wzymumon / SampleSeek

Java implementation of Sample+Seek*

• *Sample+Seek is a system published by Bolin Ding, etc. @Microsoft in SIGMOD 2016

Current support

• Restful API for query submission
• Uniform sampling based queries (COUNT, no SUM for now)
• Non-group-by queries (Will support group-by queries soon)
• Configuarable base table and sample table schemas

Run the system

Prerequisite

• Java 8
• Maven
• PostgreSQL

Build

mvn clean package

Configuration

Modify the sampleseek.yaml file according to your settings.

Run

java -jar target/cloudberry-1.0-SNAPSHOT-allinone.jar -c src/sampleseek.yaml

Send query

curl -H "Content-Type: application/json" -X POST -d '{"select": ["x", "y"], "filters": [{"attribute": "create_at", "operator": "IN", "operands": ["2017-06-01 00:00:00", "2017-07-01 00:00:00"]}]}' http://localhost:8080/query

API Definition

Non-group-by queries

{
"select": ["attributeName1", "attributeName2", ...],
"filters": [
  {"attribute": "attributeName", "operator": "IN/LT/GT/EQUAL", "operands": ["left-value", "right-value"]}, // operands list has 2 values only when operator is "IN", otherwise has only 1 value
  {...} // more filter conditions, and they are all conjunctive
]
}

group-by queries

{
"groupBy": ["attributeName1", "attributeName2", ...],
"aggFunction": "COUNT",
"aggAttribute": "attributeName",
"filters": [
  {"attribute": "attributeName", "operator": "IN/LT/GT/EQUAL", "operands": ["left-value", "right-value"]}, // operands list has 2 values only when operator is "IN", otherwise has only 1 value
  {...} // more filter conditions, and they are all conjunctive
]
}

Config file

serverConfig:
  hostname: localhost
  port: 8080

dbConfig:
  url: jdbc:postgresql://localhost/pinmap
  username: postgres
  password: pinmap

sampleConfig:
  baseTableName: tweets1 =
  baseTableColumnNames:
    - id
    - create_at
    - x # if base table uses [Point] datatype, here could use "coordinate[0]"
    - y # if base table uses [Point] datatype, here could use "coordinate[1]"
  baseTableSize: 1026036 # the number of records in base table
  sampleTableName: s_tweets1 # the table name for the sample, which will be generated when server starts
  sampleTableColumnNames: # list of columns' names in the sample table, *must be in the same order with base table
    - id
    - create_at
    - x
    - y
  sampleTableColumnTypes: # type of each column in sample table, *must be in the same order with column names
    - bigint
    - timestamp # the data types supported are: number, int, integer, bigint, timestamp, double, string
    - double
    - double

seekConfig:
  pkey: id
  measures:
    - x
    - y
  dimensions: # names of columns that can be group-by on or filter on
    - id
    - create_at
  dimensionTypes:
    - NUMERICAL
    - NUMERICAL
  epsilon: 0.05 # epsilon - error bound for approximate query result, ** whenever update this value, you need to drop the sample table manually