Druid的查询是使用REST风格的HTTP请求查询服务节点(Broker、Historical、Realtime),这些服务节点暴露REST查询接口,客户端发送Json对象请求查询接口。一般情况下,查询服务接口发布在Broker节点,基于Linux 的POST请求查询如下所示:
/**
* port: 查询请求接口对应Broker,默认8082端口
* query_json_file: 查询Json对象文件(配置)
*/
curl -X POST '<queryable_host>:<port>/druid/v2/?pretty' -H 'Content-Type:application/json' -d @<query_json_file>
1. 聚合查询 - 时间序列查询(Timeseries)、排名查询(TopN)、分组查询(GroupBy)
2. 元数据查询 - 时间范围(Time Boundary) 、段元数据(Segment Metadata)、数据源(Datasource)
3. Search查询 - Search
本节以聚合查询为主,其它查询类型比较简单,使用上相对比较少,暂不介绍。对聚合查询类型下的3种查询如何选择进行一下概述:
在可能的情况下,我们建议使用的时间序列和TopN查询代替分组查询,分组查询是Druid最灵活的的查询,但是性能最差。时间序列查询是明显快于GROUPBY查询,因为聚合不需要分组尺寸。对于分组和排序在一个单一的维度,TopN查询更优于GROUPBY。
1. 简单聚合粒度 - 支持字符串值有:all、none、second、minute、fifteen_minute、thirty_minute、hour、day、week、month、quarter、year
(1) all - 将所有块变成一块
(2) none - 不使用块数据(它实际上是使用最小索引的粒度,none意味着为毫秒级的粒度);按时间序列化查询时不建议使用none,因为所有的毫秒不存在,系统也将尝试生成0值,这往往是很多。
2. 时间段聚合粒度 - Druid指定一精确的持续时间(毫秒)和时间缀返回UTC(世界标准时间)。
3. 常用时间段聚合粒度 - 与时间段聚合粒度差不多,但是常用时间指平时我们常用时间段,如年、月、周、小时等。
下面对3种聚合粒度配置举例说明:
简单聚合粒度
查询粒度比数据采集时配置的粒度小,则不合理,也无意义,因较小粒度(相比)者无索引数据;如
查询粒度小于采集时配置的查询粒度时,则Druid的查询结果与采集数据配置的查询粒度结果一样。
假设我们存储在Druid的数据使用毫秒粒度获取,数据如下:
{"timestamp": "2013-08-31T01:02:33Z", "page": "AAA", "language" : "en"}
{"timestamp": "2013-09-01T01:02:33Z", "page": "BBB", "language" : "en"}
{"timestamp": "2013-09-02T23:32:45Z", "page": "CCC", "language" : "en"}
{"timestamp": "2013-09-03T03:32:45Z", "page": "DDD", "language" : "en"}
以"小时" 粒度提交一个groupby查询,查询配置如下:
{
"queryType":"groupBy",
"dataSource":"dataSource",
"granularity":"hour",
"dimensions":[
"language"
],
"aggregations":[
{
"type":"count",
"name":"count"
}
],
"intervals":[
"2000-01-01T00:00Z/3000-01-01T00:00Z"
]
}
按小时粒度进行的groupby查询结果中timestamp值精确到小时间,比小时粒度更小粒度值自动补填零,
以此类推按天查询,则小时及小粒度补零。timestamp值为UTC
[ {
"version" : "v1",
"timestamp" : "2013-08-31T01:00:00.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
}, {
"version" : "v1",
"timestamp" : "2013-09-01T01:00:00.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
}, {
"version" : "v1",
"timestamp" : "2013-09-02T23:00:00.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
}, {
"version" : "v1",
"timestamp" : "2013-09-03T03:00:00.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
} ]
如果指定查询粒度为 none,则返回结果与数据导入时设置粒度(queryGranularity属性值)结果一样,
此处的导入粒度为毫秒,结果如下:
[ {
"version" : "v1",
"timestamp" : "2013-08-31T01:02:33.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
}, {
"version" : "v1",
"timestamp" : "2013-09-01T01:02:33.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
}, {
"version" : "v1",
"timestamp" : "2013-09-02T23:32:45.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
}, {
"version" : "v1",
"timestamp" : "2013-09-03T03:32:45.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
} ]
如果指定查询粒度为 all,返回数组长度结果为1,结果如下:
[ {
"version" : "v1",
"timestamp" : "2000-01-01T00:00:00.000Z",
"event" : {
"count" : 4,
"language" : "en"
}
} ]
时间段聚合粒度
指定一个精确时间持续时长(毫秒表示)及时间缀,返回UTC时间;支持可选项属性origin,不指定时
