TransportClient client node openning too many tcp connections

Elastic Stack Elasticsearch
1

Hi All,

We are using java TransportClient api to connect to elasticsearch cluster. We see a lot of unclosed tcp connections lying open for longer time, is there any connection pool setting that can limit this count?

Thanks
sunder

2

The transport client is basically a mini elasticsearch node and uses the same pooling the nodes use to communicate with one another which works pretty well in my experience. Do you happen to be making a new TransportClient object very frequently?

3

Hi Nik,

Thanks for quick response.

No we are making a single TransportClient connection per jvm as an singleton, but we are running 2/3 jvm's per machine.

Thanks
sunder

4

Multiple JVMs per machine shouldn't cause any trouble.

How many connections is a lot? How are you measuring it?

5

Hi Nik,

We are checking count of established connection using the following command. Without any activity and just initializing the TransportClient at the startup and letting system rest for 8 hrs. Following is the count (4 jvm's)

netstat -an | grep 9300 | wc -l
224

With very less activity this count shoots[on one jvm] up to

netstat -anp | grep 9300 |wc -l
18732

Do you see any issue with settings? Following is the code snippet how we are initializing TransportClient

        Settings settings = Settings.settingsBuilder()
                .put("cluster.name", rb.getString("cluster.name")).build();
        TransportClient transportClient =TransportClient.builder().settings(settings).build();
        String transPortAddress = rb.getString("transport.address");
        StringTokenizer st = new StringTokenizer(transPortAddress, ",");
        while (st.hasMoreTokens()) {
            String transPortAddr = st.nextToken();
            logger.debug("transPortAddr ---{}", transPortAddr);
            transportClient = transportClient.addTransportAddress(new InetSocketTransportAddress(InetAddress.getByName(transPortAddr), 9300));
        }

The above code is loaded as singleton on JVM startup. We are using the BulkProcessor API to do add/delete documents and search api(on transport client) to do document search

Thanks
sunder