Elasticsearch date_histogram内scripted_metric空指针异常排查

我在Elasticsearch中有一个数组类型字段，要统计该数组中跨文档不重复的唯一项数量。单独使用scripted_metric聚合时正常，但嵌套在date_histogram聚合里执行就抛出空指针异常。

聚合代码（PHP格式）：

'aggs' => [
    'groupByWeek' => [
        'date_histogram' => [
            'field' => 'date',
            'calendar_interval' => '1w',
        ],
        'aggs' => [
            'count_unique_locations' => [
                'scripted_metric' => [
                     'init_script' => 'state.locations = []',
                     'map_script' => 'state.locations.addAll(doc.unique_locations_with_error)',
                     'combine_script' => 'return state.locations',
                     'reduce_script' => "
                         def locations = [];
                         for (state in states) {
                             for(location in state) {
                                 if(!locations.contains(location) && location != '' ) {
                                     locations.add(location); 
                                 }
                             }
                        }
                        return locations.length;
                    ",
                ],
            ],
        ],
    ],
],

执行查询时的错误：

{
  "error": {
    "root_cause":[],
    "type":"search_phase_execution_exception",
    "reason":"",
    "phase":"fetch",
    "grouped":true,
    "failed_shards":[],
    "caused_by":{
      "type":"script_exception",
      "reason":"runtime error",
      "script_stack":[
        "locations = []; ","            ^---- HERE"],
        "script":"def locations = []; for (state in states) { for(location in state){if(!locations.contains(location) && location != '' ) {locations.add(location); }}} return locations.length;",
      "lang":"painless",
      "position":{
        "offset":16,
        "start":4,
        "end":20
      },
      "caused_by":{
       "type":"null_pointer_exception",
       "reason":"cannot access method/field [iterator] from a null def reference"
      }
    }
  },
  "status":400
}

我推测问题和空文档有关，但不知道具体原因和修复方法。

原因及修复方法 #

原因分析 #

当date_histogram的某个时间桶没有匹配到任何文档时，scripted_metric的reduce阶段会收到包含null的states数组，遍历null时就会触发空指针异常。单独使用scripted_metric时，至少有匹配文档，所以不会出现这种情况。

修复步骤 #

1. 在reduce脚本中增加null判断：遍历states时先检查当前state是否为null，避免后续迭代出错。
2. 优化集合去重逻辑：用HashSet替代ArrayList，contains操作的时间复杂度从O(n)降到O(1)，性能更优。

修改后的reduce_script：

def locations = new HashSet();
for (state in states) {
    // 跳过空的state
    if (state == null) {
        continue;
    }
    for (location in state) {
        // 过滤空字符串并去重
        if (location != null && location != '') {
            locations.add(location);
        }
    }
}
return locations.size();

3. 可选：在map阶段增加空字段判断：避免doc.unique_locations_with_error字段不存在或为空时抛出异常，修改map_script：

if (doc.containsKey('unique_locations_with_error') && !doc.unique_locations_with_error.empty) {
    state.locations.addAll(doc.unique_locations_with_error);
}

内容的提问来源于stack exchange，提问作者Julian Kragt