corda/docs/source/database-migration.rst

Database Migration

Corda - the platform, and the installed third-party CorDapps store their data in a relational database (see api-persistence). When Corda is first installed, or when a new CorDapp is installed, associated tables, indexes, foreign-keys, etc. must be created. Similarly, when Corda is upgraded, or when a new version of a CorDapp is installed, their database schemas may have changed, but the existing data needs to be preserved or changed accordingly.

Corda supports multiple database management systems, so CorDapp developers need to keep this database portability requirement in mind when writing and testing the code. To address these concerns, Corda Enterprise provides a mechanism to make it straightforward to migrate from the old schemas to the new ones whilst preserving data. It does this by integrating a specialised database migration library. Also Corda Enterprise makes it easy to "lift" a CorDapp that does not handle the database migration (e.g.: the CorDapp developers did not include database migration scripts).

This document is addressed to node administrators and CorDapp developers.

• Node administrators need to understand how the manage the underlying database.
• CorDapp Developers need to understand how to write migration scripts.

"Database migrations" (or schema migrations) in this document, refers to the evolution of the database schema or the actual data that a Corda Node uses when new releases of Corda or CorDapps are installed. On a high level, this means that the Corda binaries will ship with scripts that cover everything from the creation of the schema for the initial install to changes on subsequent versions.

A Corda node runs on top of a database that contains internal node tables, vault tables and CorDapp tables. The database migration framework will handle all of these in the same way, as evolutions of schema and data. As a database migration framework, we use the open source library Liquibase.

Note

This advanced feature is only provided in Corda Enterprise. Whenever an upgraded version of Corda or a new version of a CorDapp is shipped that requires a different database schema to its predecessor, it is the responsibility of the party shipping the code (R3 in the case of Corda; the app developer in the case of a CorDapp) to also provide the migration scripts. Once such a change has been applied to the actual database, this fact is recorded in the database by the database migration library (see below), hence providing a mechanism to determine the 'version' of any given schema.

About Liquibase

Liquibase is a tool that implements an automated, version based database migration framework with support for a large number of databases. It works by maintaining a list of applied changesets. A changeset can be something very simple like adding a new column to a table. It stores each executed changeset with columns like id, author, timestamp, description, md5 hash, etc in a table called DATABASECHANGELOG. This changelog table will be read every time a migration command is run to determine what changesets need to be executed. It represents the "version" of the database (the sum of the executed changesets at any point). Changesets are scripts written in a supported format (xml, yml, sql), and should never be modified once they have been executed. Any necessary correction should be applied in a new changeset.

For documentation around Liquibase see: The Official website and Tutorial.

(Understanding how Liquibase works is highly recommended for understanding how database migrations work in Corda.)

Integration with the Corda node

By default, a node will not attempt to execute database migration scripts at startup (even when a new version has been deployed), but will check the database "version" and halt if the database is not in sync with the node, to avoid data corruption. To bring the database to the correct state we provide an advanced migration-tool.

Running the migration at startup automatically can be configured by specifying true in the database.runMigration node configuration setting (default behaviour is false). We recommend node administrators to leave the default behaviour in production, and use the migration tool to have better control. It is safe to run at startup if you have implemented the usual best practices for database management (e.g. running a backup before installing a new version, etc.).

Migration scripts structure

Corda provides migration scripts in an XML format for its internal node and vault tables. CorDapps should provide migration scripts for the tables they manage. In Corda, MappedSchemas (see api-persistence) manage JPA Entities and thus the corresponding database tables. So MappedSchemas are the natural place to point to the changelog file(s) that contain the changesets for those tables. Nodes can configure which MappedSchemas are included which means only the required tables are created. To follow standard best practices, our convention for structuring the changelogs is to have a "master" changelog file per MappedSchema that will only include release changelogs (see example below).

Example:

As a hypothetical scenario, let's suppose that at some point (maybe for security reasons) the owner column of the PersistentCashState entity needs to be stored as a hash instead of the X500 name of the owning party.

