Horizontally scaling Scrapyd

Question

What tool or set of tools would you use for horizontally scaling scrapyd adding new machines to a scrapyd cluster dynamically and having N instances per machine if required. Is not neccesary for all the instances to share a common job queue, but that would be awesome.

Scrapy-cluster seems promising for the job but I want a Scrapyd based solution so I listen to other alternatives and suggestions.

Answer 1

I scripted my own load balancer for Scrapyd using its API and a wrapper.

from random import shuffle
from scrapyd_api.wrapper import ScrapydAPI

class JobLoadBalancer(object):

    @classmethod
    def get_less_occupied(
            cls,
            servers_urls=settings.SERVERS_URLS,
            project=settings.DEFAULT_PROJECT,
            acceptable=settings.ACCEPTABLE_PENDING):

        free_runner = {'num_jobs': 9999, 'client': None}
        # shuffle servers optimization
        shuffle(servers_urls)
        for url in servers_urls:
            scrapyd = ScrapydAPI(target=url)
            jobs = scrapyd.list_jobs(project)
            num_jobs = len(jobs['pending'])

            if free_runner['num_jobs'] > num_jobs:
                free_runner['num_jobs'] = num_jobs
                free_runner['client'] = scrapyd
            # Optimization: if found acceptable pending operations in one server stop looking for another one
            if free_runner['client'] and free_runner['num_jobs'] <= acceptable:
                break

        return free_runner['client']

Unit test:

def setUp(self):
        super(TestFactory, self).setUp()
        # Make sure this servers are running
        settings.SERVERS_URLS = [
            'http://localhost:6800',
            'http://localhost:6900'
        ]
        self.project = 'dummy'
        self.spider = 'dummy_spider'
        self.acceptable = 0

    def test_get_less_occupied(self):
        # add new dummy jobs to first server so that choose the second one
        scrapyd = ScrapydAPI(target=settings.SERVERS_URLS[0])
        scrapyd.schedule(project=self.project, spider=self.spider)
        scrapyd.schedule(project=self.project, spider=self.spider)
        second_server_url = settings.SERVERS_URLS[1]
        scrapyd = JobLoadBalancer.get_less_occupied(
            servers_urls=settings.SERVERS_URLS,
            project=self.project,
            acceptable=self.acceptable)
        self.assertEqual(scrapyd.target, second_server_url)

This code targets an older version of scrapyd as it was written more than a year ago.