#!/usr/bin/env bash
# k3-test/run_all.sh — full runbook driver (plan §4 Phases 0–3).
# Usage:
#   tmux new -s k3
#   ./run_all.sh            # everything, K3 download starts in background immediately
#   SKIP_DOWNLOAD=1 ./run_all.sh   # if K3 already downloaded
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
source "$SCRIPT_DIR/lib/common.sh"

trap 'stop_monitors; unfill_ram' EXIT

info "k3-test run starting. Logs: $LOG_ROOT"

# ---- Phase 0: setup + kick the big download OFF in the background ------------
"$SCRIPT_DIR/setup.sh" 2>&1 | tee "$LOG_ROOT/setup.log"

if [[ -z "${SKIP_DOWNLOAD:-}" ]]; then
  info "launching K3 download in background ($K3_INCLUDE, ~1.5 TB)…"
  nohup "$SCRIPT_DIR/01_download_k3.sh" > "$LOG_ROOT/download_k3.log" 2>&1 &
  DL_PID=$!
  echo "$DL_PID" > "$LOG_ROOT/download_k3.pid"
fi

"$SCRIPT_DIR/02_download_smoke.sh" 2>&1 | tee -a "$LOG_ROOT/download_smoke.log"

# ---- Phase 0b: recon (free, while downloading) --------------------------------
"$SCRIPT_DIR/00_recon.sh" 2>&1 | tee "$LOG_ROOT/recon_driver.log"

# ---- Phase 1: smoke gate -------------------------------------------------------
if ! "$SCRIPT_DIR/10_smoke.sh" 2>&1 | tee "$LOG_ROOT/smoke_driver.log"; then
  die "SMOKE GATE FAILED — investigate logs in $LOG_ROOT before spending more"
fi

# ---- gate: wait for K3 ----------------------------------------------------------
if [[ -n "${DL_PID:-}" ]]; then
  info "waiting for K3 download (pid $DL_PID) — tail -f $LOG_ROOT/download_k3.log to watch"
  while kill -0 "$DL_PID" 2>/dev/null; do
    du -sh "${MODEL_DIR:-$HOME/models}/Kimi-K3-GGUF" 2>/dev/null | tr '\n' ' '
    df -h . | tail -1 | awk '{print "  free:",$4}'
    sleep 120
  done
  wait "$DL_PID" || die "K3 download FAILED — check $LOG_ROOT/download_k3.log"
fi

# ---- Phase 2: experiments -------------------------------------------------------
"$SCRIPT_DIR/20_k3_experiments.sh" 2>&1 | tee "$LOG_ROOT/experiments_driver.log"

# ---- Phase 3: pack --------------------------------------------------------------
"$SCRIPT_DIR/30_teardown.sh"

cat <<EOF

================================================================================
RUN COMPLETE. Sync $LOG_ROOT and the tarball off-box, then DESTROY the instance.
Report against k3-rental-test-plan.md §6 success criteria.
================================================================================
EOF