[jira] [Created] (KAFKA-5417) Clients get inconsistent connection states when SASL/SSL connection is marked CONECTED and DISCONNECTED at the same time

["dongeforever (JIRA)" <ji...@apache.org>]

dongeforever created KAFKA-5417:
-----------------------------------

             Summary: Clients get inconsistent connection states when SASL/SSL connection is marked CONECTED and DISCONNECTED at the same time
                 Key: KAFKA-5417
                 URL: https://issues.apache.org/jira/browse/KAFKA-5417
             Project: Kafka
          Issue Type: Bug
          Components: clients
    Affects Versions: 0.10.2.1
            Reporter: dongeforever
            Priority: Critical
             Fix For: 0.10.2.2


Assume the SASL or SSL Connection is established successfully, but be reset when writing data into it (This will happen frequently in LVS Proxy environment )
Selecter poll will act like follows:
try { 
   ...
    //finish connect successfully
    if (channel.finishConnect()) {
        this.connected.add(channel.id());    (1)
     }
     //the prepare will fail, for sasl or ssl will do handshake and write data
     //throw exception
    if (channel.isConnected() && !channel.ready())
                    channel.prepare();
   ....
} catch {
     close(channel);
      this.disconnected.add(channel.id());  (2)
}

The code line named (1) and (2) will mark the connection CONNECTED and DISCONNECTED at the same time.

And the NetworkClient poll will:
handleDisconnections(responses, updatedNow);   //remove the channel
handleConnections();   //mark the channel CONNECTED

So get the inconsistent ConnectionStates, and such state will block the messages sent into this channel in Sender:
For the channel will never be ready and never be connected again:

public boolean ready(Node node, long now) {
        if (node.isEmpty())
            throw new IllegalArgumentException("Cannot connect to empty node " + node);
        //return false, for the channel dose not exist actually
        if (isReady(node, now))
            return true;

        //return false, for the channel is marked CONNECTED
        if (connectionStates.canConnect(node.idString(), now))
            // if we are interested in sending to a node and we don't have a connection to it, initiate one
            initiateConnect(node, now);

        return false;
 }

So all messages sent to such channel will be expired eventually




--
This message was sent by Atlassian JIRA
(v6.3.15#6346)