5

What method do you use to version-control your database? I've committed all our database tables as separate .sql scripts to our respository (mercurial). In that way, if any member of the team makes a change to the employee table, say, I will immediately know which particular table has been modified when I updated my repository.

Such a method was described in: What are the best practices for database scripts under code control. Presently, I'm writing a python script to execute all the .sql files within the database folder, however, the issue of dependencies due to foreign-key constraints ensures we can't just run the .sql files in just any order.

The python script is to generate a file with the order in which to execute the .sql files. It will execute the .sql files in the order in which they appear in the tableorder.txt file. A table cannot be executed until its foreign key table has been executed, for example:

tableorder.txt

table3.sql
table1.sql
table7.sql and so on

Already, I have generated the dependency list for each table, from code, by parsing the result of the "show create table" mysql command. The dependency list may look thus: 

tblstate: tblcountry //tblcountry.sql must be executed before tblstate.sql etc
tblemployee: tbldepartment, tblcountry

To generate the content of the tableorder.txt, I will need an algorithm that will look thus:

function print_table(table):
  foreach table in database:
    if table.dependencies.count == 0
      print to tableorder.txt
    if table.dependencies.count > 0
      print_table(dependency) //print dependency first
end function

As you will imagine, this involves lots of recursion. I'm beginning to wonder if it's worth the effort? If there's some tool out there? What tool (or algorithm) is there to generate a list of the order to execute separate .sql tables and views taking into consideration dependencies? Is it better to version control separate .sql file for each table/view or better to version control the entire database to a single .sql file? I will appreciate any response as this has taken so many days. Thanks.

Community
  • 1
  • 1
SleepingSpider
  • 1,168
  • 4
  • 19
  • 35

4 Answers4

4

I do not use MySQL, but rather SQL Server, however, this is how I version my database:

(This is long, but in the end I hope the reasoning for me abandoning a simple schema dump as the primary way to handle database versioning is made apparent.)

  1. I make a modification to the schema and apply it to a test database.

  2. I generate delta change scripts and a dump of the schema after said scripts. (I use ApexSQL, but there are likely MySQL-specific tools to help.)

    1. The delta change scripts know how to go from the current to target schema version: ALTER TABLE existing, CREATE TABLE new, DROP VIEW old .. Multiple operations can occur within the same .SQL file as the delta is of importance.

    2. The dump of the schema is of the target schema version: CREATE TABLE a, CREATE VIEW b .. there is no "ALTER" or "DROP" here, because it is just a snapshot of the target schema. There is one .SQL file per database object as the schema is of importance.

  3. I use RoundhousE to apply the delta change scripts. (I do not use the RoundhousE "anytime script" feature as this does not correctly handle relationships.)

I learned the hard way that applying database schema changes cannot be reliably done without a comprehensive step-by-step plan and, similarly (as noted in the question), the order of relationship dependencies are important. Just storing the "current" or "end" schema is not sufficient. There are many changes that cannot be retroactively applied A->C without knowing A->B->C and some changes B might involve migration logic or corrections. SQL schema change scripts can capture these changes and allow them to be "replayed".

However, at the same time just saving the delta scripts does not provide a "simple view" of the target schema. This is why I also dump all the schema as well as the change scripts and version both. The view dump could, in theory, be used to construct the database but due to relationship dependencies (the very kind noted in the question), it may take some work and I do not use it as part of an automated schema-restore approach: yet, keeping the schema dump part of the Hg version-control allows quick identification of changes and viewing the target schema at a particular version.

The change deltas thus move forward through the revisions while the schema dump provides a view at the current revision. Because the change deltas are incremental and forward-only it is important to keep the branch dealing with these changes "clean", which is easy to do with Hg.

In one of my projects I am currently at database change number 70 - and happy and productive! - after switching to this setup. (And these are deployed changes, not just development changes!)

Happy coding.

2

You can use sqitch. Here is a tutorial for MySql, but it is actually database agnostic.

Changes are implemented as scripts native to your selected database engine... Database changes may declare dependencies on other changes—even on changes from other Sqitch projects. This ensures proper order of execution, even when you’ve committed changes to your VCS out-of-order... Change deployment is managed by maintaining a plan file. As such, there is no need to number your changes, although you can if you want. Sqitch doesn’t much care how you name your changes... Up until you tag and release your application, you can modify your change deployment scripts as often as you like. They’re not locked in just because they’ve been committed to your VCS. This allows you to take an iterative approach to developing your database schema. Or, better, you can do test-driven database development.

user2683246
  • 3,399
  • 29
  • 31
1

I'm not sure how well this answers your question, but I tend to just use mysqldump (part of the standard installation). This gives me the sql to create the tables and populate them, effectively serializing the database. Example:

> mysqldump -u username -p yourdatabase > database_dump.sql

To load a database from a dump sql file:

mysql -u username -p -e "source /path/to/database_dump.sql"

To further answer your question, I would version control each table separately only if there are multiple people working on the database in such a way that conflicts are likely to occur with just a single dump being version controlled. I've never hit a project where this is the case (the database tends to be one of the least volatile portions of the system after the initial phases of the project), so I just version control the database dump as a whole rather than each table individually.

S.C.
  • 1,152
  • 8
  • 13
0

I understand the problem but you cannot think of controlling the versions of the databases using git as if it were static code "" since it does not work, in the same way and it is not very useful to generate different files for each programmer since as you say they collide or Well, they do not have traceability, I started a project similar to how you have it, but it was one more huge problem when trying to have control over the versions and the collisions of the programmers, the solution that arrives is to generate a project where the following order is maintained

  • Enter web Login / password
  • Administration of users and profiles of what each user can do
  • Committee area -> the common and current command is sent to the database

  • Example: alter table ALTER TABLE users ADD por2 varchar (255);

the commit creates a traceability in the control system itself and the structure is sent to git starting from the initial structure for the control of changes

  • Change Control Area: it is the visualization of the commit itself plus the structure generated after the change

  • Server configuration area: the server is configured and a gitlab or github repository is added to it to carry version control in a more visual way without problems for developers

  • Backup restoration area: Send a backup and keep track of each version "Result of the change of the database structure"

This is the best handling I found without leaving the job to someone specific. I hope it helps you, I believe it in phyton which was the best I found since it uses Django and you save a lot of programming from the administrative part .. Greetings

Arelis_xzx
  • 59
  • 3