Reconnect if beanstalkd stops and restarts about fivebeans

[michaelfreund]

Having reconnect functionality after the beanstalkd server is gone away was also crucial for our infrastructure.

As we tried to implement a reconnect logic around fivebeans, we ran into multiple issues. One of them was that if we call worker.watch() inside our worker.on('close') callback function, fivebeans would also add the client listeners (on('connect'), on('error'), etc.) again, which of course would lead to an event emitter memory leak during the reconnection attempts.

Just calling worker.client.connect() during the reconnection attempts fixed the listener issue. However after the beanstalkd server was up again, all new jobs added to the queue had a mysterious hang followed by a DEADLINE_SOON after they were processed. At this point, the handler was already done, but it took a couple of seconds until the job was properly deleted.

So not sure if a reconnection behavior should be part of fivebeans, but one possible solution could be to write a simple wrapper which uses a separate connection to just check the availability of the beanstalkd server and reinitializes the worker if it was able to reconnect.

The following seems to work

const Beanworker = require('fivebeans').worker;
const Reconnect = require('net-socket-reconnect')

var beanHost = "127.0.0.1";
var beanPort = 11300;

var beanServer = Reconnect({
    host: beanHost,
    port: beanPort,
    reconnectInterval: 5000
});

var worker, options = {
    id: 'worker-test-01',
    host: beanHost,
    port: beanPort,
    handlers: {
        test: require('./handler/test')()
    },
    ignoreDefault: true
}

beanServer.on('connect', function () {
    worker = new Beanworker(options);
    worker.start(['test_tube']);
});

beanServer.on('close', function () {
    if (worker) {
        worker = null;
        delete worker;
    }
});

What do you think?

[ceejbot]

Or this is straightforward to roll by hand if necessary. Good bug.

[rjrodger]

I tried rolling it by hand, but I couldn’t get it working and then my brain hurt :)

Any pointers on how to approach this?

On 2 Jul 2014, at 18:18, C J Silverio [email protected] wrote:

Or this is straightforward to roll by hand if necessary. Good bug.

—
Reply to this email directly or view it on GitHub.

[JoshWillik]

+1 I have rolled this by hand, and it makes the code a lot more confusing.

@ceejbot Do you have plans to add this to fivebeans, or is this something you'd prefer to see implimented in a wrapper library? If it's the later, I can throw something together no problem.

[siddo420]

+1

need this behaviour as part of the library itself including reconnect and retry strategies

[gtamas]

+1 I need this too

[wltsmrz]

This thing has automatic reconnects and automatically re-uses / re-watches tubes on reconnect. It doesn't have anything advanced like an exponential backoff etc., but seems to work. Would be nice to get more mileage on it

[ceejbot]

Okay, this is pretty clearly something the module needs to do. Give me a little bit to reacquaint myself with what the heck I was thinking when I wrote it (my first node module ever, if you want to be scared) and make it reconnect. I like the server approach above.

[michaelfreund]

@ceejbot did you reacquaint with your first baby? 😁 meanwhile our solution runs in production without any problems.