DBIx::SchemaChecksum - Manage your datebase schema via checksums
version 1.102
my $sc = DBIx::SchemaChecksum->new( dbh => $dbh );
print $sc->checksum;
When you're dealing with several instances of the same database (eg. developer, testing, stage, production), it is crucial to make sure that all databases use the same schema. This can be quite an hair-pulling experience, and this module should help you keep your hair (if you're already bald, it won't make your hair grow back, sorry...)
DBIx::SchemaChecksum gets schema information (tables, columns,
primary keys, foreign keys and some more depending on your DB) and
generates a SHA1 digest. This digest can then be used to easily verify
schema consistency across different databases, and to build an update
graph of changes. Therefor, DBIx::SchemaChecksum does not requires
you to add a meta-table to your database to keep track of which
changes have already been deployed.
Caveat: The same schema might produce different checksums on different database versions.
Caveat: DBIx::SchemaChecksum only works with database engines
that support changes to the schema inside a transaction. We know this
works with PostgreSQL and SQLite. We know it does not work with MySQL
and Oracle. We don't know how other database engines behave, but would
be happy to hear about your experiences.
Please take a look at the dbchecksum script included in this distribution. It provides a nice and powerful commandline interface to make working with your schema a breeze.
So you have this genious idea for a new startup that will make you incredibly rich and famous...
Usually such ideas involve a database. So you grab your favourite database engine and start a new database:
~/Gnomes$ createdb gnomes # createdb is a postgres tool
Of course this new DB is rather empty:
gnomes=# \d
No relations found.
So you think long and hard about your database schema and write it down
~/Gnomes$ cat sql/handcrafted_schema.sql
create table underpants (
id serial primary key,
type text,
size text,
color text
);
But instead of going down the rabbit hole of manually keeping the
dev-DB on your laptop, the one on the workstation in the office, the
staging and the production one in sync (and don't forget all the
databases running on the laptops of the countless coding monkeys
you're going to hire after all the VC money starts flowing), you grab
a (free!) copy of DBIx::SchemaChecksum
~/Gnomes$ cpanm DBIx::SchemaChecksum
.. wait a bit while the giant, on which shoulders we are standing, is being assembled
Successfully installed DBIx-SchemaChecksum
42 distribution installed
Now you can create a new changes file:
~/Gnomes$ dbchecksum new_changes_file --sqlsnippetdir sql --dsn dbi:Pg:dbname=gnomes --change_name "initial schema"
New change-file ready at sql/inital_schema.sql
Let's take a look:
~/Gnomes$ cat sql/inital_schema.sql
-- preSHA1sum: 54aa14e7b7e54cce8ae07c441f6bda316aa8458c
-- postSHA1sum: xxx-New-Checksum-xxx
-- inital schema
Each changes file contains two very import "header" lines masked as a SQL comment:
preSHA1sum is the checksum of the DB schema before the changes in
this file have been applied. postSHA1sum is (you probably guessed
it) the checksum we expect after the changes have been applied.
Currently the postSHA1sum is "xxx-New-Checksum-xxx" because we have
neither defined nor run the changes yet.
So let's append the handcrafted schema from earlier to the change file:
~/Gnomes$ cat sql/handcrafted_schema.sql >> sql/inital_schema.sql
The changes file now looks like this:
~/Gnomes$ cat sql/inital_schema.sql
-- preSHA1sum: 54aa14e7b7e54cce8ae07c441f6bda316aa8458c
-- postSHA1sum: xxx-New-Checksum-xxx
-- inital schema
create table underpants (
id serial primary key,
type text,
size text,
color text
);
Let's apply this schema change, so we can finally start coding (you just can't wait to get rich, can you?)
~/Gnomes$ dbchecksum apply_changes --sqlsnippetdir sql --dsn dbi:Pg:dbname=gnomes
Apply inital_schema.sql? [y/n] [y]
post checksum mismatch!
expected
got 611481f7599cc286fa539dbeb7ea27f049744dc7
ABORTING!
Woops! What happend here? Why couldn't the change be applied? Well, we
haven't yet defined the postSHA1sum, so we cannot be sure that the
database is in the state we expect it to be.
