document connection pool. about elasticsearch-js

[numtel]

Any word on this? Do I need to do anything to use a connection pool or is it automatic? How is this still open after 3 years?

[wbyoung]

@spalger I've just come across this as well and after poking through the source, it appears that the connection pool that's implemented is a pool of connections to nodes in the Elasticsearch cluster, one per host, is that correct?

More to the point: it's not a pool of open keep-alive connections for each host to avoid the cost of establishing the HTTP connection to a host to begin with as is typically meant by connection pooling in the context of database connections. Is that true?

If one did want pooling, for instance, using generic-pool, does this seem appropriate (and does pooling these clients make sense if there may be multiple concurrent code paths that work with Elasticsearch):

import elasticsearch from 'elasticsearch';
import { createPool } from 'generic-pool';

const pool = createPool({ create, destroy }, {
  max: 40,
  min: 4,
  idleTimeoutMillis: 1000,
});

async function create(): Promise<elasticsearch.Client> {
  const client = new elasticsearch.Client({
    host: 'localhost:9200',
    log: 'trace',
    apiVersion: '5.0',
  });

  // ensure connection has been created
  await client.ping({ requestTimeout: 1000 });

  return client;
}

function destroy(client: elasticsearch.Client): Promise<void> {
  client.close(); // no way to tell when connects are actually closed

  return Promise.resolve();
}

[spalger]

@wbyoung the connection pool implemented in the client does not actually pool sockets as one might expect (naming was chosen before I implemented the standard Elasticsearch client design). That said the HttpConnection object, the default connection class used in node.js, creates a unique agentkeepalive agent that has a pool of keepalive sockets for that es host/connection/node

[spalger]

Reconnection will happen automatically, after the deadTimeout. There are several other configuration options applicable to the connection pool detailed in the configuration section.

You're right though, I do need to write more about the overall design and what is and isn't handled by the module.

[chirag04]

@spenceralger It says coming soon under connection pool in the configuration section.

[pspeter3]

+1 This would be very helpful if it is necessary

[jaredhirsch]

+1, please document the client pool behavior. Relates to #165

[nijikokun]

For the love of god document this 👍

[mfrye-intuit]

Please document this ASAP. We just wasted so much time trying to build our own pool.

[delvedor]

Hi! I'm closing this due to inactivity, if it is still an issue, please feel free to reopen! :)

[sibelius]

@wbyoung can you provide a more complete example of using generic-pool ?

from here #16 (comment)

[wbyoung]

@sibelius not better than what's above… it's been 6 years and I'm no longer working on the project that used this. Your best bet would probably be the docs for geberic-pool.

[sibelius]

it will create 1 elastic search client per pool ?