#!/usr/bin/env bash
# Overnight corpus: every size, smallest first. Self-recovering, load-guarded.
set -uo pipefail
export PATH="$HOME/.cargo/bin:$PATH"
export SDS_IMAGE=lance-blue/sds-runner-report
cd "$(dirname "$0")" || exit 1
BOT=/work/target/release/sds-bot
LOG=overnight.log
say() { echo "[$(date +%H:%M:%S)] $*" | tee -a "$LOG"; }

# Wait until the machine can take another phase. Never proceeds blindly.
settle() {
  for _ in $(seq 1 120); do
    load=$(awk '{print int($1)}' /proc/loadavg)
    free=$(free -g | awk '/^Mem:/{print $7}')
    [ "$load" -lt 8 ] && [ "$free" -ge 4 ] && return 0
    say "waiting: load $load, ${free}G free"
    sleep 60
  done
  say "machine never settled; going ahead at low concurrency anyway"
}

phase() {
  size=$1; games=$2
  say "=== $size: $games games ==="
  for attempt in 1 2 3; do
    settle
    if ./sds.sh bench --suite "suites/$size" --games "$games" --jobs 2 \
         --bot "$BOT" --candidates \
         --save-baseline "overnight-$size" \
         --notes "generated-only, uncapped, ceiling denominator" >>"$LOG" 2>&1; then
      say "$size finished"
      return 0
    fi
    say "$size attempt $attempt failed; retrying after a pause"
    sleep 120
  done
  say "$size failed three times; moving on rather than stopping the night"
}

say "start: $(uptime)"
phase 1v1 180
phase 2v2 180
phase 4v4 150
phase 8v8 120
say "all phases done"