Apache Doris Installation Guide

Choose a path

Make two decisions before you deploy.

First choose the storage architecture. Then choose who will orchestrate the processes. Those two choices produce four deployment paths.

VMManual · Integrated

FE and stateful BE processes on hosts with local disks.

VMManual · Decoupled

FE, Meta Service, stateless BE, FoundationDB, and shared storage.

K8SOperator · Integrated

DorisCluster resources manage FE and stateful BE workloads.

K8SOperator · Decoupled

The Operator manages a cloud-mode cluster and its compute groups.

Prerequisites

Remove environmental surprises first.

A deployment is only as reliable as its disks, network identity, clocks, ports, and operating-system preparation.

CPUCompatible hosts

Confirm instruction set, kernel, limits, and Java requirements.

NETReachable services

Allow FE, BE, and web service ports between required nodes.

NTPSynchronized time

Keep clocks aligned for coordination, logs, and diagnosis.

SSDDurable paths

Plan FE metadata, BE data, logs, and shared storage deliberately.

Manual integrated

Bring up control, then add storage.

Initialize the first FE, expand the FE quorum, register BE nodes, and verify membership before loading data.

First FEInitialize metadata
FE quorumAdd voting members
BE nodesRegister storage paths
VerifyCheck health and secure

Manual decoupled

Build the metadata and storage plane before compute.

A decoupled cluster adds FoundationDB, Meta Service, and a Storage Vault around the familiar FE and BE processes.

01
FoundationDB + Meta Service
Persist and coordinate data-layer metadata.
02
Cloud-mode FE + compute groups
Plan SQL and run elastic work.
03
Storage Vault
Connect durable object storage or a distributed file system.

Kubernetes integrated

Describe the cluster; let the Operator reconcile it.

The Doris Operator turns a DorisCluster resource into FE and BE workloads, services, configuration, and persistent storage.

InstallCRDs + Operator
ConfigureImages + ConfigMaps
PersistPVCs + services
ApplyDorisCluster
ObservePods + SQL health

Kubernetes decoupled

Treat dependencies as part of the platform.

Kubernetes automates Doris processes, but it does not remove responsibility for FoundationDB, shared storage, credentials, or failure planning.

FDBMetadata store

Redundant FoundationDB with monitored capacity.

MSMeta Service

Consistent cluster identity and reachable endpoints.

CGCompute groups

Stateless workers with appropriate resources and cache.

OBJShared storage

Vault credentials, permissions, durability, and throughput.

Operate and verify

Verify signals at every layer.

Good deployment work ends with health checks, security, observability, scaling rules, and a rehearsed upgrade path.

SQLMembership

Verify FE roles, BE heartbeats, and compute-group state.

SECSecurity

Set credentials, restrict exposure, and protect storage secrets.

OBSObservability

Collect metrics and logs before the first incident.

DRRecovery

Test restart, replacement, backup, and upgrade procedures.

Deployment recap

A deployment path is a set of owned tradeoffs.

Choose the least complex path that satisfies storage, elasticity, availability, and platform requirements—and validate it with real data.

Prefer integrated

Choose operational simplicity when local storage and coupled scaling fit.

Fewer platform dependencies
Stateful BE nodes
Direct local-disk ownership

Prefer decoupled

Choose independent scaling when shared storage and platform ownership are available.

Elastic compute groups
Durable shared storage
Additional metadata services