dbcut 0.6.0

Extract a lightweight subset of your relational production database for development and testing purpose.

DBcut

DBcut aims to allow the extraction of lightweight subset of relational production database for development and testing purpose.

Table of Contents

• Overview

  • Usage

• Getting started

• Under The Hood

  • Database Reflection and Loading Stategy

  • SQL from YAML

  • Extraction Graph

Overview

Its main features are:

• Extract data from large databases.

• Reinject data into another base.

• Target and source databases could be based on different DBMS (i.e., MySQL -> PostgreSQL/SQLite).

• Extraction queries simplified in YAML.

• Support nested associations.

• Json and plain SQL export.

• Reasonable performance.

• Caching of extractions to accelerate future extractions.

Usage

Usage: dbcut [OPTIONS] COMMAND1 [ARGS]... [COMMAND2 [ARGS]...]...

  Extract a lightweight subset of your production DB for development and
  testing purpose.

Options:
  -c, --config PATH    Configuration file
  --version            Show the version and exit.
  -y, --force-yes      Never prompts for user intervention
  -i, --interactive    Prompts for user intervention.
  --quiet, --no-quiet  Suppresses most warning and diagnostic messages.
  --debug              Enables debug mode.
  --verbose            Enables verbose output.
  -h, --help           Show this message and exit.

Commands:
  load        Extract and load data to the target database.
  flush       Remove ALL TABLES from the target database and recreate them
  inspect     Check databases content.
  dumpsql     Dump all SQL insert queries.
  dumpjson    Export data to json.
  clear       Remove all data (only) from the target database
  purgecache  Remove all cached queries.

Getting started

Let’s take the following database example:

We want to extract some users with all related data to our development database.

First, we have to edit the extraction file dbcut.yaml as follows:

# dbcut.yml
databases:
  source_uri: mysql://foo:bar@db-host/prod
  destination_uri: sqlite:///small-dev-database.db

queries:
  - from: user
    limit: 2

Then, we set the limit to two users, the default limit being 10.

After that, we launch the extraction command with the load command:

$ dbcut load
 ---> Reflecting database schema from mysql://foo:***@db-host/prod
 ---> Creating new sqlite:///small-dev-database.db database
 ---> Creating all tables and relations on sqlite:///small-dev-database.db

Query 1/1 :

    from: user
    limit: 2
    backref_limit: 10
    backref_depth: 5
    join_depth: 5
    exclude: []
    include: []


     ┌─ⁿ─comment
     ├─ⁿ─vote
 user┤
     └─ⁿ─user_group┐
                   └─¹─group┐
                            └─¹─role┐
                                    └─ⁿ─role_permission┐
                                                       └─¹─permission


8 tables loaded

 ---> Cache key : 4a468c3555074890b7c342c0a575f29d47145821
 ---> Executing query
 ---> Fetching objects
 ---> Inserting 31 rows

We can check the data on our new database :

$ ls
dbcut.yml  small-dev-database.db
$ sqlite3 small-dev-database.db

sqlite> SELECT id, login FROM user;
3|jerome
4|julien

sqlite> SELECT * from comment;
8|comment jerome 1|3
9|comment jerome 2|3
10|comment jerome 3|3

In the following example, we are going to retrieve roles with related groups and permissions. In order to obtain the best extraction graph, we are going to use the keyword include, which indicated to dbcut that we want to minimize the number of associated tables (Nested associations).

queries:
  - from: user
    limit: 2

  - from: role
    include:
      - group
      - permission

It is possible to empty the content of the local database before beginning the extraction with the clear command.

