MongoDB按照天数或小时聚合
需求
最近接到需求,需要对用户账户下的设备状态,分别按照天以及小时进行聚合,以此为基础绘制设备状态趋势图.
实现思路是启动定时任务,对各用户的设备状态数据分别按照小时以及天进行聚合,并存储进数据库中供用户后续查询.
涉及到的技术栈分别为:Spring Boot
,MongoDB
,Morphia
.
数据模型
@Data
@Builder
@Entity(value = "rawDevStatus", noClassnameStored = true)
// 设备状态索引
@Indexes({
// 设置数据超时时间(TTL,MongoDB根据TTL在后台进行数据删除操作)
@Index(fields = @Field("time"), options = @IndexOptions(expireAfterSeconds = 3600 * 24 * 72)),
@Index(fields = {@Field("userId"), @Field(value = "time", type = IndexType.DESC)})
})
public class RawDevStatus {
@Id
@JsonProperty(access = JsonProperty.Access.WRITE_ONLY)
private ObjectId objectId;
private String userId;
private Instant time;
@Embedded("points")
List<Point> protocolPoints;
@Data
@AllArgsConstructor
public static class Point {
/**
* 协议类型
*/
private Protocol protocol;
/**
* 设备总数
*/
private Integer total;
/**
* 设备在线数目
*/
private Integer onlineNum;
/**
* 处于启用状态设备数目
*/
private Integer enableNum;
}
}
上述代码是设备状态
实体类,其中设备状态数据是按照设备所属协议
进行区分的.
@Data
@Builder
@Entity(value = "aggregationDevStatus", noClassnameStored = true)
@Indexes({
@Index(fields = @Field("expireAt"), options = @IndexOptions(expireAfterSeconds = 0)),
@Index(fields = {@Field("userId"), @Field(value = "time", type = IndexType.DESC)})
})
public class AggregationDevStatus {
@Id
@JsonProperty(access = JsonProperty.Access.WRITE_ONLY)
private ObjectId objectId;
/**
* 用户ID
*/
private String userId;
/**
* 设备总数
*/
private Double total;
/**
* 设备在线数目
*/
private Double onlineNum;
/**
* 处于启用状态设备数目
*/
private Double enableNum;
/**
* 聚合类型(按照小时还是按照天聚合)
*/
@Property("aggDuration")
private AggregationDuration aggregationDuration;
private Instant time;
/**
* 动态设置文档过期时间
*/
private Instant expireAt;
}
上述代码是期待的聚合结果,其中构建两个索引:(1)超时索引;(2)复合索引,程序会根据用户名以及时间查询设备状态聚合结果.
聚合操作符介绍
聚合操作类似于管道,管道中的每一步操作产生的中间结果作为下一步的输入源,最终输出聚合结果.
此次聚合主要涉及以下操作:
$project
:指定输出文档中的字段.$unwind
:拆分数据中的数组;match
:选择要处理的文档数据;group
:根据key
分组聚合结果.
原始聚合语句
db.getCollection('raw_dev_status').aggregate([
{$match:
{
time:{$gte: ISODate("2019-06-27T00:00:00Z")},
}
},
{$unwind: "$points"},
{$project:
{
userId:1,points:1,
tmp: {$dateToString: { format: "%Y:%m:%dT%H:00:00Z", date: "$time" } }
}
},
{$project:
{
userId:1,points:1,
groupTime: {$dateFromString: { dateString: "$tmp", format: "%Y:%m:%dT%H:%M:%SZ", } }
}
},
{$group:
{
_id:{user_id:'$userId', cal_time:'$groupTime'},
devTotal:{'$avg':'$points.total'},
onlineTotal:{'$avg':'$points.onlineNum'},
enableTotal:{'$avg':'$points.enableNum'}
}
},
])
上述代码是按小时聚合数据,以下来逐步介绍处理思路:
(1) $match
根据小时聚合数据,因为只需要获取近24小时的聚合结果,所以对数据进行初步筛选.
