Fault Tolerance Strategy

Fault tolerance strategy in dubbo

Dubbo offers a variety of fault-tolerant scenarios when a cluster call fails, with a default failover retry.

cluster

The relationship between nodes:

  • This Invoker is the callable Service’s abstract of the Provider, and the Invoker packaging the Provider’s address and Service’s interface.
  • The Directory represent multiple Invoker,You can think of it as List<Invoker>,But unlike List,its value can be dynamically changing.such as registry push changes
  • The Cluster disguises multiple Invoker in Directory as a Invoker,The upper transparent, masquerade process contains fault-tolerant logic, call failed, try another
  • The Router is responsible for selecting subsets according to routing rules from multiple Invokers, such as read-write separation, application isolation, etc.
  • LoadBalance is responsible for selecting a specific one from multiple Invoker for this call. The selection process includes the load balancing algorithm. If the call fails, it needs to be re-selected

Cluster fault-tolerant mode

You can also customize the cluster fault tolerance strategy, see Cluster extension for more details.

Failover Cluster

Failure automatically switch, when there is failure, retry the other server (default). Usually used for read operations, but retries can result in longer delays. The times of retries can be set via retries =2 (excluding the first time).

The times of retries is configured as follows:

<dubbo:service retries="2" />

OR

<dubbo:reference retries="2" />

OR

<dubbo:reference>
    <dubbo:method name="findFoo" retries="2" />
</dubbo:reference>

Failfast Cluster

Fast failure, only made a call, failure immediately error. Usually used for non-idempotent write operations, such as adding records

Failsafe Cluster

Failure of security, anomalies, directly ignored. Usually used to write audit logs and other operations.

Failback Cluster

Failure automatically restored, failed to record the background request, regular retransmission. Usually used for message notification operations.

Forking Cluster

Multiple servers are invoked in parallel, returning as soon as one succeeds. Usually used for real-time demanding read operations, but need to waste more service resources. The maximum number of parallelism can be set with forks=2.

Broadcast Cluster

Calling all providers broadcast, one by one call, any error is reported (2.1.0+). It is usually used to notify all providers to update local resource information such as caches or logs.

Now in the broadcast call, the proportion of node call failures can be configured through broadcast.fail.percent. When this proportion is reached, BroadcastClusterInvoker will no longer call other nodes and directly throw an exception. The value of broadcast.fail.percent is in the range of 0-100. By default, an exception will be thrown when all calls fail. broadcast.fail.percent only controls whether to continue to call other nodes after failure, and does not change the result (any one will report an error). broadcast.fail.percent parameters Effective in dubbo2.7.10 and above.

Broadcast Cluster configuration broadcast.fail.percent.

broadcast.fail.percent=20 means that when 20% of the nodes fail to call, an exception will be thrown and no other nodes will be called.

@reference(cluster = "broadcast", parameters = {"broadcast.fail.percent", "20"})

Cluster mode configuration

Follow the example below to configure cluster mode on service providers and consumers

<dubbo:service cluster="failsafe" />

OR

<dubbo:reference cluster="failsafe" />

Last modified September 13, 2021: Update 3.0.2.1.md (#947) (6b3092d)