Handling MongoDB disconnect/reconnects from Node

Mike M. Lin picture Mike M. Lin · Jan 13, 2013 · Viewed 7.7k times · Source

When my MongoDB connection is idle for a few minutes, the next request ends in error. From the mongo command line client, it looks like this:

> db.users.find()
Sat Jan 12 23:42:35 Socket recv() errno:54 Connection reset by peer 107.22.25.25:47207
Sat Jan 12 23:42:35 SocketException: remote: 107.22.25.25:47207 error: 9001 socket exception [1] server [107.22.25.25:47207] 
Sat Jan 12 23:42:35 DBClientCursor::init call() failed
Sat Jan 12 23:42:35 query failed : chowology.users {} to: ds047207.mongolab.com:47207
Error: error doing query: failed
Sat Jan 12 23:42:35 trying reconnect to ds047207.mongolab.com:47207
Sat Jan 12 23:42:35 reconnect ds047207.mongolab.com:47207 ok

I see the problem against sandbox instances of MongoHQ and MongoLab.

The next request goes through fine, because of the reconnect. This is a problem in my web app, because after a few minutes of inactivity, this error will come up during a web request. There are two things that surprise me:

  1. That MongoDB connections are destroyed so regularly and frequently, and
  2. That the driver simply raises an exception as opposed to automatically retrying after reconnecting (I'm using connect-mongo which uses mongoose which, in turn, uses node-mongodb-native).

Is this everyone else's experience? How is this supposed to be handled? I'd be surprised if app developers wrap their database operations in some retry-exception-handling nonsense.

Answer

christkv picture christkv · Jan 16, 2013

You want to look at the docs for the Server object

http://mongodb.github.com/node-mongodb-native/api-generated/server.html#server

Especially the socketOptions where you can set keepAlive and the connection timeouts. By default keepalive is off and timeout is 0 or never which means the os default socket timeout is in effect (varies from os to os). Keep alive will send a packet once in awhile down the tcp socket connection to keep it alive. Sometimes firewalls are badly configured and don't send an end packet when they close a connection leaving the connection dead and in limbo which is what the monoglabs people are talking about (more often than not to be honest they are horribly configured).