|
| 1 | +#!/bin/bash |
| 2 | +# Run all variance rerun waves sequentially (designed for overnight unattended execution). |
| 3 | +# Each wave adds 1 run per task toward the target of 3 runs per config. |
| 4 | +# |
| 5 | +# Usage: |
| 6 | +# nohup ./configs/run_variance_waves.sh > logs/variance_waves_$(date +%Y%m%d_%H%M%S).log 2>&1 & |
| 7 | +# |
| 8 | +# To start from a specific wave (e.g., after wave 1 already ran): |
| 9 | +# nohup ./configs/run_variance_waves.sh --start-wave 2 > logs/variance_waves_$(date +%Y%m%d_%H%M%S).log 2>&1 & |
| 10 | + |
| 11 | +set -e |
| 12 | + |
| 13 | +SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)" |
| 14 | +REPO_ROOT="$SCRIPT_DIR/.." |
| 15 | +cd "$REPO_ROOT" |
| 16 | + |
| 17 | +START_STEP=1 |
| 18 | +while [[ $# -gt 0 ]]; do |
| 19 | + case $1 in |
| 20 | + --start-step) START_STEP="$2"; shift 2 ;; |
| 21 | + *) echo "Unknown option: $1"; exit 1 ;; |
| 22 | + esac |
| 23 | +done |
| 24 | + |
| 25 | +WAVE_DIR="configs/variance_reruns" |
| 26 | +RUNNER="./configs/run_selected_tasks.sh" |
| 27 | +LOG_DIR="logs" |
| 28 | +mkdir -p "$LOG_DIR" |
| 29 | + |
| 30 | +STARTED_AT=$(date '+%Y-%m-%d %H:%M:%S') |
| 31 | +echo "==============================================" |
| 32 | +echo "Variance Rerun — All Waves" |
| 33 | +echo "==============================================" |
| 34 | +echo "Started: $STARTED_AT" |
| 35 | +echo "Start step: $START_STEP" |
| 36 | +echo "" |
| 37 | + |
| 38 | +# Each entry: wave_num selection_file flags |
| 39 | +STEPS=( |
| 40 | + "1 wave1_both.json " |
| 41 | + "1 wave1_baseline_only.json --baseline-only" |
| 42 | + "1 wave1_mcp_only.json --full-only" |
| 43 | + "2 wave2_repoqa.json " |
| 44 | + "2 wave2_remainder.json " |
| 45 | + "2 wave2_baseline_only.json --baseline-only" |
| 46 | + "2 wave2_mcp_only.json --full-only" |
| 47 | + "3 wave3_both.json " |
| 48 | + "3 wave3_baseline_only.json --baseline-only" |
| 49 | +) |
| 50 | + |
| 51 | +TOTAL=${#STEPS[@]} |
| 52 | +STEP_NUM=0 |
| 53 | +FAILED=0 |
| 54 | + |
| 55 | +for entry in "${STEPS[@]}"; do |
| 56 | + read -r wave_num selection_file flags <<< "$entry" |
| 57 | + STEP_NUM=$((STEP_NUM + 1)) |
| 58 | + |
| 59 | + # Skip steps below start |
| 60 | + if [ "$STEP_NUM" -lt "$START_STEP" ]; then |
| 61 | + echo "[Step $STEP_NUM/$TOTAL] SKIP (step $STEP_NUM < start_step $START_STEP)" |
| 62 | + continue |
| 63 | + fi |
| 64 | + |
| 65 | + selection_path="${WAVE_DIR}/${selection_file}" |
| 66 | + if [ ! -f "$selection_path" ]; then |
| 67 | + echo "[Step $STEP_NUM/$TOTAL] SKIP — $selection_path not found" |
| 68 | + continue |
| 69 | + fi |
| 70 | + |
| 71 | + # Check if selection file has any tasks |
| 72 | + task_count=$(python3 -c "import json; print(len(json.load(open('$selection_path')).get('tasks',[])))") |
| 73 | + if [ "$task_count" -eq 0 ]; then |
| 74 | + echo "[Step $STEP_NUM/$TOTAL] SKIP — $selection_file has 0 tasks" |
| 75 | + continue |
| 76 | + fi |
| 77 | + |
| 78 | + step_log="${LOG_DIR}/wave${wave_num}_${selection_file%.json}_$(date +%Y%m%d_%H%M%S).log" |
| 79 | + |
| 80 | + echo "" |
| 81 | + echo "==============================================" |
| 82 | + echo "[Step $STEP_NUM/$TOTAL] Wave $wave_num — $selection_file ($task_count tasks) $flags" |
| 83 | + echo " Log: $step_log" |
| 84 | + echo " Started: $(date '+%Y-%m-%d %H:%M:%S')" |
| 85 | + echo "==============================================" |
| 86 | + |
| 87 | + # Pipe newline to auto-confirm the interactive prompt. |
| 88 | + # Tee to both step log and stdout so the wrapper log captures everything. |
| 89 | + if echo "" | $RUNNER \ |
| 90 | + --selection-file "$selection_path" \ |
| 91 | + --category staging \ |
| 92 | + $flags \ |
| 93 | + 2>&1 | tee "$step_log"; then |
| 94 | + echo "[Step $STEP_NUM/$TOTAL] DONE at $(date '+%Y-%m-%d %H:%M:%S')" |
| 95 | + else |
| 96 | + echo "[Step $STEP_NUM/$TOTAL] FAILED (exit $?) at $(date '+%Y-%m-%d %H:%M:%S')" |
| 97 | + FAILED=$((FAILED + 1)) |
| 98 | + # Continue to next step — don't abort the whole overnight run |
| 99 | + fi |
| 100 | +done |
| 101 | + |
| 102 | +echo "" |
| 103 | +echo "==============================================" |
| 104 | +echo "All waves complete" |
| 105 | +echo " Started: $STARTED_AT" |
| 106 | +echo " Finished: $(date '+%Y-%m-%d %H:%M:%S')" |
| 107 | +echo " Failed steps: $FAILED / $TOTAL" |
| 108 | +echo "==============================================" |
0 commit comments