老实说,我并不是很清楚为什么这3个要放在一起比较,inc和set有相似的地方,push和这2个感觉完全不一样啊。
实验中用的是ruby的mongo_ruby_driver来写程序的,为了获得比较精确的数值,我使用了benchmark这个module,反复执行set,inc,push各3000次。代码如下
#!/usr/bin/env ruby # 20140301, mongo_test.rb ### # test inc, set, push ### require "rubygems" require "mongo" require "benchmark" class MongoConnection def initialize(host, port) @mongoconn = Mongo::MongoClient.new(host, port) @db = @mongoconn.db("test") end def test() @coll = @db.collection("test") # set test @coll.insert({"num"=>0}) tm_start = Time.now.to_f # (1..3000).each { |i| @coll.update({ }, { "$set" => {"num"=>i} } ) } Benchmark.bm do |t| t.report{ (1..3000).each { |i| @coll.update({ }, { "$set" => {"num"=>i} } ) } } end # p @coll.find.to_a tm_used = Time.now.to_f - tm_start puts "set Time used(s): #{tm_used}" @coll.drop # inc test @coll.insert({"num"=>0}) tm_start = Time.now.to_f # (1..3000).each { |i| @coll.update({ }, { "$inc" => {"num"=>1} } ) } Benchmark.bm do |t| t.report{ (1..3000).each { |i| @coll.update({ }, { "$inc" => {"num"=>1} } ) } } end # p @coll.find.to_a tm_used = Time.now.to_f - tm_start puts "inc Time used(s): #{tm_used}" @coll.drop # push test @coll.insert({"num"=>[0]}) tm_start = Time.now.to_f # (1..3000).each { |i| @coll.update({ }, { "$push" => {"num"=>i} } ) } Benchmark.bm do |t| t.report{ (1..3000).each { |i| @coll.update({ }, { "$push" => {"num"=>i} } ) } } end # p @coll.find.to_a tm_used = Time.now.to_f - tm_start puts "push Time used(s): #{tm_used}" @coll.drop end end mongo_conn = MongoConnection.new("localhost", 27017) mongo_conn.test()
测试结果如下
user system total real 1.570000 0.130000 1.700000 ( 2.180069) set Time used(s): 2.1805129051208496 user system total real 1.480000 0.140000 1.620000 ( 2.059199) inc Time used(s): 2.0596039295196533 user system total real 1.850000 0.200000 2.050000 ( 6.312153) push Time used(s): 6.312557935714722
实际上,从反复执行的结果来看,set和inc基本是在同一个时间量级上,也不一定set就比inc快,也有相反的时候;而push的时间始终比较慢。这个跟rdbms的结果差异很大,rdbms的set和inc用的是update来实现,push用insert来实现,单个update语句产生的redo log要比insert多,所以一般update比insert要慢。但是,mongodb的结果与此相差很大,怀疑跟其存储引擎有关。