(2) $unwind
raw_dev_status
中的设备状态是按照协议区分的数组,因此需要对其进行展开,以便下一步进行筛选;
(3) $project
{$project:
{
userId:1,points:1,
tmp: {$dateToString: { format: "%Y:%m:%dT%H:00:00Z", date: "$time" } }
}
}
选择需要输出的数据,分别为:userId
,points
以及tmp
.
需要注意,为了按照时间聚合,对$time
属性进行操作,提取%Y:%m:%dT%H
时信息至$tmp
作为下一步的聚合依据.
(4) $project
{$project:
{
userId:1,points:1,
groupTime: {$dateFromString: { dateString: "$tmp", format: "%Y:%m:%dT%H:%M:%SZ", } }
}
}
因为上一步project
操作中,tmp
为字符串数据,最终的聚合结果需要时间戳(主要懒,不想在程序中进行转换操作).
因此,此处对$tmp
进行操作,转换为时间类型数据,即groupTime
.
(5) $group
对聚合结果进行分类操作,并生成最终输出结果.
{$group:
{
# 根据_id进行分组操作,依据是`user_id`以及`$groupTime`
_id:{user_id:'$userId', cal_time:'$groupTime'},
# 求设备总数平均值
devTotal:{'$avg':'$points.total'},
# 求设备在线数平均值
onlineTotal:{'$avg':'$points.onlineNum'},
# ...
enableTotal:{'$avg':'$points.enableNum'}
}
}
代码编写
此处ODM
选择Morphia
,亦可以使用MongoTemplate
,原理类似.
/**
* 创建聚合条件
*
* @param pastTime 过去时间段
* @param dateToString 格式化字符串(%Y:%m:%dT%H:00:00Z或%Y:%m:%dT00:00:00Z)
* @return 聚合条件
*/
private AggregationPipeline createAggregationPipeline(Instant pastTime, String dateToString, String stringToDate) {
Query<RawDevStatus> query = datastore.createQuery(RawDevStatus.class);
return datastore.createAggregation(RawDevStatus.class)
.match(query.field("time").greaterThanOrEq(pastTime))
.unwind("points", new UnwindOptions().preserveNullAndEmptyArrays(false))
.match(query.field("points.protocol").equal("ALL"))
.project(Projection.projection("userId"),
Projection.projection("points"),
Projection.projection("convertTime",
Projection.expression("$dateToString",
new BasicDBObject("format", dateToString)
.append("date", "$time"))
)
)
.project(Projection.projection("userId"),
Projection.projection("points"),
Projection.projection("convertTime",
Projection.expression("$dateFromString",
new BasicDBObject("format", stringToDate)
.append("dateString", "$convertTime"))
)
)
.group(
Group.id(Group.grouping("userId"), Group.grouping("convertTime")),
Group.grouping("total", Group.average("points.total")),
Group.grouping("onlineNum", Group.average("points.onlineNum")),
Group.grouping("enableNum", Group.average("points.enableNum"))
);
}
/**
* 获取聚合结果
*
* @param pipeline 聚合条件
* @return 聚合结果
*/
private List<AggregationMidDevStatus> getAggregationResult(AggregationPipeline pipeline) {
List<AggregationMidDevStatus> statuses = new ArrayList<>();
Iterator<AggregationMidDevStatus> resultIterator = pipeline.aggregate(
AggregationMidDevStatus.class, AggregationOptions.builder().allowDiskUse(true).build());
while (resultIterator.hasNext()) {
statuses.add(resultIterator.next());
}
return statuses;
}
//......................................................................................
// 获取聚合结果(省略若干代码)
AggregationPipeline pipeline = createAggregationPipeline(pastTime, dateToString, stringToDate);
List<AggregationMidDevStatus> midStatuses = getAggregationResult(pipeline);
if (CollectionUtils.isEmpty(midStatuses)) {
log.warn("Can not get dev status aggregation result.");
return;
}