$ dbcut -y clear load
 ---> Removing all data from sqlite:///small-dev-database.db database
 ---> Reflecting database schema from mysql://foo:***@db-host/prod?charset=utf8
 ---> Creating all tables and relations on sqlite:///small-dev-database.db

Query 1/2 :

    from: user
    limit: 2
    backref_limit: 10
    backref_depth: 5
    join_depth: 5
    exclude: []
    include: []


     ┌─ⁿ─comment
     ├─ⁿ─vote
 user┤
     └─ⁿ─user_group┐
                   └─¹─group┐
                            └─¹─role┐
                                    └─ⁿ─role_permission┐
                                                       └─¹─permission


8 tables loaded

 ---> Cache key : 4a468c3555074890b7c342c0a575f29d47145821
 ---> Using cache (2 elements)
 ---> Fetching objects
 ---> Inserting 31 rows

Query 2/2 :

    from: role
    limit: 10
    backref_limit: 10
    backref_depth: null
    join_depth: null
    exclude: []
    include:
    - group
    - permission


     ┌─ⁿ─group
 role┤
     └─ⁿ─role_permission┐
                        └─¹─permission


4 tables loaded

 ---> Cache key : 5029d84dbb2bc75a7df898dd94df93b395e91e44
 ---> Executing query
 ---> Fetching objects
 ---> Inserting 22 rows

As you can see in the first query, the cache was used and there was thus no interaction with the source database.

This query allowed the extraction of all roles:

sqlite> SELECT * from role;
1|admin
2|moderator
3|user

If we had not used the include keyword, all tables would have been extracted:

    ┌─ⁿ─role_permission┐
    │                  └─¹─permission
role┤
    └─ⁿ─group┐
             └─ⁿ─user_group┐
                           │       ┌─ⁿ─comment
                           └─¹─user┤
                                   └─ⁿ─vote

To narrow more precisely our extraction, we are now going to limit to roles that can delete a user.

queries:
  - from: user
    limit: 2

  - from: role
    include:
      - group
      - permission
    where:
      permission.codename: 'delete_user'

Only the last extraction rule is relaunched with the --last-only option.

$ dbcut -y clear load --last-only
...
 ---> Cache key : ffb664a2e69c88fa48db2680daf71d30408bd207
 ---> Executing query
 ---> Fetching objects
 ---> Inserting 14 rows

This time, only the ‘admin’ role is retrieved:

sqlite> SELECT * from role;
1|admin

Please note that the filter only applies here to role table (from) and not to the permission.

sqlite> SELECT * FROM permission";
1|delete_comment
2|delete_vote
3|delete_user
4|create_comment
5|create_vote
6|create_user

Indeed, we filter the roles based on a value from the permission table, but we do retrieved all permissions associated to this role.

In the above example, it makes sense that the admin role has all permissions.

Last but not least, we can also retrieve data in json or raw sql format !

$ dbcut dumpjson|dumpsql

[
  {
    "password": "julien",
    "vote_collection": [
      {
        "user_id": 4,
        "comment_id": 1,
        "id": 3,
        "rating": 4
      },
      {
        "user_id": 4,
        "comment_id": 3,
        "id": 6,
        "rating": 10
      },
      {
        "user_id": 4,
        "comment_id": 6,
        "id": 13,
        "rating": 10
      }
    ],
    "comment_collection": [],
    "id": 4,
    "login": "julien",
    "user_group_collection": [
      {
        "user_id": 4,
        "group": {
          "name": "Utilisateur",
          "role": {
            "id": 3,
            "role_permission_collection": [
              {
                "permission": {
                  "id": 4,
                  "codename": "create_comment",
                  "role_permission_collection": []
                },

PRAGMA foreign_keys = OFF;