This means, as a CorDapp developer we have to do these generic steps:

1. In the PersistentCashState entity we need to replace

@Column(name = "owner_name")
var owner: AbstractParty,

with:

@Column(name = "owner_name_hash", length = MAX_HASH_HEX_SIZE)
var ownerHash: String,

2. Add a owner_key_hash column to the contract_cash_states table. (Each JPA Entity usually defines a table name as a @Table annotation.)
3. Run an update to set the owner_key_hash to the hash of the owner_name. This is needed to convert the existing data to the new (hashed) format.
4. Delete the owner_name column

Steps 2. 3. and 4. can be expressed very easily like this:

<changeSet author="R3.Corda" id="replace owner_name with owner_hash">
    <addColumn tableName="contract_cash_states">
        <column name="owner_name_hash" type="nvarchar(130)"/>
    </addColumn>
    <update tableName="contract_cash_states">
        <column name="owner_name_hash" valueComputed="hash(owner_name)"/>
    </update>
    <dropColumn tableName="contract_cash_states" columnName="owner_name"/>
</changeSet>

The PersistentCashState entity is included in the CashSchemaV1 schema, so based on the above mentioned convention we create a file cash.changelog-v2.xml with the above changeset and include in cash.changelog-master.xml.

@CordaSerializable
object CashSchemaV1 : MappedSchema(
        schemaFamily = CashSchema.javaClass, version = 1, mappedTypes = listOf(PersistentCashState::class.java)) {

    override val migrationResource = "cash.changelog-master"

<databaseChangeLog>
    <!--the original schema-->
    <include file="migration/cash.changelog-init.xml"/>

    <!--added now-->
    <include file="migration/cash.changelog-v2.xml"/>
</databaseChangeLog>

As we can see in this example, database migrations can "destroy" data, so it is therefore good practice to backup the database before executing the migration scripts.

Migration tool

The migration tool is distributed as a standalone jar file named db-migration-tool-${corda_version}.jar. It is intended to be used by Corda Enterprise node administrators.

Currently it has these features:

1. It allows running the migration on the database (--execute-migration )
2. Offers the option to inspect the actual SQL statements that will be run as part of the current migration (--dry-run )
3. Sometimes, when a node or the migration tool crashes while running migrations, Liquibase will not release the lock. This can happen during some long database operation, or when an admin kills the process. ( This cannot happen during normal operation of a node. Only during the migration process.) See: <http://www.liquibase.org/documentation/databasechangeloglock_table.html>. The tool provides a "release-lock" command that would forcibly unlock the db migration.
4. When a CorDapp that does not is ready to be deployed on a Corda Enterprise production node, using this tool, the CorDapp can be "lifted" (--create-migration-sql-for-cordapp). The reason this is needed is because those CorDapps don't handle this enterprise level concern. See below for details.

It has the following command line options:

Option	Description
--help	Print help message
--mode	Either 'NODE' or 'DOORMAN'. By default 'NODE'
--base-directory(*)	The node directory
--config-file	The name of the config file, by default 'node.conf'
--doorman-jar-path	For internal use only
--create-migration-sql-for-cordapp	Create migration files for a CorDapp. You can specify the fully qualified name of the MappedSchema class. If not specified it will generate the migration for all schemas that don't have migrations. The output directory is the base-directory, where a migration folder is created.
--dry-run	Output the database migration to the specified output file. The output directory is the base-directory. You can specify a file name or 'CONSOLE' if you want to send the output to the console.
--execute-migration	This option will run the db migration on the configured database. This is the only command that will actually write to the database.
--release-lock	Releases whatever locks are on the database change log table, in case shutdown failed.

For example:

java -jar db-migration-tool-R3.CORDA-3.0-DP3-RC01.jar --base-directory /path/to/node --execute-migration

Note

When running the migration tool, prefer using absolute paths when specifying the "base-directory".

Examples

The first time you set up your node, you will want to create the necessary database tables. Run the normal installation steps. Using the db migration tool, attempt a dry-run to inspect the output SQL:

--base-directory /path/to/node --dry-run

The output sql from the above command can be executed directly on the database or this command can be run:

--base-directory /path/to/node --execute-migration

At this point the node can be started successfully.

When upgrading, deploy the new version of Corda. Attempt to start the node. If there are database migrations in the new release, then the node will exit and will show how many changes are needed. You can then use the same commands as above, either to do a dry run or execute the migrations.

