section> Computing
  • Auto Scaling
  • Bare Metal Server
  • Dedicated Host
  • Elastic Cloud Server
  • FunctionGraph
  • Image Management Service
Network
  • Direct Connect
  • Domain Name Service
  • Elastic IP
  • Elastic Load Balancing
  • Enterprise Router
  • NAT Gateway
  • Private Link Access Service
  • Secure Mail Gateway
  • Virtual Private Cloud
  • Virtual Private Network
  • VPC Endpoint
Storage
  • Cloud Backup and Recovery
  • Cloud Server Backup Service
  • Elastic Volume Service
  • Object Storage Service
  • Scalable File Service
  • Storage Disaster Recovery Service
  • Volume Backup Service
Application
  • API Gateway (APIG)
  • Application Operations Management
  • Application Performance Management
  • Distributed Message Service (for Kafka)
  • Simple Message Notification
Data Analysis
  • Cloud Search Service
  • Data Lake Insight
  • Data Warehouse Service
  • DataArts Studio
  • MapReduce Service
  • ModelArts
  • Optical Character Recognition
Container
  • Application Service Mesh
  • Cloud Container Engine
  • Cloud Container Instance
  • Software Repository for Containers
Databases
  • Data Replication Service
  • Distributed Cache Service
  • Distributed Database Middleware
  • Document Database Service
  • GeminiDB
  • Relational Database Service
  • TaurusDB
Management & Deployment
  • Cloud Create
  • Cloud Eye
  • Cloud Trace Service
  • Config
  • Log Tank Service
  • Resource Formation Service
  • Tag Management Service
Security Services
  • Anti-DDoS
  • Cloud Firewall
  • Database Security Service
  • Dedicated Web Application Firewall
  • Host Security Service
  • Identity and Access Management
  • Key Management Service
  • Web Application Firewall
Other
  • Enterprise Dashboard
  • Marketplace
  • Price Calculator
  • Status Dashboard
APIs
  • REST API
  • API Usage Guidelines
  • Endpoints
Development and Automation
  • SDKs
  • Drivers and Tools
  • Terraform
  • Ansible
  • Cloud Create
Architecture Center
  • Best Practices
  • Blueprints
IaaSComputingAuto ScalingBare Metal ServerDedicated HostElastic Cloud ServerFunctionGraphImage Management ServiceNetworkDirect ConnectDomain Name ServiceElastic IPElastic Load BalancingEnterprise RouterNAT GatewayPrivate Link Access ServiceSecure Mail GatewayVirtual Private CloudVirtual Private NetworkVPC EndpointStorageCloud Backup and RecoveryCloud Server Backup ServiceElastic Volume ServiceObject Storage ServiceScalable File ServiceStorage Disaster Recovery ServiceVolume Backup ServicePaaSApplicationAPI Gateway (APIG)Application Operations ManagementApplication Performance ManagementDistributed Message Service (for Kafka)Simple Message NotificationData AnalysisCloud Search ServiceData Lake InsightData Warehouse ServiceDataArts StudioMapReduce ServiceModelArtsOptical Character RecognitionContainerApplication Service MeshCloud Container EngineCloud Container InstanceSoftware Repository for ContainersDatabasesData Replication ServiceDistributed Cache ServiceDistributed Database MiddlewareDocument Database ServiceGeminiDBRelational Database ServiceTaurusDBManagementManagement & DeploymentCloud CreateCloud EyeCloud Trace ServiceConfigLog Tank ServiceResource Formation ServiceTag Management ServiceSecuritySecurity ServicesAnti-DDoSCloud FirewallDatabase Security ServiceDedicated Web Application FirewallHost Security ServiceIdentity and Access ManagementKey Management ServiceWeb Application FirewallOtherOtherEnterprise DashboardMarketplacePrice CalculatorStatus Dashboard

