Prerequisites
This page covers requirements shared across all deployment types. Environment-specific requirements (Kubernetes version, Helm, Docker, --net=host, libfuse) are listed inline in each installation guide.
All Deployments
Alluxio components run on Java 11 — the JDK is bundled in the provided container images; no host installation needed.
Hardware
CPU architecture
x86_64 or ARM64
OS
Linux (kernel 5.4 or higher)
Networking
Open the following ports between all hosts running Alluxio components, and between your application nodes and the Alluxio workers:
2379
etcd
Client requests
2380
etcd
Peer communication
19998
Coordinator
RPC
19999
Coordinator
REST API / Prometheus metrics
29998
Worker
S3 API
29999
Worker
RPC
30000
Worker
REST API / Prometheus metrics
49998
FUSE
RPC
49999
FUSE
Prometheus metrics
On Kubernetes, these ports are managed by pod-to-pod network policy. Verify that your CNI plugin does not block intra-cluster traffic on these ports.
Resource Sizing
These are starting points. Adjust based on dataset size, client concurrency, and target throughput.
K8s node
CPU / Memory
8 cores / 32 GB
16 cores / 64 GB
Coordinator
CPU / Memory
4 cores / 12 GB
8 cores / 16 GB
Worker
CPU / Memory
4 cores / 12 GB
8 cores / 24 GB
FUSE
CPU / Memory
4 cores / 8 GB
8 cores / 20 GB
Worker cache disk
Per worker
100 GB
100 GB+
The K8s node row applies only to Kubernetes deployments. For Docker / bare-metal, size each host to the component it runs.
etcd
For evaluation, a single etcd node is sufficient.
For production, deploy 3 nodes on dedicated, stable hosts:
Avoid nodes with heavy I/O load — etcd is sensitive to disk latency
Avoid spot or preemptible instances — if quorum is lost the Alluxio cluster must be recreated
Quorum rule: With 3 nodes, at least 2 must be running at all times
On Kubernetes, the Alluxio Operator provisions etcd automatically. An external etcd cluster is also supported — see Appendix: Use External etcd.
Next Steps
Last updated