Connectivity issues with Java client

(Shawn Johnson) #1

I have a simple two node cluster, and I'm trying to run a custom Java client (indexer-client-node0) on the same server as one of the 'data' (XX4) nodes. It happens that data-node2 (XX5) is master at the moment - which may to have some relevancy here?

I was able to connect the client previously when there was only one node, and the client and data nodes were on the same server. The problems began when the second data node was added.

Relevant log entries from client:

14:02:36,967 DEBUG [netty:895] [indexer-client-node0] connected to node [{#zen_unicast_1#}{10.10.XXX.XX4}{10.10.XXX.XX4:9300}]

14:02:38,259 DEBUG [zen:918] [indexer-client-node0] filtered ping responses: (filter_client[true], filter_data[false]) {none}

14:02:39,969 DEBUG [netty:1032] [indexer-client-node0] disconnecting from [{#zen_unicast_1#}{10.10.XXX.XX4}{10.10.XXX.XX4:9300}] due to explicit disconnect call

14:02:39,972 INFO [zen:443] [indexer-client-node0] failed to send join request to master [{elastic02}{e29o0UG4SmOWrtEcfs1u2Q}{10.10.XXX.XX5}{10.10.XXX.XX5:9300}], reason [RemoteTransportException[[elastic02][10.10.XXX.XX5:9300][internal:discovery/zen/join]]; nested: ConnectTransportException[[indexer-client-node0][] connect_timeout[30s]]; nested: NotSerializableExceptionWrapper[Connection refused: /]; ]]

Version: Elasticsearch 2.1.1
Both server elasticsearch.yml have: _eth0_

One settings I'm not sure about is - I have only one entry on each servers' elasticsearch.yml - and they are pointing to each other. Not quite sure if this is correct or not?

node1: elastic-data02

node2: elastic-data01

(Mark Walkom) #2


It is correct.

This looks relevant. It seems the client is using port 9301, which is good, but why is it connecting to localhost?
What's your connection string in the client look like?

(Shawn Johnson) #3

Thanks @warkolm.

client elasticsearch.yml 10.10.XXX.XX4:9300

However, the above is actually not as IP, instead it's set as the domain name - in case that matters?

(Mark Walkom) #4

That will depend on how the domain name is resolved, what does ping hostdomainname return?

(Shawn Johnson) #5

Yes, it does resolve. Actually the data nodes also refer to one another by domain name. I also (just) switched the client-node to IP referral - appears to be the same results.

(David Pilato) #6

You should prefer the Transport client. Node client won't be supported in the future.

Note that the node client does not load anymore any elasticsearch.yml file from the classpath.

(Shawn Johnson) #7

@dadoonet - Thanks for the reply. Is that mentioned anywhere? I thought what I had read to this point was that the a Node-type client was preferred?


The obvious first choice is to look at the client Elasticsearch provides natively.

(Isabel Drost-Fromm) #8

For a brief analysis on which native client to use see here:

or here:

To quote from the advise in the page above: "If you’re using Java, the Transport client should be chosen over the Node client unless the performance gain from using a Node client turns out to be large enough to warrant the additional network complexity. Use benchmarks to verify the performance gains."

(Shawn Johnson) #9

Thanks all for your replies, switching to Transport client seems to be working much smoother.

(system) #10