The same is true when installing or upgrading a CorDapp. Do a dry run, check the SQL, then trigger a migration.

Node administrator installing a CorDapp targeted at the open source node

The open source Corda codebase does not have support for Liquibase, so CorDapps contributed by the OS community will not have this concern addressed by their original developers.

To help Corda Enterprise users, we offer support in the migration tool for "lifting" a CorDapp to support Liquibase.

These are the steps:

1. Deploy the CorDapp on your node (copy the jar into the cordapps folder)
2. Find out the name of the MappedSchema containing the new contract state entities.
3. Call the migration tool: --base-directory /path/to/node --create-migration-sql-for-cordapp com.example.MyMappedSchema This will generate a file called my-mapped-schema.changelog-master.sql in a folder called migration in the base-directory. In case you don't specify the actual MappedSchema name, the tool will generate one SQL file for each schema defined in the CorDapp
4. Inspect the file(s) to make sure it is correct. This is a standard SQL file with some Liquibase metadata as comments.
5. Create a jar with the migration folder (by convention it could be named: originalCorDappName-migration.jar), and deploy this jar together with the CorDapp.
6. To make sure that the new migration will be used, do a dry run with the migration tool and inspect the output file.

Node administrator deploying a new version of a CorDapp developed by the OS community

This is a slightly more complicated scenario. You will have to understand the changes (if any) that happened in the latest version. If there are changes that require schema adjustments, you will have to write and test those migrations. The way to do that is to create a new changeset in the existing changelog for that CorDapp (generated as above). See Liquibase Sql Format

CorDapp developer developing a new CorDapp

CorDapp developers who decide to store contract state in custom entities can create migration files for the MappedSchema they define.

There are 2 ways of associating a migration file with a schema:

1. By overriding val migrationResource: String and pointing to a file that needs to be in the classpath.
2. By putting a file on the classpath in a migration package whose name is the hyphenated name of the schema (all supported file extensions will be appended to the name).

CorDapp developers can use any of the supported formats (XML, SQL, JSON, YAML) for the migration files they create. In case CorDapp developers distribute their CorDapps with migration files, these will be automatically applied when the CorDapp is deployed on a Corda Enterprise node. If they are deployed on an open source Corda node, then the migration will be ignored, and the database tables will be generated by Hibernate. In case CorDapp developers don't distribute a CorDapp with migration files, then the organisation that decides to deploy this CordApp on a Corda Enterprise node has the responsibility to manage the database.

During development or demo on the default H2 database, then the CorDapp will just work when deployed even if there are no migration scripts, by relying on the primitive migration tool provided by Hibernate, which is not intended for production.

Warning

A very important aspect to be remembered is that the CorDapp will have to work on all supported Corda databases. It is the responsibility of the developers to test the migration scripts and the CorDapp against all the databases. In the future we will provide aditional tooling to assist with this aspect.

When developing a new version of an existing CorDapp, depending on the changes to the PersistentEntities, a changelog will have to be created as per the Liquibase documentation and the example above.

Troubleshooting

When seeing problems acquiring the lock, with output like this:

Waiting for changelog lock....
Waiting for changelog lock....
Waiting for changelog lock....
Waiting for changelog lock....
Waiting for changelog lock....
Waiting for changelog lock....
Waiting for changelog lock....
Liquibase Update Failed: Could not acquire change log lock.  Currently locked by SomeComputer (192.168.15.X) since 2013-03-20 13:39
SEVERE 2013-03-20 16:59:liquibase: Could not acquire change log lock.  Currently locked by SomeComputer (192.168.15.X) since 2013-03-20 13:39
liquibase.exception.LockException: Could not acquire change log lock.  Currently locked by SomeComputer (192.168.15.X) since 2013-03-20 13:39
        at liquibase.lockservice.LockService.waitForLock(LockService.java:81)
        at liquibase.Liquibase.tag(Liquibase.java:507)
        at liquibase.integration.commandline.Main.doMigration(Main.java:643)
        at liquibase.integration.commandline.Main.main(Main.java:116)

then the advice at this StackOverflow question may be useful. You can run --base-directory /path/to/node --release-lock to force Liquibase to give up the lock.