Skip to content

Latest commit

 

History

History
368 lines (309 loc) · 13.6 KB

README.md

File metadata and controls

368 lines (309 loc) · 13.6 KB
Raw

Data Connector Agent for DuckDB

This directory contains an DuckDB implementation of a data connector agent. It can use local DuckDB database files as referenced by the "db" config field.

This is a basically a SQLite to DuckDB replacement of https://github.com/hasura/sqlite-dataconnector-agent.

Current Status

The DuckDB agent is a functioning port of the SQLite DataConnect Agent to use DuckDB as a backend instead. Currently, it only works on DuckDB native tables, but provides functionality required for basic read-only usage.

  • GraphQL Schema from Tables
  • GraphQL Schema from Views
  • GraphQL Schema from remotely accessed files
  • Preliminary support for nested types
  • GraphQL Queries
  • Relationships
  • Aggregations

Aim

This DataConnector allows you to utilize DuckDB's broad array of data importers to expose common analytics file formats through GraphQL. Through the cleverness and type-safe schemas of DuckDB, the data does not even need to be present in DuckDB. This allows you to create GraphQL schemas for combinations of remote files and define relationships on them, utilizing DuckDB as a processing engine.

This will, eventually be updated to take advantage of DuckDB's exceptional OLAP capabilities, and can already perform some rapid aggregations in some instances. Further development will allow

Example

This pointless example demonstrates pulling and joining data externally as well as locally from multiple file formats Consider the following schema in /data/sample.duckdb (note the creation of a local JSON file):

COPY (FROM VALUES
  ('MSFT', 'Microsoft', 'Hayward'),
  ('GOOG', 'Google', 'Hayward'),
  ('APPL', 'Apple', 'San Francisco'),
  ('DATA', 'Tableau', 'San Jose') AS v(symbol, name, city)
) TO 'companies.json';
CREATE VIEW companies AS FROM 'companies.json';
CREATE VIEW holdings AS FROM 'https://duckdb.org/data/prices.parquet';
CREATE VIEW weather AS
  SELECT column0 AS city, column4 as date, column1 as low, column2 as high, column3 as rainfall
  FROM read_csv('https://duckdb.org/data/weather.csv');

You will need to configure the following relationships manually (there is an example configuration export at the end of this page):

  holdings.company: (ticker) -[object]-> companies (symbol)
  companies.holdings: (symbol) -[array]-> holdings (ticker)
  companies.weather: (city) -[array]-> weather (city)

The following query is possible:

{
  holdings {
    ticker
    shares
    company {
      name
      weather(where: {rainfall: {_is_null: false}},
        	    order_by: {date:desc}, limit: 1) { 
        date
        rainfall
      }
      weather_aggregate {
        aggregate {
          max { high }
          min { low }
        }
      }
    }
  }
}

Additionally, as long as the view schemas remain consistent, the definition of the view itself can be transparently changed within the DuckDB database or materialized as a table:

DROP VIEW companies;
CREATE TABLE companies AS FROM 'companies.json';

Or added into the init configuration of the connector (this is not a great way to do this, but it works):

{
  ...
  init: "CREATE TEMPORARY VIEW companies AS (FROM VALUES ('MSFT', 'Microsoft', 'Hayward'), ('GOOG', 'Google', 'Hayward'), ('APPL', 'Apple', 'San Francisco'), ('DATA', 'Tableau', 'San Jose') AS v(symbol, name, city) );"
  ...
}

Lastly, it's possible to use DuckDB's httpfs extension and an in-memory database to keep the database outside of the connector:

{
  db: ":memory:",
  init: "LOAD httpfs; ATTACH 'http://192.168.1.2:8080/chinook.duckdb'; set search_path='chinook,main'; CREATE OR REPLACE TEMPORARY VIEW companies AS FROM 'http://192.168.7.209:8082/companies.json';"
}

Capabilities

The DuckDB agent currently supports an unknown set of capabilities. It's a direct drop-in replacement for the SQLite agent.

