Does bindListeners() has to be recalled when a worker dies?

[MickL]

If yes, is it no problem to just call it again?

workers[i].on('exit', (code, signal) => {
    winstonCluster.bindListeners();
});

[ryankurte]

Yes the handler does have to be re-registered, no I think you probably shouldn't call it again.

Check out the relevant line here which will afaik cause multiple handlers to be registered to the same event on each existing thread.

If you wanted to remedy this, you could split it into a bindListener function which is called by the master on creation of all threads with bindListeners and also on the creation of a new thread following a dead worker. You'll also have to overwrite the logger in the new worker thread.

[MickL]

Could you provide an example?

My code is very default:

  // Get number of cpu-cores
  const num_processes = require('os').cpus().length;

  // Set worker file
  cluster.setupMaster({
    exec: './server/worker.ts'
  });

  // This stores our workers. We need to keep them to be able to reference
  // them based on source IP address. It's also useful for auto-restart, for example
  let workers = [];

  // Helper function for spawning worker at index 'i'
  let spawn = i => {
    logger.info(`Spawning worker ${i}`);
    workers[i] = cluster.fork();

    // Restart worker on exit
    workers[i].on('exit', (code, signal) => {
      if (signal) {
        logger.warn(`Worker was killed by signal: ${signal}`);
      } else if (code !== 0) {
        logger.warn(`Worker exited with error code: ${code}`);
      }
      spawn(i);
    });
  };

  // Spawn workers.
  for (let i = 0; i < num_processes; i++) {
    spawn(i);
  }

  // Bind winston to workers
  winstonCluster.bindListeners();

[ralphschuler]

In PR #13 there was a singe bindListener(worker, instance=null) function added to overcome that problem.