TCP Extend Errors

Question

Hi All,
I am getting the below errors in my TCP-Extend Proxy nodes.
Oracle Coherence GE 3.5.3/465p3 <Error> (thread=Proxy:ExendTcpProxyService:TcpAcceptorWorker:7, member=19): Extend*TCP has determined that the connection to "Socket[addr=/11.11.56.123,port=18733,localport=17062]" must be closed to maintain system stability: This connection is 1 messages behind (112820921 bytes); the limit is 60000 messages (100000000 bytes).
What does this mean ? Any serious issues in the cluster ?

Answer 1

Hi user594809 ,
Coherence will close an Extend connection if it thinks that the client is not reading messages off of its queue fast enough as this can destabilise the extend proxy. The limit is 60000 messages behind or 100Mbytes behind.
In your case you have tried to return too much data in a single go (112,820,921 bytes) - I have seen this is often the result of an invocation service call or something similar. What tends to happen in this case is that your client has started to read the message back and while doing this Coherence closes the connection so you get stream exceptions in the client.
If you really need to return such a large amount of data there are a couple of settings you can add to your <tcp-acceptor> configuration on the client
For example:
<tcp-acceptor>
    <local-address>
        <address>localhost</address>
        <port>10000</port>
    </local-address>
    <limit-buffer-size>100000000</limit-buffer-size>
    <suspect-buffer-size>10000000</suspect-buffer-size>
</tcp-acceptor>The limit-buffer-size is the maximum size of the queue in bytes before the connection gets closed - The default is 100Mbytes
The suspect-buffer-size is the size of the queue that Coherence starts to log warnings - The default is 10Mbytes
If you increase the limit-buffer-size you should increase the suspect-buffer-size too.
Note there is a reason that these settings are there in the first place and changing them to allow very large results to be returned could have an adverse affect on your extend proxy node and your cluster so it is up to you to test things properly under load.
JK