为什么 Mongo Spark 连接器为查询返回不同且不正确的计数? [英] Why Mongo Spark connector returns different and incorrect counts for a query?

问题描述

我正在评估一个项目的 Mongo Spark 连接器，但结果不一致.我在笔记本电脑上本地使用 MongoDB 服务器版本 3.4.5、Spark(通过 PySpark)版本 2.2.0、Mongo Spark 连接器版本 2.11；2.2.0.对于我的测试数据库，我使用 Enron 数据集 http://mongodb-enron-email.s3-website-us-east-1.amazonaws.com/ 我对 Spark SQL 查询很感兴趣，当我开始运行简单的计数测试查询时，每次运行都会收到不同的计数.这是我的 mongo shell 的输出:

I'm evaluating Mongo Spark connector for a project and I'm getting the inconsistent results. I use MongoDB server version 3.4.5, Spark (via PySpark) version 2.2.0, Mongo Spark Connector version 2.11;2.2.0 locally on my laptop. For my test DB I use the Enron dataset http://mongodb-enron-email.s3-website-us-east-1.amazonaws.com/ I'm interested in Spark SQL queries and when I started to run simple test queries for count I received different counts for each run. Here is output from my mongo shell:

> db.messages.count({'headers.To': '[e-mail]'})
203

以下是我的 PySpark shell 的一些输出:

Here are some output from my PySpark shell:

In [1]: df = spark.read.format("com.mongodb.spark.sql.DefaultSource").option("uri", "mongodb://127.0.0.1/enron_mail.messages").load()
In [2]: df.registerTempTable("messages")
In [3]: res = spark.sql("select count(*) from messages where headers.To='[e-mail]'")
In [4]: res.show()
+--------+                                                                      
|count(1)|
+--------+
|     162|
+--------+
In [5]: res.show()
+--------+                                                                      
|count(1)|
+--------+
|     160|
+--------+
In [6]: res = spark.sql("select count(_id) from messages where headers.To='[e-mail]'")
In [7]: res.show()
+----------+                                                                    
|count(_id)|
+----------+
|       161|
+----------+
In [8]: res.show()
+----------+                                                                    
|count(_id)|
+----------+
|       162|
+----------+

我在 Google 中搜索了有关此问题的信息，但没有找到任何有用的信息.如果有人有任何想法为什么会发生这种情况以及如何正确处理这种情况，请分享您的想法.我有一种感觉，可能是我遗漏了某些东西，或者某些东西没有正确配置.

I searched in Google about this issue but I didn't find anything helpful. If someone has any ideas why this could happen and how to handle this correctly please share your ideas. I have a feeling that maybe I missed something or maybe something wasn't configured properly.

更新:我解决了我的问题.计数不一致的原因是 MongoDefaultPartitioner 包装了使用随机采样的 MongoSamplePartitioner.老实说，这对我来说是一个很奇怪的默认设置.我个人更喜欢使用缓慢但一致的分区程序.分区器选项的详细信息可以在官方配置选项文档中找到.

UPDATE: I solved my issue. The reason of inconsistent counts was the MongoDefaultPartitioner which wraps MongoSamplePartitioner which uses random sampling. To be honest this is quite a weird default as for me. I personally would prefer to have a slow but a consistent partitioner instead. The details for partitioner options can be found in the official configuration options documentation.

更新:将解决方案复制到答案中.

UPDATE: Copied the solution into an answer.

推荐答案

我解决了我的问题.计数不一致的原因是 MongoDefaultPartitioner 包装了使用随机采样的 MongoSamplePartitioner.老实说，这对我来说是一个很奇怪的默认设置.我个人更喜欢使用缓慢但一致的分区程序.分区器选项的详细信息可以在官方 配置选项 文档中找到.

I solved my issue. The reason of inconsistent counts was the MongoDefaultPartitioner which wraps MongoSamplePartitioner which uses random sampling. To be honest this is quite a weird default as for me. I personally would prefer to have a slow but a consistent partitioner instead. The details for partitioner options can be found in the official configuration options documentation.

代码:

val df = spark.read
  .format("com.mongodb.spark.sql.DefaultSource")
  .option("uri", "mongodb://127.0.0.1/enron_mail.messages")
  .option("partitioner", "spark.mongodb.input.partitionerOptions.MongoPaginateBySizePartitioner ")
  .load()

这篇关于为什么 Mongo Spark 连接器为查询返回不同且不正确的计数?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！