Doug Turnbull
Category: Solr
I frequently hit the wall of needing to work asynchronously with Solr requests in Python. Ill have some code that blocks on a Solr HTTP request, waits for it to complete, then execute a second request. Something like this code
import requests#Search 1solrResp = requests.get('http://mysolr.com/solr/statedecoded/search?q=law')for doc in solrResp.json()['response']['docs']: print doc['catch_line']#Search 2solrResp = requests.get('http://mysolr.com/solr/statedecoded/search?q=shoplifting')for doc in solrResp.json()['response']['docs']: print doc['catch_line'](were using the Requests library to do HTTP):
Being able to parallelize work is especially helpful with scripts that index documents into Solr. I need to scale my work up so that Solr, not network access, is the indexing bottleneck.
Unfortunately, Python isnt exactly Javascript or Go when it comes to doing asynchronous programming. But the gevent coroutine library can help us a bit with that. Under the hood gevent uses the libevent library. Built on top of native async calls (select, poll, etc – the original async ), libevent nicely leverages a lot of low-level async functionality.
Working with gevent is fairly straightforward. One slight sticking point is the gevent.monkey.patch_all() which patches a lot of the standard library to cooperate better with gevents asychrony. It sounds scary, but I have yet to have a problem with the monkey patched implementations.
Without further ado, heres how you use gevents to do parallel Solr requests:
import requestsfrom gevent import monkeyimport geventmonkey.patch_all()class Searcher(object): """ Simple wrapper for doing a search and collecting the results """ def __init__(self, searchUrl): self.searchUrl = searchUrl def search(self): solrResp = requests.get(self.searchUrl) self.docs = solrResp.json()['response']['docs']def searchMultiple(urls): """ Use gevent to execute the passed in urls; dump the results""" searchers = [Searcher(url) for url in urls] # Gather a handle for each task handles = [] for searcher in searchers: handles.append(gevent.spawn(searcher.search)) # Block until all work is done gevent.joinall(handles) # Dump the results for searcher in searchers: print "Search Results for %s" % searcher.searchUrl for doc in searcher.docs: print doc['catch_line']searchUrls = ['http://mysolr.com/solr/statedecoded/search?q=law', 'http://mysolr.com/solr/statedecoded/search?q=shoplifting']searchMultiple(searchUrls)Lots more code, and not nearly as pretty as the equivalent Javascript, but it gets the job done. The meat of the code is these lines:
# Gather a handle for each taskhandles = []for searcher in searchers: handles.append(gevent.spawn(searcher.search))# Block until all work is donegevent.joinall(handles)We tell gevent to spawn searcher.search. This gives us a handle to the spawned task. We can then optionally wait for all the spawned tasks to complete, then dump the results.
Thats about it! As always, comment if you have any thoughts on pointers. And let us know how we can help with any part of your Solr search application!