BEGIN;
INSERT OR IGNORE INTO permission (id, codename) VALUES (4, 'create_comment');
INSERT OR IGNORE INTO permission (id, codename) VALUES (5, 'create_vote');
INSERT OR IGNORE INTO permission (id, codename) VALUES (1, 'delete_comment');
INSERT OR IGNORE INTO permission (id, codename) VALUES (2, 'delete_vote');
INSERT OR IGNORE INTO role (id, name) VALUES (3, 'user');
INSERT OR IGNORE INTO role (id, name) VALUES (2, 'moderator');
INSERT OR IGNORE INTO user (id, login, password) VALUES (4, 'julien', 'julien');
INSERT OR IGNORE INTO user (id, login, password) VALUES (3, 'jerome', 'jerome');
INSERT OR IGNORE INTO "group" (id, name, role_id) VALUES (3, 'Utilisateur', 3);
INSERT OR IGNORE INTO "group" (id, name, role_id) VALUES (2, 'Moderateur', 2);
INSERT OR IGNORE INTO comment (id, content, user_id) VALUES (8, 'comment jerome 1', 3);
INSERT OR IGNORE INTO comment (id, content, user_id) VALUES (9, 'comment jerome 2', 3);
INSERT OR IGNORE INTO comment (id, content, user_id) VALUES (10, 'comment jerome 3', 3);
INSERT OR IGNORE INTO role_permission (id, role_id, permission_id) VALUES (12, 3, 4);
INSERT OR IGNORE INTO role_permission (id, role_id, permission_id) VALUES (13, 3, 5);
INSERT OR IGNORE INTO role_permission (id, role_id, permission_id) VALUES (7, 2, 4);
INSERT OR IGNORE INTO role_permission (id, role_id, permission_id) VALUES (8, 2, 5);
INSERT OR IGNORE INTO role_permission (id, role_id, permission_id) VALUES (10, 2, 1);
INSERT OR IGNORE INTO role_permission (id, role_id, permission_id) VALUES (11, 2, 2);
INSERT OR IGNORE INTO user_group (id, user_id, group_id) VALUES (4, 4, 3);
INSERT OR IGNORE INTO user_group (id, user_id, group_id) VALUES (3, 3, 2);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (3, 4, 4, 1);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (6, 10, 4, 3);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (13, 10, 4, 6);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (2, 5, 3, 1);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (5, 1, 3, 2);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (7, 10, 3, 3);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (10, 6, 3, 1);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (11, 5, 3, 5);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (12, 6, 3, 6);
INSERT OR IGNORE INTO vote (id, rating, user_id, comment_id) VALUES (19, 10, 3, 10);
COMMIT;

Under The Hood

Database Reflection and Loading Stategy

DBcut heavily uses SQLAlchemy, the SQL toolkit and Object Relational Mapper for Python. The ORM makes it possible to free ourselves from the SQL direct manipulation, but that is not all. SQLAlchemy offers a range of toolkits that enable us to programmatically build all SQL queries useful to DBcut. This include both the schema creation and all of its properties, the select, join and insert queries… no matter which DBMS is used (PostgreSQL, MySQL, SQLite, oracle etc.).

One of the most important features of DBcut is that the user does not need to know or provide the source database schema to use it. First of all, DBcut will inspect the source database and retrieve all metadata. This action is what we call: Database Reflection.

The MetaData object store all the collection of metadata entities. DBcut will alter this MetaData object to make it compatible with most DBMS. For example, the names of indexes or foreign keys can be too long for SQLite but not for MYSQL. Sometimes, it also changes the types of the column to make it match what is expected in the target database. (mysql.TINYINT became SMALLINT in SQLite and PostgreSQL)

Once the MetaData object is complete, we can create the new database which is almost identical to the source database (except some compatibility adjustments)

DBcut will generate and launch extraction request on the source database. The data thus obtained will be detached from the first SQLAlchemy session to be attached to the new session in the target database. This is where the SQLAlchemy magic happens: the same request will be used to extract data from the source database and to load them into the target database. Indeed, in the first case (query/fetch), it will be translated into SQL SELECT queries and in the second case, into SQL INSERT statements (load).

SQL from YAML

One of the goals of DBcut is to allow quick writing of extraction requests. Most of the time, to write an extraction request, not much information is needed: only the main table name, hoping to retrieve the maximum number of related data as possible.

The idea was to find a sufficiently concise syntax that allows us to build the most complete extraction requests with the minimum effort.

