Partition placement groups divide instances into logical partitions, each placed on a separate rack with independent power and network, allowing large distributed systems to isolate failures to a single partition while scaling to thousands of instances.
A partition placement group distributes instances across logical partitions, where each partition resides on a separate rack with its own power and network connectivity. Partitions can span multiple Availability Zones, providing both fault isolation and high scalability. This strategy is designed for large distributed systems like Hadoop, Cassandra, and Kafka that need to tolerate rack-level failures while maintaining service availability. You can specify the number of partitions (up to 7) when creating the group, and each partition can host multiple instances. Partitions offer the benefits of spread groups (fault isolation) with the scale of cluster groups, supporting thousands of instances.
You can specify up to 7 partitions per Availability Zone when creating the group
Each partition spans a single Availability Zone; multiple partitions can be in different AZs
No limit on instances per partition, but total instances limited by EC2 quotas
Each partition is isolated from failures in other partitions
Instances in the same partition share the same rack and network
You can specify the partition number when launching an instance, or let AWS automatically place it
Partition numbers are visible via
aws ec2 describe-instances
Apache Hadoop (HDFS NameNode, DataNodes across partitions)
Apache Cassandra or ScyllaDB ring deployments
Apache Kafka with replicated partitions across brokers
Elasticsearch clusters with data nodes distributed across partitions
Large-scale container orchestration (Kubernetes control plane and workers)
Any distributed system that replicates data across failure domains
To maximize fault tolerance in a partition placement group, distribute your application's replicas across different partitions. For example, in a 3-node Cassandra cluster, place each node in a different partition. If one partition experiences a rack failure, only one node is lost, and the remaining two nodes can maintain quorum. Similarly, for Kafka, place partition leaders and followers across different partitions to ensure that leader failure doesn't affect all replicas.
- https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/placement-groups.html#placement-groups-partition
- https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/placement-groups.html#placement-groups-limitations-partition
- https://awscli.amazonaws.com/v2/documentation/api/latest/reference/ec2/describe-placement-groups.html