MapReduce Service

  • Using CarbonData
  • Using CDL
  • Using ClickHouse
  • Using DBService
  • Using Flink
    • Using Flink from Scratch
    • Viewing Flink Job Information
    • Flink Configuration Management
    • Security Configuration
    • Security Hardening
    • Security Statement
    • Using the Flink Web UI
    • Deleting Residual Information About Flink Tasks
    • Flink Log Overview
    • Flink Performance Tuning
    • Common Flink Shell Commands
    • Reference
    • Flink Restart Policy
  • Using Flume
  • Using HBase
  • Using HDFS
  • Using HetuEngine
  • Using Hive
  • Using Hudi
  • Using Hue
  • Using IoTDB
  • Using Kafka
  • Using Loader
  • Using MapReduce
  • Using Oozie
  • Using Ranger
  • Using Spark2x
  • Using Tez
  • Using Yarn
  • Using ZooKeeper
  • Appendix
  • Change History
  • Component Operation Guide (LTS)
  • Using Flink
  • Flink Restart Policy

Flink Restart Policy¶

Overview¶

Flink supports different restart policies to control whether and how to restart a job when a fault occurs. If no restart policy is specified, the cluster uses the default restart policy. You can also specify a restart policy when submitting a job. For details about how to configure such a policy on the job development page of MRS 3.1.0 or later, see Managing Jobs on the Flink Web UI.

The restart policy can be specified by configuring the restart-strategy parameter in the Flink configuration file Client installation directory/Flink/flink/conf/flink-conf.yaml or can be dynamically specified in the application code. The configuration takes effect globally. Restart policies include failure-rate and the following two default policies:

  • No restart: If CheckPoint is not enabled, this policy is used by default.

  • Fixed-delay: If CheckPoint is enabled but no restart policy is configured, this policy is used by default.

No restart Policy¶

When a fault occurs, the job fails and does not attempt to restart.

Configure the parameter as follows:

restart-strategy: none

fixed-delay Policy¶

When a fault occurs, the job attempts to restart for a fixed number of times. If the number of attempts exceeds the times you specified, the job fails. The restart policy waits for a fixed period of time between two consecutive restart attempts.

In the following example, a job fails if the job attempts to restart for three times at an interval of 10 seconds. Configure the parameters as follows:

restart-strategy: fixed-delay
restart-strategy.fixed-delay.attempts: 3
restart-strategy.fixed-delay.delay: 10 s

failure-rate Policy¶

When a job fails, the job restarts directly. If the failure rate exceeds the value you configured, the job is considered as failed. The restart policy waits for a fixed period of time between two consecutive restart attempts.

In the following example, a job is considered as failed if the job attempts to restart for three times at an interval of 10 minutes. Configure the parameters as follows:

restart-strategy: failure-rate
restart-strategy.failure-rate.max-failures-per-interval: 3
restart-strategy.failure-rate.failure-rate-interval: 10 min
restart-strategy.failure-rate.delay: 10 s

Choosing a Restart Policy¶

  • If you do not want to retry a failed job, select the No restart policy.

  • To retry a failed job, select the failure-rate policy. If the fixed-delay policy is used, the number of job failures may reach the maximum number of retries due to hardware faults such as network and memory faults. As a result, the job fails.

    To prevent repeated restarts when the failure-rate policy is used, configure parameters as follows:

    restart-strategy: failure-rate
    restart-strategy.failure-rate.max-failures-per-interval: 3
    restart-strategy.failure-rate.failure-rate-interval: 10 min
    restart-strategy.failure-rate.delay: 10 s
    
  • Prev
  • Next
last updated: 2025-07-09 15:07 UTC - commit: cb943fa3145d5c3e150bb4fa1a987d24c3077fe9
Edit pageReport Documentation Bug
Page Contents
  • Flink Restart Policy
    • Overview
    • No restart Policy
    • fixed-delay Policy
    • failure-rate Policy
    • Choosing a Restart Policy
© T-Systems International GmbH
  • Contact
  • Data privacy
  • Disclaimer of Liabilities
  • Imprint