When you author a sql change, you will always have to first apply the
change to figure out the new postSHA1sum. As soon as
DBIx::SchemaChecksum tells you the checksum the DB will have after
the change is applied, you have to add the new checksum to your
changes file:
~/Gnomes$ vim sql/inital_schema.sql
# replace xxx-New-Checksum-xxx with 611481f7599cc286fa539dbeb7ea27f049744dc7
~/Gnomes$ head -2 sql/inital_schema.sql
-- preSHA1sum: 54aa14e7b7e54cce8ae07c441f6bda316aa8458c
-- postSHA1sum: 611481f7599cc286fa539dbeb7ea27f049744dc7
Now we can try again:
~/Gnomes$ dbchecksum apply_changes --sqlsnippetdir sql --dsn dbi:Pg:dbname=gnomes
Apply inital_schema.sql? [y/n] [y]
post checksum OK
No more changes
Yay, this looks much better!
Now you can finally start to collect underpants!
Some weeks later (you have now convinced a friend to join you in your quest for fortune) a git pull drops a new file into your sql directory. It seems that your colleague needs some tweaks to the database:
~/Gnomes$ cat sql/underpants_need_washing.sql
-- preSHA1sum: 611481f7599cc286fa539dbeb7ea27f049744dc7
-- postSHA1sum: 094ef4321e60b50c1d34529c312ecc2fcbbdfb51
-- underpants need washing
ALTER TABLE underpants ADD COLUMN needs_washing BOOLEAN NOT NULL DEFAULT false;
Seems reasonable, so you apply it:
~/Gnomes$ dbchecksum apply_changes --sqlsnippetdir sql --dsn dbi:Pg:dbname=gnomes
Apply underpants_need_washing.sql? [y/n] [y]
post checksum OK
No more changes
Now that was easy!
DBIx::SchemaChecksum uses MooseX::App to power the commandline
interface. We use the Config and ConfigHome plugins, so you can
pack some of the flags into a config file, for even less typing (and typos):
~/Gnomes$ cat dbchecksum.yml
global:
sqlsnippetdir: sql
dsn: dbi:Pg:dbname=gnomes
Now run:
~/Gnomes$ dbchecksum apply_changes --config dbchecksum.yml
db checksum 094ef4321e60b50c1d34529c312ecc2fcbbdfb51 matching sql/underpants_need_washing.sql
Or you can store the config file into your ~/.dbchecksum/config.yml:
~/Gnomes$ cat ~/.dbchecksum/config.yml
global:
sqlsnippetdir: sql
dsn: dbi:Pg:dbname=gnomes
And it magically works:
~/Gnomes$ dbchecksum apply_changes
db checksum 094ef4321e60b50c1d34529c312ecc2fcbbdfb51 matching sql/underpants_need_washing.sql
This section is left empty as an exercise for the reader!
sqlsnippetdir points to a directory containing so-called changes files. For a file to be picked up by dbchecksum it needs to use
the extension .sql.
The file itself has to contain a header formated as sql comments, i.e.
starting with --. The header has to contain the preSHA1sum and
should include the postSHA1sum.
If the postSHA1sum is missing, we assume that you don't know it yet and try to apply the change. As the new checksum will not match the empty postSHA1sum the change will fail. But we will report the new checksum, which you can now insert into the changes file.
After the header, the changes file should list all sql commands you
want to apply to change the schema, seperated by a semicolon ;,
just as you would type them into your sql prompt.
-- preSHA1sum: b1387d808800a5969f0aa9bcae2d89a0d0b4620b
-- postSHA1sum: 55df89fd956a03d637b52d13281bc252896f602f
CREATE TABLE nochntest (foo TEXT);
Not all commands need to actually alter the schema, you can also
include sql that just updates some data. In fact, some schmema changes
even require that: for example, if you want to add a NOT NULL
constraint to a column, you first have to make sure that the column in
fact does not contain a NULL.
-- preSHA1sum: c50519c54300ec2670618371a06f9140fa552965
-- postSHA1sum: 48dd6b3710a716fb85b005077dc534a8f9c11cba
UPDATE foo SET some_field = 42 WHERE some_field IS NULL;
ALTER TABLE foo ALTER some_filed SET NOT NULL;
Functions usually contain semicolons inside the function definition,
so we cannot split the file on semicolon. Luckily, you can specifiy a different splitter using -- split-at. We usually use ---- (again, the SQL comment marker) so the changes file is still valid SQL.
-- preSHA1sum c50519c54300ec2670618371a06f9140fa552965
-- postSHA1sum 48dd6b3710a716fb85b005077dc534a8f9c11cba
-- split-at ------
ALTER TABLE underpants
ADD COLUMN modified timestamp with time zone DEFAULT now() NOT NULL;
------
CREATE FUNCTION update_modified() RETURNS trigger
LANGUAGE plpgsql
AS $$
BEGIN
if NEW <> OLD THEN
NEW.modified = now();
END IF;
RETURN NEW;
END;
$$;
------
CREATE TRIGGER underpants_modified
BEFORE UPDATE ON underpants
FOR EACH ROW EXECUTE PROCEDURE update_modified();
We have been using DBIx::SchemaChecksum since 2008 and encountered
a few issues. Here are our solutions:
Sometimes two databases will produce different checksums. This can be
caused by a number of things. A good method to figure out what's
causing the problem is running <dbchecksum checksum --show_dump some_name>>
on the databases causing the problem. Then you can use
diff or vim -d to inspect the raw dump.
Some problems we have encountered, and how to fix them:
-
Manual changes
Somebody did a manual change to a database (maybe an experiment on a local DB, or some quick-fix on a live DB).
Fix: Revert the change. Maybe make a proper change file if the change makes sense for the project.
-
Bad search-path
The
search_pathsof the DBs differ. This can cause subtile diferences in the way keys and references are reported, thus causing a different checksum.Fix: Make sure all DBs use the same
search_path. -
Other schema-related troubles
Maybe the two instances use different values for
--schemata?Fix: Use the same
--schemataeverywhere. Put them in a config-file or write a wrapper script. -
Just weird diffs
Maybe the systems are using different version of the database server, client,
DBIorDBD::*. While we try hard to filter out version-specific differences, this might still cause problems.Fix: Use the same versions on all machines.
Sometimes it's impossible to get DBIx::SchemaChecksum installed on
the database server (or on some other machine, I have horrible
recollections about a colleague using Windows..). And the sysadmin
won't let you access the database over the network...
Fix: Prepare all changes on your local machine, and run them manually on the target machine.
~/Gnomes$ dbchecksum show_update_path --from_checksum 54aa14e7b7e54cce8ae07c441f6bda316aa8458c
inital_schema.sql (611481f7599cc286fa539dbeb7ea27f049744dc7)
underpants_need_washing.sql (094ef4321e60b50c1d34529c312ecc2fcbbdfb51)
No update found that's based on 094ef4321e60b50c1d34529c312ecc2fcbbdfb51.
Now you could import the changes manually on the server. But it's even
easier using the --output flag:
~/Gnomes$ dbchecksum show_update_path --output psql --dbname gnomes --from_checksum 54aa14e7b7e54cce8ae07c441f6bda316aa8458c
psql gnomes -1 -f inital_schema.sql
psql gnomes -1 -f underpants_need_washing.sql
# No update found that's based on 094ef4321e60b50c1d34529c312ecc2fcbbdfb51.
You could pipe this into changes.sh and then run that.
Or use --output concat:
~/Gnomes$ dbchecksum show_update_path --output concat --from_checksum 54aa14e7b7e54cce8ae07c441f6bda316aa8458c > changes.sql
~/Gnomes$ cat changes.sql
-- file: inital_schema.sql
-- preSHA1sum: 54aa14e7b7e54cce8ae07c441f6bda316aa8458c
-- postSHA1sum: 611481f7599cc286fa539dbeb7ea27f049744dc7
-- inital schema
create table underpants (
id serial primary key,
type text,
size text,
color text
);
-- file: underpants_need_washing.sql
-- preSHA1sum: 611481f7599cc286fa539dbeb7ea27f049744dc7
-- postSHA1sum: 094ef4321e60b50c1d34529c312ecc2fcbbdfb51
-- underpants need washing
ALTER TABLE underpants ADD COLUMN needs_washing BOOLEAN NOT NULL DEFAULT false;
-- No update found that's based on 094ef4321e60b50c1d34529c312ecc2fcbbdfb51.
Happyness!
