If you are not migrating an existing Orca deployment, refer to Orca SQL Setup instead.
Migrate from Redis to SQL
Migrating without downtime from Redis to SQL is a three-step process:
- Deploy Orca with the
DualExecutionRepositorywriting to both Redis and SQL.
- Deploy a new Orca cluster with migrators enabled and queue processing disabled.
- Once all executions have been migrated, delete migration cluster and disable
DualExecutionRepository is running, writes will be routed to either Redis or SQL.
Executions will only be migrated to SQL once they’ve completed (either successfully or terminally): This keeps the migration story simple.
As such, the migration agents will need to run for awhile.
At Netflix, we ran the migration cluster for two weeks, as we had long pipeline executions due to canaries.
You may only need to run the migration cluster for an hour.
NOTE: Deploying the migrators as a separate cluster is optional, however the migration process is memory hungry, so you may need to devote more resources to the Orca process.
Building atop the baseline configuration above, add the following to orca.yml:
executionRepository: dual: enabled: true primaryClass: com.netflix.spinnaker.orca.sql.telemetry.SqlInstrumentedExecutionRepository previousClass: com.netflix.spinnaker.orca.telemetry.RedisInstrumentedExecutionRepository sql: enabled: true redis: enabled: true
Note that both repositories are enabled. Orca will fail to start up if the
DualExecutionRepository is misconfigured.
Deploy a Migration Cluster
At Netflix, we deployed
orca-main, the cluster that serves our production traffic, as well as
orca-main-sqlmigration, which does not receive API traffic nor process the work queue. It’s sole purpose is to shovel bits from Redis to SQL.
To perform a deploy a migration cluster, add the following configuration to
--- spring: profiles: sqlmigration pollers: orchestrationMigrator: enabled: true intervalMs: 1800000 pipelineMigrator: enabled: true intervalMs: 1800000 queue: redis: enabled: false keiko: queue: enabled: false
You will need to launch this migration cluster with
Spring Profiles allow you to configure a service to startup with different configurations.
Once all executions have been migrated, you can deploy Orca once again without
DualExecutionRepository and delete the migration cluster.
Orca emits a few handfuls of metrics that are specific to SQL.
SqlHealthcheckQueueActivator class, which will disable work queue processing if SQL connectivity goes unhealthy, emits one metric:
sql.queueActivator.invocations with a tag of
ExecutionRepository will emit a bunch of invocation and timing metrics with the following patterns:
If you are using the default HikariCP connection pool:
If you are using the MariaDB driver extension: