Availability and reliability

更新时间:
复制 MD 格式

AnalyticDB for PostgreSQL protects your data and keeps instances running through three complementary layers: dual-replica compute redundancy, triple-replica disk storage, and remote backup. Each layer addresses a different class of failure — from a single node crashing to a full data center outage.

Failure scenarioProtection mechanism
Single compute node failureDual-replica primary/secondary architecture
Individual disk failureTriple-replica (triple-modular redundancy) disk storage
Data center outage or accidental deletionRemote backup and restoration

How it works

AnalyticDB for PostgreSQL is a distributed database built on a massively parallel processing (MPP) architecture. Distributing workloads across compute nodes improves query performance, but also increases the probability of individual node failures. The following mechanisms address this at each layer of the stack.

Dual-replica compute redundancy

Each instance runs a primary/secondary architecture with two replicas per compute node. If a single replica set has at least one available compute node, the instance continues serving queries normally. This covers the most common failure scenario: a single node going offline.

Triple-replica disk storage

Each data file is stored across three independent disks in a triple-modular redundancy configuration. This prevents data loss from individual disk failures, independent of compute-level redundancy.

Remote backup

For data center-level failures or accidental data deletion — scenarios where local redundancy is not sufficient — AnalyticDB for PostgreSQL stores remote backups that you can use to restore your instance.

Automatic failover and recovery

AnalyticDB for PostgreSQL monitors the health status of every node in real time. When the primary compute node becomes unavailable, a hot standby node takes over automatically, keeping the instance available without manual intervention.

After a node failure, the system recovers the node to restore full multi-replica redundancy. The recovery method depends on the root cause:

Root causeRecovery action
Physical machine failureMigrate the affected compute nodes to another physical machine
Suspended processRestart the process

Backup and restoration

Remote backup protects against data center-level failures and accidental data deletion.

Supported backup types

Backup typeStatusImpact on running workloads
Full logical backupAvailable
Incremental backupComing soonNo data locking; minimal impact

Incremental backup, once available, will complement full logical backup by reducing disruption to running workloads.

Restoration granularity

Restore data at the granularity that matches your situation:

GranularityDescription
InstanceRestore the entire instance from a backup
DatabaseRestore a specific database within an instance
SchemaRestore a specific schema within a database