默认开始时间(1970-01-01T00:00:00Z)
/**持续时间段2小时,从1970-01-01T00:00:00Z开始*/
{"type": "duration", "duration": 7200000}
/**持续时间1小时,从origin开始*/
{"type": "duration", "duration": 3600000, "origin": "2012-01-01T00:30:00Z"}
以上简单聚合粒度的示例数据为例,提交groupby查询,持续时间段为24小时,查询配置如下:
{
"queryType":"groupBy",
"dataSource":"dataSource",
"granularity":{"type": "duration", "duration": "86400000"},
"dimensions":[
"language"
],
"aggregations":[
{
"type":"count",
"name":"count"
}
],
"intervals":[
"2000-01-01T00:00Z/3000-01-01T00:00Z"
]
}
查询结果:
[ {
"version" : "v1",
"timestamp" : "2013-08-31T00:00:00.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
}, {
"version" : "v1",
"timestamp" : "2013-09-01T00:00:00.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
}, {
"version" : "v1",
"timestamp" : "2013-09-02T00:00:00.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
}, {
"version" : "v1",
"timestamp" : "2013-09-03T00:00:00.000Z",
"event" : {
"count" : 1,
"language" : "en"
}
} ]
常用时间段聚合粒度
略...
示例简单查看使用方式:
查询过滤(Selector filte)
等价于:WHERE <dimension_string> = '<dimension_value_string>'
"filter": { "type": "selector", "dimension": <dimension_string>, "value": <dimension_value_string> }
正则表达过滤(Regular expression filter)
与Selector filte差不多,只是这里使用正则表达式,表达式为标准的Java正则表达式规范
"filter": { "type": "regex", "dimension": <dimension_string>, "pattern": <pattern_string> }
逻缉表达过滤(Logical expression filters)
AND
"filter": { "type": "and", "fields": [<filter>, <filter>, ...] }
OR
"filter": { "type": "or", "fields": [<filter>, <filter>, ...] }
NOT
"filter": { "type": "not", "field": <filter> }
IN过滤(In filter)
SQL查询
SELECT COUNT(*) AS 'Count' FROM `table` WHERE `outlaw` IN ('Good', 'Bad', 'Ugly')
{
"type": "in",
"dimension": "outlaw",
"values": ["Good", "Bad", "Ugly"]
}
范围过滤(Bound filter)
Bound filter 过滤比较值大小或小于某值,默认按字符串比较,使用数据比较需要设置alphaNumeric 属
性为true;默认 Bound filter为非严格性(类闭区间),如 inputString <= upper && inputSting >= lower
{
"type": "bound",
"dimension": "age",
"lower": "21",
"upper": "31" ,
"alphaNumeric": true
}
上述表示等价如:21 <= age <= 31
Bound filter 严格性,需要设置lowerStrict or/and upperStrict 属性值为true如下:
{
"type": "bound",
"dimension": "age",
"lower": "21",
"lowerStrict": true,
"upper": "31" ,
"upperStrict": true,
"alphaNumeric": true
}
等价如:21 < age < 31
Count aggregator
查询返回匹配过滤条件的数据行数,需要注意的是:Druid进行Count查询的数据量并不一定等于数据采
集时导入的数据量,因为Druid在采集数据并导入时已经对数据进行了聚合。
{ "type" : "count", "name" : <output_name> }
Sum aggregator
longSum aggregator:计算值为有符号位64位整数
{ "type" : "longSum", "name" : <output_name>, "fieldName" : <metric_name> }
doubleSum aggregator:与longSum类似,计算值为64位浮点型
{ "type" : "doubleSum", "name" : <output_name>, "fieldName" : <metric_name> }
Min / Max aggregators
doubleMin aggregator
{ "type" : "doubleMin", "name" : <output_name>, "fieldName" : <metric_name> }
doubleMax aggregator
{ "type" : "doubleMax", "name" : <output_name>, "fieldName" : <metric_name> }
longMin aggregator