The YAML came to us naturally as it is pleasant to read, easy to understand and to edit for humans.

The dbcut.yml file is both used to configure DBcut and to write extraction requests.

databases:
  source_uri: mysql://chinook:chinook@192.168.66.66/chinook
  destination_uri: sqlite:///chinook.db

queries:
  - from: customer_customer

To write an extraction request, only the keyword from is mandatory. However, other keywords can be added to reduce the size of data to retreive.

- from: contracts_customer
  where:
    brand: 2
  limit: 100
  backref_limit: 500
  backref_depth: 2
  join_depth: 5
  exclude:
  - django_admin_log
  - django_session
  include: []

Unlike the SQL queries, an extraction request using DBcut automatically and recursively loads all associated relations (See Extraction Graph). All these options are filtering and reducing options that prevents from slowing down the extraction process.

Finally, with the scope of making the extraction requests as compact as possible, we can add default values to most of these options:

default_limit: 100
default_backref_limit: 500

default_backref_depth: 2
default_join_depth: 5

global_exclude:
  - django_admin_log
  - django_session

Extraction Graph

To build an extraction request, we first build its extraction graph.

An extraction graph is a subset of the complete graph of database relations. Every node represents a table, and each link represents a relation between two tables. The link direction is defined by the foreign key.

To build this graph, we use the MetaData object (See Database Reflection and Loading Stategy).

Let’s use the following database schema:

The retrieved metadata during the database reflection are used to build the following complete graph of relations:

To build the extraction graph, we browse the complete graph starting from the table used in the from instruction. The browsing only stops if :

• the link has already been browsed

• the table is explicitly excluded

• the maximum depth is reached

For the following request:

queries:
  - from: customer_customer

The generated extraction graph is:

Please note that we handle the two types of relations : one-to-many relations (noted 1 in the extraction graph) and many-to-many relations (noted n).

CHANGELOG

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Version 0.6.0

Released on April 13th 2021

Added

• Better support for tables without primary key

• Load all many-to-many relationships just like one-to-many ones

Version 0.5.0

Released on April 09th 2021

Added

• Added support of SQLAlchemy 1.4

Version 0.4.1

Released on April 01st 2021

Changed

• Downgrade sqlalchemy requirement to <=1.4 versions to prevent crashing

Version 0.4.0

Released on March 12th 2021

Changed

• Ensure that generated names for SQLAlchemy relationships do not conflict with columns names

• Always generate unique indexes on sqlite

• Translate mysql current_timestamp() to CURRENT_TIMESTAMP on sqlite

Version 0.3.1

Released on March 05th 2021

Added

• New experimental class Recorder that record all SQL interactions in order to replay them offline (for testing purpose)

Version 0.2.0

Released on August 21st 2020

Added

• Always enable SQLAlchemy post_update to avoid circular dependecies

• Disable all cache if cache config key is set to no

• Do not globally exclude already loaded relations from the other extraction graph branches

• Improved cyclical relations loading in the extraction graph

Changed

• Store cache by dbcut version

Removed

• Removed marshmallow serialization and prefer builtin json

Version 0.1.6

Released on August 20th 2020

Changed

• flush purges the cache only when –with-cache is passed

Fixed

• Load only transient mapper objects to force sqlalchemy to generate sql insert queries

Version 0.1.5

Released on August 11th 2020

Changed

• dumpsql prints only create and insert sql statements

Fixed

• Fixed dumpjson regression

• Fixed query caching mechanism

• Prepared all mapper objects correctly when metadata is cached

• Various minor bug fixes

Version 0.1.4

Released on May 07th 2020

Fixed

• Fixed TypeError exception

• Defined a max length for indexes on TEXT column on mysql databases

Version 0.1.3

Released on November 27th 2019

Changed

• clear cmd delete only existing table.

Fixed

• Determistic cache key generation.

First release on PyPI.