#!/bin/bash
# Runs the whole CTS query list (list-queries.sh) under dawn.node on lavapipe,
# split round-robin across N shards (default: one per core) so a slow region
# of the suite (texture sampling, say) spreads across shards instead of
# landing in one. Each shard runs through run-shard.sh in its own background
# process; one hung or crashed query only stalls its own shard. Results land
# under results/dawn-node/<shard>/ (gitignored, never committed).
#
# Resumable as long as shard_count stays the same across runs: the round-robin
# assignment is deterministic, so each query lands in the same shard
# directory every time, and run-shard.sh skips what that shard already
# finished. A different shard_count reassigns queries to new directories and
# redoes them; it does not corrupt anything.
#
# Usage: run-dawn-node.sh [shard_count]
set -u
SELF_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
SHARDS="${1:-$(nproc)}"
OUT_ROOT="$SELF_DIR/results/dawn-node"
mkdir -p "$OUT_ROOT"

"$SELF_DIR/list-queries.sh" > "$OUT_ROOT/queries.txt"
total=$(wc -l < "$OUT_ROOT/queries.txt")

# Round-robin, not contiguous ranges: query N goes to shard N mod SHARDS.
for i in $(seq 0 $((SHARDS - 1))); do
  id=$(printf '%02d' "$i")
  mkdir -p "$OUT_ROOT/$id"
  awk -v n="$SHARDS" -v i="$i" 'NR % n == i' "$OUT_ROOT/queries.txt" > "$OUT_ROOT/$id/queries.txt"
done

if [ ! -f "$OUT_ROOT/run-manifest.txt" ]; then
  echo "start=$(date -Is)" > "$OUT_ROOT/run-manifest.txt"
fi
{
  echo "total_queries=$total"
  echo "shards=$SHARDS"
} >> "$OUT_ROOT/run-manifest.txt"

pids=()
for i in $(seq 0 $((SHARDS - 1))); do
  id=$(printf '%02d' "$i")
  "$SELF_DIR/run-shard.sh" "$OUT_ROOT/$id" < "$OUT_ROOT/$id/queries.txt" &
  pids+=("$!")
done
for pid in "${pids[@]}"; do
  wait "$pid"
done

echo "end=$(date -Is)" >> "$OUT_ROOT/run-manifest.txt"