Package io.confluent.kafka.replication.push

Class PushSession

java.lang.Object
io.confluent.kafka.replication.push.PushSession
public abstract class PushSession extends Object
Identifies a push replication session for some partition replica. In addition to the session identifiers (leader epoch, replica epoch, replication session ID), it also contains callbacks for session-related updates - one called whenever an AppendRecords request for the session has been successfully responded to, and one called whenever the push session has ended.

  • Constructor Details

    • PushSession

      public PushSession(int leaderEpoch, long replicaEpoch, long replicationSessionId, org.apache.kafka.common.Node replicaNode)

  • Method Details

    • onAppendRecordsResponse

      public abstract boolean onAppendRecordsResponse(long logEndOffset, long logStartOffset)
      Called on a successful AppendRecords response for that partition replica. Includes the LEO and the optional LSO for the updated partition data on the follower side.
      Returns:
      whether the high watermark or low watermark increased

    • onPushSessionEnded

      public abstract void onPushSessionEnded()

      Called when the push session has ended.

      This can happen due to either:

      • a leader-side event that ends the session, regardless of whether it requires notifying the follower or not - even if a "session ended" signal will be sent, the internal processing on the leader will be eagerly triggered;
      • AppendRecords response processing that detects a follower-side non-retriable error that ends the session e.g. detecting a stale leader or replica epoch;

    • tryCompleteDelayedRequests

      public abstract void tryCompleteDelayedRequests()
      Try to complete any pending requests in purgatory. Must be invoked after the onAppendRecordsResponse(long, long) handler indicates that the high watermark or low watermark increased. This is decoupled from onAppendRecordsResponse to not have potentially expensive purgatory completions block the pusher thread or replica state updates, while also allowing the pusher thread to schedule purgatory completions on a separate worker pool.

    • leaderEpoch

      public int leaderEpoch()

    • replicaEpoch

      public long replicaEpoch()

    • replicationSessionId

      public long replicationSessionId()

    • replicaNode

      public org.apache.kafka.common.Node replicaNode()

    • matchesRequestData

      public boolean matchesRequestData(long requestReplicaEpoch, org.apache.kafka.common.message.AppendRecordsRequestData.PartitionData data)

    • toString

      public String toString()
      Overrides:
      toString in class Object