

SCRIPT=/tmp/zillion2/ruoliu/Monet/script_examples/sft_no_text_stage2_8gpu_multinode.sh
MASTER_LOG=/tmp/zillion2/ruoliu/logs/sft_stage2_master.log
WORKER_LOG=/tmp/zillion2/ruoliu/logs/sft_stage2_worker.log

# Clean previous artifacts
rm -rf /tmp/zillion2/ruoliu/ckpts/Monet_no_text_ckpts/stage2_notext_latent8_ce4.0_al2.0_emph2.0
ssh h100-b 'rm -rf /tmp/zillion2/ruoliu/ckpts/Monet_no_text_ckpts/stage2_notext_latent8_ce4.0_al2.0_emph2.0'

# Move old logs aside
mv "$MASTER_LOG" "${MASTER_LOG}.prev" 2>/dev/null || true
mv "$WORKER_LOG" "${WORKER_LOG}.prev" 2>/dev/null || true
ssh h100-b "mkdir -p /tmp/zillion2/ruoliu/logs"

# Start worker on h100-b first (it will wait for master)
echo "Launching worker on h100-b..."
ssh -f h100-b "NODE_RANK=1 MASTER_ADDR=10.89.17.7 nohup bash $SCRIPT > $WORKER_LOG 2>&1 < /dev/null &"

# Start master here
echo "Launching master on $(hostname)..."
NODE_RANK=0 MASTER_ADDR=10.89.17.7 nohup bash "$SCRIPT" > "$MASTER_LOG" 2>&1 < /dev/null &
echo "MASTER_PID=$!"
echo "Both launched. Tail $MASTER_LOG and (on h100-b) $WORKER_LOG."