Production Deployment

This section is not meant to be an exhaustive guide to running your Schema Registry in production, but it covers the key things to consider before putting your cluster live. Three main areas are covered:

  • Logistical considerations, such as hardware recommendations and deployment strategies
  • Configuration more suited to a production environment
  • Post-deployment considerations, such multi-data center setup

Hardware

If you’ve been following the normal development path, you’ve probably been playing with Schema Registry on your laptop or on a small cluster of machines laying around. But when it comes time to deploying Schema Registry to production, there are a few recommendations that you should consider. Nothing is a hard-and-fast rule.

Memory

Schema Registry uses Kafka as a commit log to store all registered schemas durably, and maintains a few in-memory indices to make schema lookups faster. A conservative upper bound on the number of unique schemas registered in a large data-oriented company like LinkedIn is around 10,000. Assuming roughly 1000 bytes heap overhead per schema on average, heap size of 1GB would be more than sufficient.

CPUs

CPU usage in Schema Registry is light. The most computationally intensive task is checking compatibility of two schemas, an infrequent operation which occurs primarily when new schemas versions are registered under a subject.

If you need to choose between faster CPUs or more cores, choose more cores. The extra concurrency that multiple cores offers will far outweigh a slightly faster clock speed.

Disks

Schema Registry does not have any disk resident data. It currently uses Kafka as a commit log to store all schemas durably and holds in-memory indices of all schemas. Therefore, the only disk usage comes from storing the log4j logs.

Network

A fast and reliable network is obviously important to performance in a distributed system. Low latency helps ensure that nodes can communicate easily, while high bandwidth helps shard movement and recovery. Modern data-center networking (1 GbE, 10 GbE) is sufficient for the vast majority of clusters.

Avoid clusters that span multiple data centers, even if the data centers are colocated in close proximity. Definitely avoid clusters that span large geographic distances.

Larger latencies tend to exacerbate problems in distributed systems and make debugging and resolution more difficult.

Often, people might assume the pipe between multiple data centers is robust or low latency. But this is usually not true and network failures might happen at some point. Please refer to our recommended Multi-DC Setup.

JVM

We recommend running JDK 1.7 u51, and using the G1 collector. If you do this (and we highly recommend it), make sure you’re on u51. We tried out u21 in testing, but we had a number of problems with the GC implementation in that version. Our recommended GC tuning looks like this:

-Xms1g -Xmx1g -XX:PermSize=48m -XX:MaxPermSize=48m -XX:+UseG1GC -XX:MaxGCPauseMillis=20 \
       -XX:InitiatingHeapOccupancyPercent=35

Important Configuration Options

The full set of configuration options are documented in Configuration Options.

However, there are some logistical configurations that should be changed for production. These changes are necessary because there is no way to set a good default (because it depends on your cluster layout).

kafkastore.connection.url Zookeeper url for the Kafka cluster

  • Type: string
  • Importance: high

port Port to listen on for new connections.

  • Type: int
  • Default: 8081
  • Importance: high

host.name Hostname to publish to ZooKeeper for clients to use. In IaaS environments, this may need to be different from the interface to which the broker binds. If this is not set, it will use the value returned from java.net.InetAddress.getCanonicalHostName().

  • Type: string
  • Default: host.name
  • Importance: high

Note

Configure min.insync.replicas on the Kafka server for the schemas topic that stores all registered schemas to be higher than 1. For example, if the kafkastore.topic.replication.factor is 3, then set min.insync.replicas on the Kafka server for the kafkastore.topic to 2. This ensures that the register schema write is considered durable if it gets committed on at least 2 replicas out of 3. Furthermore, it is best to set unclean.leader.election.enable to false so that a replica outside of the isr is never elected leader (potentially resulting in data loss).

Don’t Touch These Settings!

Storage settings

Schema Registry stores all schemas in a Kafka topic defined by kafkastore.topic. Since this Kafka topic acts as the commit log for the Schema Registry database and is the source of truth, writes to this store need to be durable. Schema Registry ships with very good defaults for all settings that affect the durability of writes to the Kafka based commit log. Whenever in doubt, leave these settings alone.

kafkastore.topic The single partition topic that acts as the durable log for the data

  • Type: string
  • Default: “_schemas”
  • Importance: high

kafkastore.topic.replication.factor The desired replication factor of the schema topic. The actual replication factor will be the smaller of this value and the number of live Kafka brokers.

  • Type: int
  • Default: 3
  • Importance: high

kafkastore.timeout.ms The timeout for an operation on the Kafka store. This is the maximum time that a register call blocks.

  • Type: int
  • Default: 500
  • Importance: medium

Kafka & ZooKeeper

Please refer to Zookeeper for recommendations on operationalizing Kafka and ZooKeeper.

Multi-DC Setup

Overview

Spanning multiple datacenters with your Schema Registry provides additional protection against data loss and improved latency. The recommended multi-datacenter deployment designates one datacenter as “master” and all others as “slaves”. If the “master” datacenter fails and is unrecoverable, a “slave” datacenter will need to be manually designated the new “master” through the steps in the Run Book below.

Important Settings

kafkastore.connection.url kafkastore.connection.url should be identical across all schema registry nodes. By sharing this setting, all Schema Registry instances will point to the same ZooKeeper cluster.

schema.registry.zk.namespace Namespace under which schema registry related metadata is stored in Zookeeper. This setting should be identical across all nodes in the same schema registry.

master.eligibility A schema registry server with master.eligibility set to false is guaranteed to remain a slave during any master election. Schema Registry instances in a “slave” data center should have this set to false, and Schema Registry instances local to the shared Kafka cluster should have this set to true.

Setup

Assuming you have Schema Registry running, here are the recommended steps to add Schema Registry instances in a new “slave” datacenter (call it DC B):

  • In DC B, make sure Kafka has unclean.leader.election.enable set to false.
  • In Kafka in DC B, create the _schemas topic. It should have 1 partition, kafkastore.topic.replication.factor of 3, and min.insync.replicas at least 2.
  • In DC B, run MirrorMaker with Kafka in the “master” datacenter as the source and Kafka in DC B as the target.
  • In the Schema Registry config files in DC B, set kafkastore.connection.url and schema.registry.zk.namespace to match the instances already running, and set master.eligibility to false.
  • Start your new Schema Registry instances with these configs.

Run Book

Let’s say you have Schema Registry running in multiple datacenters, and you have lost your “master” datacenter; what do you do? First, note that the remaining Schema Registry instances will continue to be able to serve any request which does not result in a write to Kafka. This includes GET requests on existing ids and POST requests on schemas already in the registry.

  • If possible, revive the “master” datacenter by starting Kafka and Schema Registry as before.
  • If you must designate a new datacenter (call it DC B) as “master”, update the Schema Registry config files so that kafkastore.connection.url points to the local ZooKeeper, and change master.eligibility to true. The restart your Schema Registry instances with these new configs in a rolling fashion.