The DuckDB agent currently has tested support the following capabilities:

  • GraphQL Schema from Tables
  • GraphQL Schema from Views
  • GraphQL Schema from Macros
  • GraphQL Schema from remotely accessed files
  • Preliminary support for nested types
  • Support for nested operators and nested types
  • GraphQL Queries
  • Relationships
  • Aggregations
  • Query explains
  • Prometheus Metrics
  • Exposing Foreign-Key Information
  • Mutations
  • Native (Interpolated) Queries
  • Subscriptions
  • Streaming Subscriptions

The SQLite agent currently supports the following capabilities:

  • GraphQL Schema
  • GraphQL Queries
  • Relationships
  • Aggregations
  • Prometheus Metrics
  • Exposing Foreign-Key Information
  • Mutations
  • Native (Interpolated) Queries
  • Subscriptions
  • Streaming Subscriptions

Note: You are able to get detailed metadata about the agent's capabilities by GETting the /capabilities endpoint of the running agent.

Requirements

Build & Run

npm install
npm run build
npm run start

Or a simple dev-loop via entr:

echo src/**/*.ts | xargs -n1 echo | DB_READONLY=y entr -r npm run start

Docker Build & Run

> docker build . -t dc-duckdb-agent:latest
> docker run -it --rm -v db.duckdb:/db.duckdb -p 8100:8100 dc-sqlite-agent:latest

You will want to mount a volume with your database(s) so that they can be referenced in configuration.

Options / Environment Variables

These are from the SQLite agent and have not yet been checked for DuckDB

Note: Boolean flags {FLAG} can be provided as 1, true, t, yes, y, or omitted and default to false.

ENV Variable Name Format Default Info
PORT INT 8100 Port for agent to listen on.
PERMISSIVE_CORS {FLAG} false Allows all requests - Useful for testing with SwaggerUI. Turn off on production.
DB_CREATE {FLAG} false Allows new databases to be created.
DB_READONLY {FLAG} false Makes databases readonly.
DB_ALLOW_LIST DB1[,DB2]* Any Allowed Restrict what databases can be connected to.
DB_PRIVATECACHE {FLAG} Shared Keep caches between connections private.
DEBUGGING_TAGS {FLAG} false Outputs xml style tags in query comments for deugging purposes.
PRETTY_PRINT_LOGS {FLAG} false Uses pino-pretty to pretty print request logs
LOG_LEVEL fatal | error | info | debug | trace | silent info The minimum log level to output
METRICS {FLAG} false Enables a /metrics prometheus metrics endpoint.
QUERY_LENGTH_LIMIT INT Infinity Puts a limit on the length of generated SQL before execution.
DATASETS {FLAG} false Enable dataset operations
DATASET_DELETE {FLAG} false Enable DELETE /datasets/:name
DATASET_TEMPLATES DIRECTORY ./dataset_templates Directory to clone datasets from.
DATASET_CLONES DIRECTORY ./dataset_clones Directory to clone datasets to.
MUTATIONS {FLAG} false Enable Mutation Support.

Agent usage

The agent is configured as per the configuration schema. The valid configuration properties are:

Property Type Default
db string
tables string[] null
init string null
include_sqlite_meta_tables boolean false
explicit_main_schema boolean false

The only required property is db which specifies a local sqlite database to use.

The init property allows specifying a number of SQL statements to run upon initializing a connection. For example: LOAD json; SET file_search_path='/data,/other-data'; SET autoload_known_extensions=false; SET lock_configuration=true;

Would ensure that the json extension is available and files can be referenced relative to /data or /other-data but that no additional extensions or changes can be made to the connecition configuration after initialization.

The schema is exposed via introspection, but you can limit which tables are referenced by

  • Explicitly enumerating them via the tables property, or
  • Toggling the include_sqlite_meta_tables to include or exclude sqlite meta tables.

The explicit_main_schema field can be set to opt into exposing tables by their fully qualified names (ie ["main", "MyTable"] instead of just ["MyTable"]).

Dataset

The dataset used for testing the reference agent is sourced from:

Datasets

Datasets support is enabled via the ENV variables:

  • DATASETS
  • DATASET_DELETE
  • DATASET_TEMPLATES
  • DATASET_CLONES

Templates will be looked up at ${DATASET_TEMPLATES}/${template_name}.sqlite or ${DATASET_TEMPLATES}/${template_name}.sql. The .sqlite templates are just SQLite database files that will be copied as a clone. The .sql templates are SQL script files that will be run against a blank SQLite database in order to create a clone.

Clones will be copied to ${DATASET_CLONES}/${clone_name}.sqlite.

Testing Changes to the Agent

Ensure you run the agent with DATASETS=1 DATASET_DELETE=1 MUTATIONS=1 in order to enable testing of mutations.

Then run:

cabal run dc-api:test:tests-dc-api -- test --agent-base-url http://localhost:8100 sandwich --tui

From the HGE repo.

Known Issues

  • Using "returning" in insert/update/delete mutations where you join across relationships that are affected by the insert/update/delete mutation itself may return inconsistent results. This is because of this issue with SQLite: https://sqlite.org/forum/forumpost/9470611066

TODO

  • Replace SQLite library with DuckDB in src/db.js
  • Add ability to provide initialization statements for DuckDB
  • Fix docker container segfaulting when attempting to use the npm-provided duckdb package.
  • Address the absense of JSON_EACH from DuckDB
  • Add Duckdb database for testing
  • Replace parsing of SQLITE_SCHEMA with use of the information_schema tables for introspection
  • Expose and enable views in the introspected GraphQL schema
  • Expose and enable table functions in the introspected GraphQL schema
  • Replace common init values with configuration options
  • Revise env variables
  • Re-enable parameterized queries (disabled due to conflicting/weird API differences)
  • Test the un-tested features provided from the SQLite connector

SQLite TODO (For Reference)

  • Prometheus metrics hosted at /metrics
  • Pull reference types from a package rather than checked-in files
  • Health Check
  • DB Specific Health Checks
  • Schema
  • Capabilities
  • Query
  • Array Relationships
  • Object Relationships
  • Ensure everything is escaped correctly - https://sequelize.org/api/v6/class/src/sequelize.js~sequelize#instance-method-escape
  • Or... Use parameterized queries if possible - https://sequelize.org/docs/v6/core-concepts/raw-queries/#bind-parameter
  • Run test-suite from SDK
  • Remove old queries module
  • Relationships / Joins
  • Rename resultTT and other badly named types in the schema.ts module
  • Add ENV Variable for restriction on what databases can be used
  • Update to the latest types
  • Port back to hge codebase as an official reference agent
  • Make escapeSQL global to the query module
  • Make CORS permissions configurable
  • Optional DB Allowlist
  • Fix SDK Test suite to be more flexible about descriptions
  • READONLY option
  • CREATE option
  • Don't create DB option
  • Aggregate queries
  • Verbosity settings
  • Cache settings
  • Missing WHERE clause from object relationships
  • Reuse find_table_relationship in more scenarios
  • ORDER clause in aggregates breaks SQLite parser for some reason
  • Check that looped exist check doesn't cause name conflicts
  • NOT EXISTS IS NULL != EXISTS IS NOT NULL
  • Mutation support

Further Examples

Exmaple metadata from above schema:

{
  "resource_version": 93,
  "metadata": {
    "version": 3,
    "sources": [
      {"name": "DuckDB Test",
        "kind": "DuckDB",
        "configuration": {
          "template": null,
          "timeout": null,
          "value": {
            "db": "/data/sample.duckdb",
            "explicit_main_schema": false,
            "include_sqlite_meta_tables": false,
            "init": "SET file_search_path='/data';"
          }
        },
        "tables": [
          {"table": ["companies"],
            "array_relationships": [
              {"name": "Holdings",
                "using": {
                  "manual_configuration": {
                    "column_mapping": {
                      "symbol": "ticker"
                    },
                    "insertion_order": null,
                    "remote_table": ["holdings"]
              }}},
              {"name": "Weather",
                "using": {
                  "manual_configuration": {
                    "column_mapping": {
                      "city": "city"
                    },
                    "insertion_order": null,
                    "remote_table": ["weather"]
              }}}]
          },
          {"table": ["holdings"],
            "object_relationships": [
              {"name": "Company",
                "using": {
                  "manual_configuration": {
                    "column_mapping": {
                      "ticker": "symbol"
                    },
                    "insertion_order": null,
                    "remote_table": ["companies"]
          }}}]
      }]
  }]}
}