HTTP Handler

The Databend HTTP handler is a REST API that used to send query statement for execution on the server and to receive results back to the client.

The HTTP handler is hosted by databend-query, it can be specified by using --http_handler_host and --http_handler_port(This defaults to 8000).

HTTP Methods

Overview

This handler return results in pages with long-polling.

1. Start with A POST to /v1/query with JSON of type QueryRequest which contains the SQL to execute, returns a JSON of type QueryResponse.
2. Use fields of QueryResponse for further processing:
   1. A GET to the next_uri returns the next page of query results. It returns QueryResponse too, processing it the same way until next_uri is null.
   2. (optional) A GET to the kill_uri to kill the query. Return empty body.
   3. (optional) A GET to the stats_uri to get stats only at once (without long-polling), return QueryResponse with empty data field.

Please note that you should keep using the latest next_uri to get the next page of results before the query is finished, otherwise you may miss some results or leak session resources until session timeout. The next_uri will be null when you have received all the results of the query.

Quick Example

curl -u root: \
  --request POST \
  '127.0.0.1:8001/v1/query/' \
  --header 'Content-Type: application/json' \
  --data-raw '{"sql": "SELECT avg(number) FROM numbers(100000000)"}'

the SQL will be run with default session and pagination settings, mainly:

1. use a new one-off session with the default database.
2. each request wait for at most 1 second for results before return.

for more advanced configs, see the Reference below:

you are expected to get JSON like this (formatted):

{
  "id": "b22c5bba-5e78-4e50-87b0-ec3855c757f5",
  "session_id": "5643627c-a900-43ac-978f-8c76026d9944",
  "session": {},
  "schema": [
    {
      "name": "avg(number)",
      "type": "Nullable(Float64)"
    }
  ],
  "data": [["49999999.5"]],
  "state": "Succeeded",
  "error": null,
  "stats": {
    "scan_progress": {
      "rows": 100000000,
      "bytes": 800000000
    },
    "write_progress": {
      "rows": 0,
      "bytes": 0
    },
    "result_progress": {
      "rows": 1,
      "bytes": 9
    },
    "total_scan": {
      "rows": 100000000,
      "bytes": 800000000
    },
    "running_time_ms": 446.748083
  },
  "affect": null,
  "stats_uri": "/v1/query/b22c5bba-5e78-4e50-87b0-ec3855c757f5",
  "final_uri": "/v1/query/b22c5bba-5e78-4e50-87b0-ec3855c757f5/final",
  "next_uri": "/v1/query/b22c5bba-5e78-4e50-87b0-ec3855c757f5/final",
  "kill_uri": "/v1/query/b22c5bba-5e78-4e50-87b0-ec3855c757f5/kill"
}

Query Request

QueryRequest

field	type	Required	Default	description
sql	string	Yes		the sql to execute
session_id	string	No		used only when reuse server-side session
session	SessionState	No
pagination	Pagination	No		a uniq query_id for this POST request

SessionState

field	type	Required	Default	description
database	string	No	"default"	set current_database
keep_server_session_secs	int	No	0	secs the Session will be retain after the last query finished
settings	map(string, string)	No	0

OldSession

field	type	Required	Default	description
id	string	Yes		session_id from QueryResponse.session_id

Pagination: critical conditions for each HTTP request to return (before all remaining result is ready to return)

field	type	Required	Default	description
wait_time_secs	u32	No	1	long polling time

Query Response

QueryResponse:

field	type	description
state	string	choices: "Running","Failed", "Succeeded"
error	QueryError	error of the sql parsing or execution
id	string	a uniq query_id for this POST request
data	array	each item is a row of results
schema	array	An ordered sequence of Field
affect	Affect	the affect of some queries
session_id	String
session	SessionState

Field:

field	type
name	string
type	string

Stats:

field	type	description
running_time_ms	float	million secs elapsed since query begin to execute internally, stop timing when query Finished (state != Running)
scan_progress	QueryProgress	query scan progress

Progress:

field	type
read_rows	int
read_bytes	int

Error:

field	type	description
stats	int	error code used inside databend
message	string	error message
backtrace	string

Affect:

field	type	description
type	string	ChangeSetting/UseDB
...		according to type

Response Status Code

The usage of status code for different kinds of errors:

code	error
200	if sql is invalid or failed, the detail is in the error field of the JSON
404	"query_id" or "page" not found
400	invalid request format

Check the response body for error reason as a string when status code is not 200.

data format

all field value in .data is represented in string, client need to interpreter the values with the help of information in the schema field.

client-side session

Dur to the stateless nature of HTTP, it is hard to maintain session in server side. client need to config the session in the session field when starting a new Request.

{
  "sql": "select 1",
  "session": {
    "database": "db2",
    "settings": {
      "max_threads": "1"
    }
  }
}

all the values of settings should be string.

If the SQL is set or use, the session will be change, can carry back to client in side response, client need to record it and put it in the following Request.

QueryAffect (Experimental)

For each SQL, client get an optional table-formed result. Client also get info about progress about rows/bytes read/write. Due to the limit of them, client may not get all interesting information about the Query. So we add QueryAffect to carry some extra information about the Query.

