mirror of https://github.com/apache/druid.git
30 lines
921 B
Markdown
30 lines
921 B
Markdown
---
|
|
layout: default
|
|
---
|
|
Time boundary queries return the earliest and latest data points of a data set. The grammar is:
|
|
|
|
<code>{
|
|
"queryType" : "timeBoundary",
|
|
"dataSource": "sample_datasource"
|
|
}
|
|
</code>
|
|
|
|
There are 3 main parts to a time boundary query:
|
|
|
|
|property|description|required?|
|
|
|--------|-----------|---------|
|
|
|queryType|This String should always be “timeBoundary”; this is the first thing Druid looks at to figure out how to interpret the query|yes|
|
|
|dataSource|A String defining the data source to query, very similar to a table in a relational database|yes|
|
|
|context|An additional JSON Object which can be used to specify certain flags.|no|
|
|
|
|
The format of the result is:
|
|
|
|
<code>[ {
|
|
"timestamp" : "2013-05-09T18:24:00.000Z",
|
|
"result" : {
|
|
"minTime" : "2013-05-09T18:24:00.000Z",
|
|
"maxTime" : "2013-05-09T18:37:00.000Z"
|
|
}
|
|
} ]
|
|
</code>
|