Node.js集群-检测工作人员卡住了吗？

Question

I'm using node.js with cluster, typically with 2 cpu's which translate to one master and two workers. I am having a sneaky problem, where occasionaly (very rarely), one of the workers gets 'stuck' for some reason, and the other bares all of the load. I am not sure of the cause and still investigating (no memory leak, no stack overflow, no exception).

When looking at the processes using top bash command on linux, I can clearly see that one of the node processes is steady at 100% cpu load.

What I want to ask of you guys today, is whether you know of a way to detect this situation (when one worker is at 100%) so I can kill it off.

Answer 1

Check out usage package. Something like this should work. I skipped cluster and worker setup.

var usage = require('usage');    
setInterval(function() {
    usage.lookup(worker.process.pid, function(err, result) {
        console.log(result);
        if(result.cpu > 90){
            worker.kill();
        }
    });
}, 5000)

Answer 2

OK, So here goes. Turns out my worker gets absolutely stuck. Don't know why, but it may be a cluster problem (what you call a cluster %^&$) Anyway, I had to monitor workers by the master. What I did is use cron to report from each worker every minute to the master, like so:

process.send({id:cluster.worker.id})

The master would receive that message and know that this worker is alive and well. The master then keeps a count of missing worker responses. After 5 minutes, the worker is killed if the count reaches 0 (decremented once every minute)

This is how I achieved (my own) goal of killing a stuck worker after a few minutes. This is not a complete solution, and I still don't know what causes the workers to get stuck without any exception. But that is life right now.