Note that QueryAffect is for advanced user and is not stable. It is not recommended to QueryAffect to maintain session.

Example for Session and QueryAffect:

set statement:

{
  "sql": "set max_threads=1;",
  "session": {
    "database": "db1",
    "settings": {
      "max_threads": "6"
    }
  }
}

response:

{
  "affect": {
    "type": "ChangeSetting",
    "key": "max_threads",
    "value": "1",
    "is_global": false
  },
  "session": {
    "database": "db1",
    "settings": {
      "max_threads": "1"
    }
  }
}

use statement:

{
  "sql": "use db2",
  "session": {
    "database": "db1",
    "settings": {
      "max_threads": "6"
    }
  }
}

response:

{
  "affect": {
    "type": "UseDB",
    "name": "db2"
  },
  "session": {
    "database": "db2",
    "settings": {
      "max_threads": "1"
    }
  }
}

Stage Attachment

Databend allows you to insert or update data from a staged file into a table by utilizing the INSERT INTO or REPLACE INTO statement with its HTTP Handler.

Example: Inserting Data from Staged File

create table t_insert_stage(a int null, b int default 2, c float, d varchar default 'd');

Upload values.csv to a stage:

values.csv

1,1.0
2,2.0
3,3.0
4,4.0

Request /v1/upload_to_stage

curl -H "stage_name:my_int_stage" -F "upload=@./values.csv" -XPUT http://root:@localhost:8000/v1/upload_to_stage

Insert with the uploaded file:

curl -d '{"sql": "insert into t_insert_stage (a, c) values", "stage_attachment": {"location": "@my_int_stage/values.csv", "file_format_options": {}, "copy_options": {}}}' -H 'Content-type: application/json' http://root:@localhost:8000/v1/query

tip

You can specify the file format and various copy-related settings with the FILE_FORMAT and COPY_OPTIONS available in the COPY INTO command. When purge is set to true, the original file will only be deleted if the data update is successful.

Verify the inserted data:

select * from t_insert_stage;
+------+------+------+------+
| a    | b    | c    | d    |
+------+------+------+------+
|    1 |    2 |  1.0 | d    |
|    2 |    2 |  2.0 | d    |
|    3 |    2 |  3.0 | d    |
|    4 |    2 |  4.0 | d    |
+------+------+------+------+

Example: Replacing Data with Staged File

First, create a table called "sample":

CREATE TABLE sample
(
    Id      INT,
    City    VARCHAR,
    Score   INT,
    Country VARCHAR DEFAULT 'China'
);

Then, create an internal stage and upload a sample CSV file called sample_3_replace.csv to the stage:

CREATE STAGE s1 FILE_FORMAT = (TYPE = CSV);

curl -u root: -H "stage_name:s1" -F "upload=@sample_3_replace.csv" -XPUT "http://localhost:8000/v1/upload_to_stage"
{"id":"b8305187-c816-4bb5-8350-c441b85baaf9","stage_name":"s1","state":"SUCCESS","files":["sample_3_replace.csv"]}

LIST @s1;
name                |size|md5|last_modified                |creator|
--------------------+----+---+-----------------------------+-------+
sample_3_replace.csv|  83|   |2023-06-12 03:01:56.522 +0000|       |

Use REPLACE INTO to insert data from the staged CSV file through the HTTP handler:

tip

You can specify the file format and various copy-related settings with the FILE_FORMAT and COPY_OPTIONS available in the COPY INTO command. When purge is set to true, the original file will only be deleted if the data update is successful.

curl -s -u root: -XPOST "http://localhost:8000/v1/query" --header 'Content-Type: application/json' -d '{"sql": "REPLACE INTO sample (Id, City, Score) ON(Id) VALUES", "stage_attachment": {"location": "@s1/sample_3_replace.csv", "copy_options": {"purge": "true"}}}'
{"id":"92182fc6-11b9-461b-8fbd-f82ecaa637ef","session_id":"f5caf18a-5dc8-422d-80b7-719a6da76039","session":{},"schema":[],"data":[],"state":"Succeeded","error":null,"stats":{"scan_progress":{"rows":5,"bytes":83},"write_progress":{"rows":5,"bytes":277},"result_progress":{"rows":0,"bytes":0},"total_scan":{"rows":0,"bytes":0},"running_time_ms":143.632441},"affect":null,"stats_uri":"/v1/query/92182fc6-11b9-461b-8fbd-f82ecaa637ef","final_uri":"/v1/query/92182fc6-11b9-461b-8fbd-f82ecaa637ef/final","next_uri":"/v1/query/92182fc6-11b9-461b-8fbd-f82ecaa637ef/final","kill_uri":"/v1/query/92182fc6-11b9-461b-8fbd-f82ecaa637ef/kill"}

Verify the inserted data:

SELECT * FROM sample;
id|city       |score|country|
--+-----------+-----+-------+
 1|'Chengdu'  |   80|China  |
 3|'Chongqing'|   90|China  |
 6|'HangZhou' |   92|China  |
 9|'Changsha' |   91|China  |
10|'Hong Kong'|   88|China  |

client implementations

The official client bendsql is mainly base on HTTP handler.

The most simple example of http handler client implementation is in sqllogictest for databend.