[
https://issues.apache.org/jira/browse/KAFKA-13778?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
zhaobo updated KAFKA-13778:
---------------------------
Description:
The design purpose of the code is that only the leader broker can determine the
preferred read-replica.
{code:java}
// If we are the leader, determine the preferred read-replica val
preferredReadReplica = clientMetadata.flatMap( metadata =>
findPreferredReadReplica(partition, metadata, replicaId, fetchInfo.fetchOffset,
fetchTimeMs))
{code}
But in fact, since the broker does not judge whether it is the leader or not,
the follower will also execute the preferred read-replica selection.
{code:java}
partition.leaderReplicaIdOpt.flatMap { leaderReplicaId =>
// Don't look up preferred for follower fetches via normal replication and
if (Request.isValidBrokerId(replicaId))
None
else { {code}
was:
The design purpose of the code is that only the leader broker can determine the
preferred read-replica.
{code:java}
//代码占位符
{code}
But in fact, since the broker does not judge whether it is the leader or not,
the follower will also execute the preferred read-replica selection.
{code:java}
//代码占位符
partition.leaderReplicaIdOpt.flatMap { leaderReplicaId =>
// Don't look up preferred for follower fetches via normal replication and
// don't look up preferred read replica while fetch from follower replica
if (Request.isValidBrokerId(replicaId))
None
else { {code}
> Fix follower broker also always execute preferred read-replica selection
> ------------------------------------------------------------------------
>
> Key: KAFKA-13778
> URL: https://issues.apache.org/jira/browse/KAFKA-13778
> Project: Kafka
> Issue Type: Bug
> Components: core
> Affects Versions: 3.2.0
> Reporter: zhaobo
> Assignee: zhaobo
> Priority: Minor
>
> The design purpose of the code is that only the leader broker can determine
> the preferred read-replica.
>
> {code:java}
> // If we are the leader, determine the preferred read-replica val
> preferredReadReplica = clientMetadata.flatMap( metadata =>
> findPreferredReadReplica(partition, metadata, replicaId,
> fetchInfo.fetchOffset, fetchTimeMs))
> {code}
>
> But in fact, since the broker does not judge whether it is the leader or not,
> the follower will also execute the preferred read-replica selection.
> {code:java}
> partition.leaderReplicaIdOpt.flatMap { leaderReplicaId =>
> // Don't look up preferred for follower fetches via normal replication and
> if (Request.isValidBrokerId(replicaId))
> None
> else { {code}
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