{ "type" : "longMin", "name" : <output_name>, "fieldName" : <metric_name> }
longMax aggregator
{ "type" : "longMax", "name" : <output_name>, "fieldName" : <metric_name> }
类似聚合(Approximate Aggregations)
基数聚合(Cardinality aggregator)
计算Druid多种维度基数,Cardinality aggregator使用HyperLogLog评估基数,这种聚合比带有索引的
hyperUnique聚合慢,运行在一个维度列,意味着不能从数据集中删除字符串维度来提高聚合;一般我们
强力推荐使用hyperUnique aggregator而不是Cardinality aggregator,格式如下:
{
"type": "cardinality",
"name": "<output_name>",
"fieldNames": [ <dimension1>, <dimension2>, ... ],
"byRow": <false | true> # (optional, defaults to false)
}
. 维度值聚合-当设置属性byRow为false(默认值)时,通过合并所有给定的维度列来计算值集合。
对于单维度,等价如下:
SELECT COUNT(DISTINCT(dimension)) FROM <datasource>
对于多维度,等价如下:
SELECT COUNT(DISTINCT(value)) FROM (
SELECT dim_1 as value FROM <datasource>
UNION
SELECT dim_2 as value FROM <datasource>
UNION
SELECT dim_3 as value FROM <datasource>
)
. 行聚合-当设置属性byRow为true时,根所不同维度的值合并来计算行值,等价如下:
SELECT COUNT(*) FROM ( SELECT DIM1, DIM2, DIM3 FROM <datasource> GROUP BY DIM1, DIM2, DIM3 )
许多不同国家的人出生地或来自哪里,用druid配置如下:
{
"type": "cardinality",
"name": "distinct_countries",
"fieldNames": [ "coutry_of_origin", "country_of_residence" ]
}
HyperUnique aggregator
已经被“hyperunique”在创建索引时聚合的维度值使用HyperLogLog计算估计,更多资料请参考官网
{ "type" : "hyperUnique", "name" : <output_name>, "fieldName" : <metric_name> }
1. Arithmetic post-aggregators
2. Field accessor post-aggregator
3. Constant post-aggregator
4. JavaScript post-aggregator
5. HyperUnique Cardinality post-aggregator
Arithmetic post-aggregators
算术后聚合应用已提供的函数从左到右获取字段,这些字段可聚合或后聚合;支持+
, -
, *
, /
, and quotient。
算术后聚合可以指定ordering属性,用于聚合结果排序(对topN查询很有用 ):
(1) 如果无ordering属性(或null),使用默认的浮点排序。
(2) numericFirst 首先返回有限值,其次是NaN,最后返回无限值。
算术后聚合语法如下:
postAggregation : {
"type" : "arithmetic",
"name" : <output_name>,
"fn" : <arithmetic_function>,
"fields": [<post_aggregator>, <post_aggregator>, ...],
"ordering" : <null (default), or "numericFirst">
}
Field accessor post-aggregator - fieldName引用aggregator定义的名称
{ "type" : "fieldAccess", "name": <output_name>, "fieldName" : <aggregator_name> }
Constant post-aggregator - 返回指定值
{ "type" : "constant", "name" : <output_name>, "value" : <numerical_value> }
属性 | 描述 | 必填项 |
queryType | 字符串类型,时间序列 "timeseries" | 是 |
dataSource | 字符串类型,数据源(类似数据库表) | 是 |
descending | 排序标志,默认为 "false"(升序) | 否 |
intervals | 查询时间范围跨度,JSON对象,ISO-8601区间 | 是 |
granularity | 定义查询结果块粒度 | 是 |
filter | 过滤条件 | 否 |
aggregations | 聚合 | 是 |
postAggregations | 后聚合 | 否 |
context | 上下文 | 否 |
{ "queryType": "timeseries", "dataSource": "sample_datasource", "granularity": "day", "descending": "true", "filter": { "type": "and", "fields": [ { "type": "selector", "dimension": "sample_dimension1", "value": "sample_value1" }, { "type": "or", "fields": [ { "type": "selector", "dimension": "sample_dimension2", "value": "sample_value2" }, { "type": "selector", "dimension": "sample_dimension3", "value": "sample_value3" } ] } ] }, "aggregations": [ { "type": "longSum", "name": "sample_name1", "fieldName": "sample_fieldName1" }, { "type": "doubleSum", "name": "sample_name2", "fieldName": "sample_fieldName2" } ], "postAggregations": [ { "type": "arithmetic", "name": "sample_divide", "fn": "/", "fields": [ { "type": "fieldAccess", "name": "postAgg__sample_name1", "fieldName": "sample_name1" }, { "type": "fieldAccess", "name": "postAgg__sample_name2", "fieldName": "sample_name2" } ] } ], "intervals": [ "2012-01-01T00:00:00.000/2012-01-03T00:00:00.000" ] }
上述配置了过滤条件,2个聚合,后聚合器将2个聚合结果进行相除。查询结果如下,查询结果存储在属性result,以键值对方式存储:
[
{
"timestamp": "2012-01-01T00:00:00.000Z",
"result": { "sample_name1": <some_value>, "sample_name2": <some_value>, "sample_divide": <some_value> }
},
{
"timestamp": "2012-01-02T00:00:00.000Z",
"result": { "sample_name1": <some_value>, "sample_name2": <some_value>, "sample_divide": <some_value> }
}
]
TopN在每个节点将顶上K个结果排名,在Druid默认情况下最大值为1000。在实践中,如果你要求前1000个项顺序排名,那么从第1-999个项的顺序正确性是100%,其后项的结果顺序没有保证。你可以通过增加threshold值来保证顺序准确。
属性 | 描述 | 必填项 |
queryType | 字符串类型,时间序列 "topN" | 是 |
dataSource | 字符串类型,数据源(类似数据库表) | 是 |
intervals | 查询时间范围跨度,JSON对象,ISO-8601区间 | 是 |
granularity | 定义查询结果块粒度 | 是 |
filter | 过滤条件 | 否 |
aggregations | 聚合 | 是 |
postAggregations | 后聚合 | 否 |
dimension | 查询的维度(列) | 是 |
threshold | 返回Top N个结果 | 是 |
metric | 字符串或Json对象指定度量对Top N个结果排序 | 是 |
context | 上下文 | 否 |
属性 | 描述 | 必填项 |
type | 数字排序 | 是 |
metric | 排序字段 | 是 |
数据排序(Numeric TopNMetricSpec) - 最简单的规范指定一个字符串值指示排序TopN结果的度量
"metric": "<metric_name>"
metric属性通常配置为Json对象,上述等价于:
"metric": {
"type": "numeric",
"metric": "<metric_name>"
}
topN query 配置示例如下:
{
"queryType": "topN",
"dataSource": "sample_data",
"dimension": "sample_dim",
"threshold": 5,
"metric": "count",
"granularity": "all",
"filter": {
"type": "and",
"fields": [
{
"type": "selector",
"dimension": "dim1",
"value": "some_value"
},
{
"type": "selector",
"dimension": "dim2",
"value": "some_other_val"
}
]
},
"aggregations": [
{
"type": "longSum",
"name": "count",
"fieldName": "count"
},
{
"type": "doubleSum",
"name": "some_metric",
"fieldName": "some_metric"
}
],
"postAggregations": [
{
"type": "arithmetic",
"name": "sample_divide",
"fn": "/",
"fields": [
{
"type": "fieldAccess",
"name": "some_metric",
"fieldName": "some_metric"
},
{
"type": "fieldAccess",
"name": "count",
"fieldName": "count"
}
]
}
],
"intervals": [
"2013-08-31T00:00:00.000/2013-09-03T00:00:00.000"
]
}
查询前Top 5个结果,按count排序:
[
{
"timestamp": "2013-08-31T00:00:00.000Z",
"result": [
{
"dim1": "dim1_val",
"count": 111,
"some_metrics": 10669,
"average": 96.11711711711712
},
{
"dim1": "another_dim1_val",
"count": 88,
"some_metrics": 28344,
"average": 322.09090909090907
},
{
"dim1": "dim1_val3",
"count": 70,
"some_metrics": 871,
"average": 12.442857142857143
},
{
"dim1": "dim1_val4",
"count": 62,
"some_metrics": 815,
"average": 13.14516129032258
},
{
"dim1": "dim1_val5",
"count": 60,
"some_metrics": 2787,
"average": 46.45
}
]
}
]
待续。。。