0. run command

```
./bin/spark-submit \
--master "spark://YOUR_HOST_NAME:7077" \
--deploy-mode client \
--executor-memory 3G \
--num-executors 2 \
--conf "spark.mongodb.input.uri=mongodb://127.0.0.1:27017/test" \
--conf "spark.mongodb.output.uri=mongodb://127.0.0.1:27017/test" \
--packages org.mongodb.spark:mongo-spark-connector_2.11:2.2.0 \
/YOUR/PATH/TO/THIS/joinErr.py
```

1. insert some data into test db

```
db.calc.drop()

db.calc.insert([

{devId: "001", prdId: "product1", size: 10}

,

{devId: "002", prdId: "product2", size: 20}

])

db.calc.find()

{ "_id" : ObjectId("59f284a8a22c47422c2059d2"), "devId" : "001", "prdId" : "product1", "size" : 10 } { "_id" : ObjectId("59f284a8a22c47422c2059d3"), "devId" : "002", "prdId" : "product2", "size" : 20 }

```

2. run the command in section 0, its output show as below

```
==> collect sumDF: [Row(devId=u'001', prdId=u'product1', size=10), Row(devId=u'002', prdId=u'product2', size=20)]
==> collect dpDF: [Row(devId=u'001', prdId=u'product1', size=1), Row(devId=u'002', prdId=u'product2', size=2)]
==> final sumDF.collect: [Row(devId=u'002', prdId=u'product2', size=22), Row(devId=u'001', prdId=u'product1', size=11)]
```
Yep!, the infal sumDF is what I wanted. but look in the mongo shell:

```
db.calc.find()

{ "_id" : ObjectId("59f2855896b438192f213ecf"), "devId" : "001", "prdId" : "product1", "size" : NumberLong(1) } { "_id" : ObjectId("59f2855896b438192f213ece"), "devId" : "002", "prdId" : "product2", "size" : NumberLong(2) }

```
it seems that sumDF is replaced by dpDF.

3. if sumDF is create by hand or read throud Pymongo, it works as intended. just as test-case2 and test-case3 indicated.