[ 
https://issues.apache.org/jira/browse/PHOENIX-8038?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Lokesh Khurana resolved PHOENIX-8038.
-------------------------------------
    Resolution: Fixed

> Remove CLUSTER_ROLE_BASED_MUTATION_BLOCK_ENABLED and always block mutations 
> in the ACTIVE_TO_STANDBY role
> ---------------------------------------------------------------------------------------------------------
>
>                 Key: PHOENIX-8038
>                 URL: https://issues.apache.org/jira/browse/PHOENIX-8038
>             Project: Phoenix
>          Issue Type: Sub-task
>            Reporter: Lokesh Khurana
>            Assignee: Lokesh Khurana
>            Priority: Major
>
> h3. Problem
> The property {{phoenix.cluster.role.based.mutation.block.enabled}} 
> ({{CLUSTER_ROLE_BASED_MUTATION_BLOCK_ENABLED}}) controls the mutation block 
> during a consistent failover. When the local role is {{ACTIVE_TO_STANDBY}}, 
> {{IndexRegionObserver#preBatchMutate}} rejects HA client writes with 
> {{MutationBlockedIOException}}.
> A consistent failover is not correct without this block. If the property is 
> false, the old active cluster continues to accept writes during the cutover. 
> Consistent failover deployments always set this property to true. Thus the 
> property gives no value. It only adds a configuration that can cause an 
> incorrect failover.
> h3. Current code (PHOENIX-7562-feature-new @ c9ea66c177)
> * {{QueryServices.java:456}}: the property name.
> * {{QueryServicesOptions.java:505}} and {{:650}}: the default value and the 
> {{setIfUnset}} entry.
> * {{HAGroupStoreManager.java:185}}: reads the property into 
> {{mutationBlockEnabled}}.
> * {{HAGroupStoreManager.java:210}}: {{isMutationBlocked()}} returns false 
> when the property is false.
> * {{IndexRegionObserver.java:982}}: the only caller. The exception message 
> names the property.
> * {{MutationBlockedIOException.java:23}}: the Javadoc names the property.
> h3. Proposed change
> # Remove {{CLUSTER_ROLE_BASED_MUTATION_BLOCK_ENABLED}} from {{QueryServices}}.
> # Remove {{DEFAULT_CLUSTER_ROLE_BASED_MUTATION_BLOCK_ENABLED}} and its 
> {{setIfUnset}} entry from {{QueryServicesOptions}}.
> # Remove the {{mutationBlockEnabled}} field from {{HAGroupStoreManager}}. 
> {{isMutationBlocked()}} always checks the cluster role.
> # Update the exception message in {{IndexRegionObserver}} and the Javadoc of 
> {{MutationBlockedIOException}}. Do not refer to the property.
> # Remove the property from these ITs: 
> {{SyncReplicationRolloutFeatureOnMissingGroupIT}}, 
> {{SyncReplicationRolloutStep1IT}}, {{BypassedMutationBlockMetricsIT}}, 
> {{IndexRegionObserverMutationBlockingIT}}, {{HAGroupMetricsIT}}, 
> {{HAGroupStoreManagerIT}}, {{ParallelPhoenixConnectionIT}}.
> # Remove {{HAGroupStoreManagerIT#testMutationBlockDisabled}}. The disabled 
> mode does not exist after this change.
> # Update the class comment in {{SyncReplicationRolloutStep1IT}} (line 54), 
> which names the property.
> h3. Behavior after the change
> The block continues to operate only when the 
> {{SYNCHRONOUS_REPLICATION_ENABLED}} master switch is true. It applies only to 
> batches that carry {{_HAGroupName}} ({{IndexRegionObserver.java:965}}). Thus 
> a cluster that does not use consistent failover has no change.
> h3. Out of scope
> * The {{master}} and {{5.3}} branches. These branches keep the property for 
> the PHOENIX-7493 graceful failover path.
> * {{HA_GROUP_STALE_FOR_MUTATION_CHECK_ENABLED}}.



--
This message was sent by Atlassian Jira
(v10.3.26#10030026)

Reply via email to