cqlmigrate is a library for performing schema migrations on a cassandra cluster including:
It is best used as an application dependency, but can also be used standalone.
repositories {
mavenCentral()
}
compile 'uk.sky:cqlmigrate:0.13.0'
<dependency>
<groupId>uk.sky</groupId>
<artifactId>cqlmigrate</artifactId>
<version>0.13.0</version>
</dependency>
The locks keyspace and table needs to be created before running any migrations.
CREATE KEYSPACE IF NOT EXISTS cqlmigrate WITH replication = {'class': 'REPLICATION_CLASS', 'replication_factor': REPLICATION_FACTOR };
CREATE TABLE IF NOT EXISTS cqlmigrate.locks (name text PRIMARY KEY, client text);
To apply all .cql
files located in /cql
in the classpath:
import com.datastax.oss.driver.api.core.*;
// Configure locking for coordination of multiple nodes
CassandraLockConfig lockConfig = CassandraLockConfig.builder()
.withTimeout(Duration.ofSeconds(3))
.withPollingInterval(Duration.ofMillis(500))
.withConsistencyLevel(ConsistencyLevel.ALL)
.withLockKeyspace("cqlmigrate")
.build();
// Configure cql migrator
CqlMigratorConfig cqlMigratorConfig = CqlMigratorConfig.builder()
.withLockConfig(lockConfig)
.withReadConsistencyLevel(ConsistencyLevel.LOCAL_ONE)
.withWriteConsistencyLevel(ConsistencyLevel.ALL)
.withTableCheckerInitDelay(Duration.ofSeconds(5))
.withTableCheckerTimeout(Duration.ofMinutes(1))
.build()
// Create a Cassandra session for cassandra driver 4.x
CqlSession session = CqlSession.builder()
.addContactPoints(cassandraHosts)
.withLocalDatacenter("datacenter1")
.withAuthProvider(new ProgrammaticPlainTextAuthProvider("username", "password"))
.build();
// Create a migrator and run it
CqlMigrator migrator = CqlMigratorFactory.create(lockConfig);
Path schemas = Paths.get(ClassLoader.getSystemResource("/cql").toURI());
migrator.migrate(session, "my_keyspace", asList(schemas));
The migrator will look for a bootstrap.cql
file for setting up the keyspace.
$ java -Dhosts=localhost,192.168.1.1 -Dport=9042 -DlocalDC=DC1 -Dkeyspace=my_keyspace -Ddirectories=cql-common,cql-local -jar cqlmigrate.jar
Specify credentials, if required, using -Dusername=<username>
and -Dpassword=<password>
.
Specify optional properties, if required, using
-Dprecheck=<true/false>
default false
-DtableCheckerTimeout=<duration>
default PT1M
(This was introduced for AWS Keyspaces as when new table is created it might take time to allocate resources and if additional changes or data are inserted while table is in CREATING
state, it will fail. If cluster is no AWS Keyspaces it will not wait.)-DtableCheckerInitDelay=<duration>
default PT5S
(Supplement previous option as it might take time AWs Keyspaces update their system so we need to wait before we check first time.)-DreadCL
default LOCAL_ONE
-DwriteCL
default ALL
Checks all nodes are up and their schemas are in agreement.
Tries to acquire a lock for the keyspace. If it can't initially be acquired it will continue to retry at a set polling time until the timeout is reached.
Looks for a bootstrap.cql
file and runs it first. This file should contain the keyspace definition:
CREATE KEYSPACE my_keyspace
WITH replication = {'class': 'SimpleStrategy', 'replication_factor': 1 };
Applies .cql
files one by one, sorted by filename in ascending order. It is suggested to prefix
the files with a datetime to order them:
/cql/2015-05-02-19:20-create-airplanes-table.cql
/cql/2015-05-03-14:19-add-manufacturer-column.cql
Any previously applied files will be skipped. For AWS Keyspaces, it will wait after each successfully applied file for tables to get into ACTIVE state for initial
Releases the lock.
This table is used to determine what has been previously applied.
SELECT * FROM schema_updates;
filename | applied_on | checksum
----------------------------------+--------------------------+------------------------------------------
0001-create-tables.cql | 2015-04-08 12:10:04+0100 | ec19dfac7ede62b2a40c0f39706b237cd5c30da6
0002-dataset.cql | 2015-04-08 12:10:04+0100 | 4fa2d6c4fae9950f0c9140ae2eb57fe689192b4a
0003-initial-date.cql | 2015-04-08 12:10:04+0100 | 19d0c9522b6464a06b18192c6e04233f83e78a84
(3 rows)
It also maintains a checksum to ensure the script hasn't changed since it was last applied.
The locks keyspace replication class and factor can be configured using the LocksConfig. This table is used to keep track of what locks are currently in place, and relies on Cassandra's lightweight transactions.
SELECT * FROM locks;
name | client
-------------------------------------+--------------------------------------
airplanes_keyspace.schema_migration | 2a4ec2ae-d3d1-4b33-86a9-eb844e35eeeb
(1 rows)
Each lock will be deleted by cqlmigrate
once the migration is complete.
This project has been tested against the following versions:
Cassandra is an eventually consistent, AP database, and so applying schema updates are not as simple as a traditional relational database.
Certain schema changes can cause data corruption on cassandra. Be very careful changing a schema for a table that is being actively written to. Generally, adding columns is safe (but be careful with collection types). Test the migration before rolling out to production.
AP properties of Cassandra also apply to schema updates - so it is possible for a cluster to have an
inconsistent schema across nodes in case of split brain or other situation. cqlmigrate
tries to
alleviate this with appropriate consistency levels.
There are a number of ways to add comments to your cql
files.
For inline comments prepend --
to your comment, e.g:
-- Select Queries
SELECT * FROM schema_updates;
For multiline comments wrap them with /*
and */
, e.g:
/*
Added by John Smith
19th September 2017
*/
SELECT * FROM schema_updates;
Originally developed by the Cirrus team at Sky.