Asynchronous Replication

You can perform deployment using unidirectional (master-follower) or bidirectional (multi-master) asynchronous replication between universes (also known as data centers).

For information on two data center (2DC) deployment architecture and supported replication scenarios, see Two data center (2DC) deployments.

Setting Up Universes

You can create source and target universes as follows:

1. Create the yugabyte-source universe by following the procedure from Manual deployment.
2. Create tables for the APIs being used by the source universe.
3. Create the yugabyte-target universe by following the procedure from Manual deployment.
4. Create tables for the APIs being used by the target universe. These should be the same tables as you created for the source universe.
5. Proceed to setting up unidirectional or bidirectional replication.

Setting Up Unidirectional Replication

After you created the required tables, you can set up asynchronous replication as follows:

• Look up the source universe UUID and the table IDs for the two tables and the index table:

  • To find a universe's UUID, check /varz for --cluster_uuid. If it is not available in this location, check the same field in the cluster configuration.

  • To find a table ID, execute the following command as an admin user:

    yb-admin list_tables include_table_id
    

• Run the following yb-admin setup_universe_replication command from the YugabyteDB home directory in the source universe:

  ./bin/yb-admin \
    -master_addresses <target_universe_master_addresses> \
    setup_universe_replication <source_universe_uuid> \
      <source_universe_master_addresses> \
      <table_id>,[<table_id>..]
  

  For example:

  ./bin/yb-admin \
    -master_addresses 127.0.0.11:7100,127.0.0.12:7100,127.0.0.13:7100 \
    setup_universe_replication e260b8b6-e89f-4505-bb8e-b31f74aa29f3 \
      127.0.0.1:7100,127.0.0.2:7100,127.0.0.3:7100 \
      000030a5000030008000000000004000,000030a5000030008000000000004005,dfef757c415c4b2cacc9315b8acb539a
  

The preceding command contains three table IDs: the first two are YSQL for the base table and index, and the third is the YCQL table.

Also, be sure to specify all master addresses for source and target universes in the command.

If you need to set up bidirectional replication, see instructions provided in Setting Up Bidirectional Replication. Otherwise, proceed to Loading Data into the Source Universe.

Setting Up Bidirectional Replication

To set up bidirectional replication, repeat the procedure described in Setting Up Unidirectional Replication applying the steps to the yugabyte-target universe. You need to set up each yugabyte-source to consume data from yugabyte-target.

When completed, proceed to Loading Data.

Loading Data into the Source Universe

Once you have set up replication, load data into the source universe as follows:

• Download the YugabyteDB workload generator JAR file yb-sample-apps.jar from GitHub.

• Start loading data into yugabyte-source by following examples for YSQL or YCQL:

  • YSQL:

  java -jar yb-sample-apps.jar --workload SqlSecondaryIndex --nodes 127.0.0.1:5433
  

  • YCQL:

  java -jar yb-sample-apps.jar --workload CassandraBatchKeyValue --nodes 127.0.0.1:9042
  

  Note that the IP address needs to correspond to the IP of any T-Servers in the cluster.

• For bidirectional replication, repeat the preceding step in the yugabyte-target universe.

When completed, proceed to Verifying Replication.

Verifying Replication

You can verify replication by stopping the workload and then using the COUNT(*) function on the yugabyte-target to yugabyte-source match.

Unidirectional Replication

For unidirectional replication, connect to the yugabyte-target universe using the YSQL shell (ysqlsh) or the YCQL shell (ycqlsh), and confirm that you can see the expected records.

Bidirectional Replication

For bidirectional replication, repeat the procedure described in Unidirectional Replication, but reverse the source and destination information, as follows:

1. Run yb-admin setup_universe_replication on the yugabyte-target universe, pointing to yugabyte-source.
2. Use the workload generator to start loading data into the yugabyte-target universe.
3. Verify replication from yugabyte-target to yugabyte-source.

To avoid primary key conflict errors, keep the key ranges for the two universes separate. This is done automatically by the applications included in the yb-sample-apps.jar.

Replication Lag

Replication lag is computed at the tablet level as follows:

replication lag = hybrid_clock_time - last_read_hybrid_time

hybrid_clock_time is the hybrid clock timestamp on the source's tablet-server, and last_read_hybrid_time is the hybrid clock timestamp of the latest record pulled from the source.

An example script determine_replication_lag.sh calculates the replication lag. The script requires the jq package.

The following example generates a replication lag summary for all tables on a cluster. You can also request an individual table.

To obtain a summary of all command options, execute determine_repl_latency.sh -h .

Setting Up Replication with TLS

Source and Target Universes have the Same Certificates

If both universes use the same certificates, run yb-admin setup_universe_replication and include the -certs_dir_name flag. Setting that to the target universe's certificate directory will make replication use those certificates for connecting to both universes.

For example:

Source and Target Universes have Different Certificates

When both universes use different certificates, you need to store the certificates for the producer universe on the target universe:

1. Ensure that use_node_to_node_encryption is set to true on all masters and tservers on both the source and target.

2. For each master and tserver on the target universe, set the gflag certs_for_cdc_dir to the parent directory where you will store all the source universe's certs for replication.

3. Find the certificate authority file used by the source universe (ca.crt). This should be stored within the --certs_dir.

4. Copy this file to each node on the target. It needs to be copied to a directory named: <certs_for_cdc_dir>/<source_universe_uuid>.

   For example, if you previously set certs_for_cdc_dir=/home/yugabyte/yugabyte_producer_certs, and the source universe's ID is 00000000-1111-2222-3333-444444444444, then you would need to copy the cert file to /home/yugabyte/yugabyte_producer_certs/00000000-1111-2222-3333-444444444444/ca.crt.

5. Finally, set up replication using yb-admin setup_universe_replication, making sure to also set the -certs_dir_name flag to the directory with the target universe's certificates (this should be different from the directory used in the previous steps).

   For example, if you have the target's certificates in /home/yugabyte/yugabyte-tls-config, then you would run:

   ./bin/yb-admin -master_addresses 127.0.0.11:7100,127.0.0.12:7100,127.0.0.13:7100 \
     -certs_dir_name /home/yugabyte/yugabyte-tls-config \
     setup_universe_replication 00000000-1111-2222-3333-444444444444 \
     127.0.0.1:7100,127.0.0.2:7100,127.0.0.3:7100 \
     000030a5000030008000000000004000,000030a5000030008000000000004005,dfef757c415c4b2cacc9315b8acb539a
   

Bootstrapping a sink cluster

Documentation coming soon. More details in #6870.

Schema migration

This section describs how to execute DDL operations, after replication has been already configured for some tables.

Stopping user writes

Some use cases can afford to temporarily stop incoming user writes. For such cases, the typical runbook would be:

• Stop any new incoming user writes.
• Wait for all changes to get replicated to the sink cluster. This can be observed by replication lag dropping to 0.
• Apply the DDL changes on both sides.
  • Alter replication for any newly created tables, eg: after having used CREATE TABLE / CREATE INDEX.
• Resume user writes.

Using backup and restore

In the event you cannot stop incoming user traffic, then the recommended approach would be to apply DDLs on the source cluster and use bootstrapping a sink cluster flow above. A more detailed runbook would be as follows:

• Stop replication, in advance of DDL changes.
• Apply all your DDL changes to the source cluster.
• Take a backup of the source cluster, of all the relevant tables that you intend to replicate changes for.
  • Make sure to use the bootstrapping a sink cluster flow, as described above.
• Restore this backup on the sink cluster.
• Setup replication again, for all of the relevant tables.
  • Make sure to pass in the bootstrap_ids, as described above.