You will only need those methods if you want to use the library itself instead of using the dbchecksum wrapper script.
my $sha1_hex = $self->checksum();
Gets the schemadump and runs it through Digest::SHA1, returning the current checksum.
my $schemadump = $self->schemadump;
Returns a string representation of the whole schema (as a Data::Dumper Dump).
Lazy Moose attribute.
my $hashref = $self->_build_schemadump_schema( $schema );
This is the main entry point for checksum calculations per schema. Method-modifiy it if you need to alter the complete schema data structure before/after checksumming.
Returns a HashRef like:
{
tables => $hash_ref
}
my $hashref = $self->_build_schemadump_tables( $schema );
Iterate through all tables in a schema, calling _build_schemadump_table for each table and collecting the results in a HashRef
my $hashref = $self->_build_schemadump_table( $schema, $table );
Get metadata on a table (columns, primary keys & foreign keys) via DBI introspection.
This is a good place to method-modify if you need some special processing for your database
Returns a hashref like
{
columns => $data,
primary_keys => $data,
foreign_keys => $data,
}
my $hashref = $self->_build_schemadump_column( $schema, $table, $column, $raw_dbi_data );
Does some cleanup on the data returned by DBI.
my $update_info = $self->update_path
Lazy Moose attribute that returns the data structure needed by apply_sql_update.
_build_update_path reads in all files ending in ".sql" in $self->sqlsnippetdir.
It builds something like a linked list of files, which are chained by their
preSHA1sum and postSHA1sum.
my ($pre, $post) = $self->get_checksums_from_snippet( $filename );
Returns a list of the preSHA1sum and postSHA1sum for the given file in sqlnippetdir.
The file has to contain this info in SQL comments, eg:
-- preSHA1sum: 89049e457886a86886a4fdf1f905b69250a8236c
-- postSHA1sum: d9a02517255045167053ea92dace728e1389f8ca
alter table foo add column bar;
Database handle (DBH::db). Moose attribute
The database catalog searched for data. Not implemented by all DBs. See DBI::table_info
Default %.
Moose attribute
An Arrayref containing names of schematas to include in checksum calculation. See DBI::table_info
Default %.
Moose attribute
Path to the directory where the sql change files are stored.
Moose attribute
Be verbose or not. Default: 0
Additional options for the specific database driver.
These options define how to connect to your database.
Required. The Data Source Name (DSN) as used by DBI to connect to your database.
Some examples: dbi:SQLite:dbname=sqlite.db,
dbi:Pg:dbname=my_project;host=db.example.com;port=5433,
dbi:Pg:service=my_project_dbadmin
Username to use to connect to your database.
Password to use to connect to your database.
These options define which parts of the schema are relevant to the checksum
Default: %
Needed during DBI introspection. Pg does not need it.
Default: % (all schemata)
If you have several schemata in your database, but only want to consider some for the checksum, use --schemata to list the ones you care about. Can be specified more than once to list several schemata:
dbchecksum apply --schemata foo --schemata bar
Some database drivers might implement further options only relevant
for the specific driver. As of now, this only applies to
DBIx::SchemaChecksum::Driver::Pg, which defines the driveropts
triggers, sequences and functions
"dbchecksum" in bin for a command line frontend powered by MooseX::App
There are quite a lot of other database schema management tools out there, but nearly all of them need to store meta-info in some magic table in your database.
You can find more information on the rational, usage & implementation in the slides for my talk at the Austrian Perl Workshop 2012, available here: http://domm.plix.at/talks/dbix_schemachecksum.html
Thanks to
- Klaus Ita and Armin Schreger for writing the initial core code. I just glued it together and improved it a bit over the years.
- revdev, a nice little software company run by Koki, domm
(http://search.cpan.org/~domm/) and Maroš (http://search.cpan.org/~maros/) from 2008 to 2011. We initially wrote
DBIx::SchemaChecksumfor our work at revdev. - validad.com which grew out of
revdev and still uses (and supports)
DBIx::SchemaChecksumevery day. - Farhad from Spherical Elephant for nagging me into writing proper docs.
- Thomas Klausner [email protected]
- Maroš Kollár [email protected]
- Klaus Ita [email protected]
This software is copyright (c) 2012 by Thomas Klausner, Maroš Kollár, Klaus Ita.
This is free software; you can redistribute it and/or modify it under the same terms as the Perl 5 programming language system itself.