feat: add docker image cleanup script for benchmark images

bash/cleanup_docker_images.sh

@@ -0,0 +1,205 @@
+#!/bin/bash
+# ============================================================
+# 清理评测后遗留的 Docker 镜像
+#
+# 用途：
+#   SWE-bench、Terminal-Bench 等 benchmark 会拉取大量 Docker 镜像，
+#   评测完后很占磁盘。本脚本按命名空间批量删除这些镜像。
+#
+# 用法：
+#   # 默认只预览，不会真的删除
+#   bash bash/cleanup_docker_images.sh --preset swe
+#
+#   # 确认删除
+#   bash bash/cleanup_docker_images.sh --preset swe --yes
+#
+#   # 删除 terminal-bench 镜像
+#   bash bash/cleanup_docker_images.sh --preset terminal --yes
+#
+#   # 删除所有常见 benchmark 镜像
+#   bash bash/cleanup_docker_images.sh --preset all-bench --yes
+#
+#   # 自定义匹配规则（支持多个，逗号分隔）
+#   bash bash/cleanup_docker_images.sh --pattern 'swebench/*,alexgshaw/*' --yes
+#
+#   # 同时清理 dangling 镜像和未使用卷
+#   bash bash/cleanup_docker_images.sh --preset all-bench --yes --prune
+# ============================================================
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+# 预设的 benchmark 镜像命名空间
+PRESET_SWE=('swebench/*')
+PRESET_TERMINAL=('alexgshaw/*')
+PRESET_ALL_BENCH=()
+PRESET_ALL_BENCH+=("${PRESET_SWE[@]}")
+PRESET_ALL_BENCH+=("${PRESET_TERMINAL[@]}")
+
+DRY_RUN=1
+PRESET=''
+PATTERNS=''
+PRUNE=0
+
+usage() {
+    sed -n '7,34p' "$0"
+}
+
+while [[ $# -gt 0 ]]; do
+    case "$1" in
+        --preset)
+            PRESET="${2:-}"
+            shift 2
+            ;;
+        --pattern)
+            PATTERNS="${2:-}"
+            shift 2
+            ;;
+        --yes)
+            DRY_RUN=0
+            shift
+            ;;
+        --prune)
+            PRUNE=1
+            shift
+            ;;
+        -h|--help)
+            usage
+            exit 0
+            ;;
+        *)
+            echo "Unknown option: $1"
+            usage
+            exit 1
+            ;;
+    esac
+done
+
+# 解析出要删除的 pattern 列表
+PATTERN_LIST=()
+if [[ -n "$PRESET" ]]; then
+    case "$PRESET" in
+        swe)
+            PATTERN_LIST+=("${PRESET_SWE[@]}")
+            ;;
+        terminal)
+            PATTERN_LIST+=("${PRESET_TERMINAL[@]}")
+            ;;
+        all-bench)
+            PATTERN_LIST+=("${PRESET_ALL_BENCH[@]}")
+            ;;
+        *)
+            echo "ERROR: unknown preset '$PRESET'. Available: swe, terminal, all-bench"
+            exit 1
+            ;;
+    esac
+fi
+
+if [[ -n "$PATTERNS" ]]; then
+    IFS=',' read -ra CUSTOM_PATTERNS <<< "$PATTERNS"
+    PATTERN_LIST+=("${CUSTOM_PATTERNS[@]}")
+fi
+
+if [[ ${#PATTERN_LIST[@]} -eq 0 ]]; then
+    echo "ERROR: 请指定 --preset 或 --pattern"
+    usage
+    exit 1
+fi
+
+echo "============================================================"
+echo "清理目标 pattern:"
+for p in "${PATTERN_LIST[@]}"; do
+    echo "  - $p"
+done
+echo "============================================================"
+
+# 收集所有匹配且没有被运行中容器使用的镜像 ID
+MATCHED_IMAGES=()
+RUNNING_IMAGES=$(docker ps --format '{{.Image}}' | sort -u)
+
+for pattern in "${PATTERN_LIST[@]}"; do
+    while IFS= read -r img; do
+        [[ -z "$img" ]] && continue
+        repo="${img%:*}"
+        # 检查是否有运行中容器在使用该镜像
+        if echo "$RUNNING_IMAGES" | grep -qx "$img"; then
+            echo "SKIP (running container uses): $img"
+            continue
+        fi
+        # 避免重复
+        if [[ ! " ${MATCHED_IMAGES[*]} " =~ " ${img} " ]]; then
+            MATCHED_IMAGES+=("$img")
+        fi
+    done < <(docker images --format '{{.Repository}}:{{.Tag}}' --filter=reference="$pattern")
+done
+
+if [[ ${#MATCHED_IMAGES[@]} -eq 0 ]]; then
+    echo "没有匹配到可删除的镜像"
+    exit 0
+fi
+
+echo ""
+echo "匹配到 ${#MATCHED_IMAGES[@]} 个镜像："
+for img in "${MATCHED_IMAGES[@]}"; do
+    size=$(docker images --format '{{.Size}}' "$img" | head -1)
+    echo "  $img ($size)"
+done
+
+# 计算标记总大小（注意共享层会被多次计算，仅作参考）
+TOTAL_SIZE_GB=$(docker images --format '{{.Size}} {{.Repository}}:{{.Tag}}' \
+    | awk -v img_list="${MATCHED_IMAGES[*]}" '
+    BEGIN {
+        n = split(img_list, imgs, " ");
+        for (i=1; i<=n; i++) target[imgs[i]] = 1;
+    }
+    {
+        size = $1;
+        img = $2;
+        if (!(img in target)) next;
+        unit = substr(size, length(size)-1);
+        val = substr(size, 1, length(size)-2);
+        if (unit == "GB") sum += val * 1024 * 1024 * 1024;
+        else if (unit == "MB") sum += val * 1024 * 1024;
+        else if (unit == "kB") sum += val * 1024;
+    }
+    END {
+        printf "%.2f", sum / 1024 / 1024 / 1024;
+    }')
+
+echo ""
+echo "标记总大小约: ${TOTAL_SIZE_GB} GB（含共享层，实际释放会小于该值）"
+
+if [[ "$DRY_RUN" -eq 1 ]]; then
+    echo ""
+    echo "当前是预览模式，不会删除。如需删除请加上 --yes"
+    exit 0
+fi
+
+echo ""
+echo "开始删除..."
+FAILED=0
+for img in "${MATCHED_IMAGES[@]}"; do
+    if docker rmi -f "$img" >/dev/null 2>&1; then
+        echo "  DELETED $img"
+    else
+        echo "  FAILED  $img"
+        FAILED=1
+    fi
+done
+
+if [[ "$PRUNE" -eq 1 ]]; then
+    echo ""
+    echo "清理 dangling 镜像和未使用卷..."
+    docker image prune -f
+    docker volume prune -f
+fi
+
+echo ""
+echo "============================================================"
+if [[ "$FAILED" -eq 0 ]]; then
+    echo "清理完成"
+else
+    echo "部分镜像删除失败"
+fi
+echo "============================================================"