Well OK in a nutshell you want the result set for every user prepared and email to that user right.
This is a form of ETL where those result sets need to be posted somewhere. Say you create a table based on the result set prepared for each user. You may have many raw target tables at the end of the first ETL. How does this differ from using forEach? Performance wise forEach may not be optimal. Can you take the sample tables and try your method? HTH Mich Talebzadeh, Lead Solutions Architect/Engineering Lead Palantir Technologies Limited London United Kingdom view my Linkedin profile <https://www.linkedin.com/in/mich-talebzadeh-ph-d-5205b2/> https://en.everybodywiki.com/Mich_Talebzadeh *Disclaimer:* Use it at your own risk. Any and all responsibility for any loss, damage or destruction of data or any other property which may arise from relying on this email's technical content is explicitly disclaimed. The author will in no case be liable for any monetary damages arising from such loss, damage or destruction. On Wed, 26 Apr 2023 at 04:10, Marco Costantini < marco.costant...@rocketfncl.com> wrote: > Hi Mich, > First, thank you for that. Great effort put into helping. > > Second, I don't think this tackles the technical challenge here. I > understand the windowing as it serves those ranks you created, but I don't > see how the ranks contribute to the solution. > Third, the core of the challenge is about performing this kind of > 'statement' but for all users. In this example we target Mich, but that > reduces the complexity by a lot! In fact, a simple join and filter would > solve that one. > > Any thoughts on that? For me, the foreach is desirable because I can have > the workers chain other actions to each iteration (send email, send HTTP > request, etc). > > Thanks Mich, > Marco. > > On Tue, Apr 25, 2023 at 6:06 PM Mich Talebzadeh <mich.talebza...@gmail.com> > wrote: > >> Hi Marco, >> >> First thoughts. >> >> foreach() is an action operation that is to iterate/loop over each >> element in the dataset, meaning cursor based. That is different from >> operating over the dataset as a set which is far more efficient. >> >> So in your case as I understand it correctly, you want to get order for >> each user (say Mich), convert the result set to json and send it to Mich >> via email >> >> Let us try this based on sample data >> >> Put your csv files into HDFS directory >> >> hdfs dfs -put users.csv /data/stg/test >> hdfs dfs -put orders.csv /data/stg/test >> >> Then create dataframes from csv files, create temp views and do a join on >> result sets with some slicing and dicing on orders table >> >> #! /usr/bin/env python3 >> from __future__ import print_function >> import sys >> import findspark >> findspark.init() >> from pyspark.sql import SparkSession >> from pyspark import SparkContext >> from pyspark.sql import SQLContext, HiveContext >> from pyspark.sql.window import Window >> >> def spark_session(appName): >> return SparkSession.builder \ >> .appName(appName) \ >> .enableHiveSupport() \ >> .getOrCreate() >> >> def main(): >> appName = "ORDERS" >> spark =spark_session(appName) >> # get the sample >> users_file="hdfs://rhes75:9000/data/stg/test/users.csv" >> orders_file="hdfs://rhes75:9000/data/stg/test/orders.csv" >> users_df = >> spark.read.format("com.databricks.spark.csv").option("inferSchema", >> "true").option("header", "true").load(users_file) >> users_df.printSchema() >> """ >> root >> |-- id: integer (nullable = true) >> |-- name: string (nullable = true) >> """ >> >> print(f"""\n Reading from {users_file}\n""") >> users_df.show(5,False) >> orders_df = >> spark.read.format("com.databricks.spark.csv").option("inferSchema", >> "true").option("header", "true").load(orders_file) >> orders_df.printSchema() >> """ >> root >> |-- id: integer (nullable = true) >> |-- description: string (nullable = true) >> |-- amount: double (nullable = true) >> |-- user_id: integer (nullable = true) >> """ >> print(f"""\n Reading from {orders_file}\n""") >> orders_df.show(50,False) >> users_df.createOrReplaceTempView("users") >> orders_df.createOrReplaceTempView("orders") >> # Create a list of orders for each user >> print(f"""\n Doing a join on two temp views\n""") >> >> sqltext = """ >> SELECT u.name, t.order_id, t.description, t.amount, t.maxorders >> FROM >> ( >> SELECT >> user_id AS user_id >> , id as order_id >> , description as description >> , amount AS amount >> , DENSE_RANK() OVER (PARTITION by user_id ORDER BY amount) AS >> RANK >> , MAX(amount) OVER (PARTITION by user_id ORDER BY id) AS >> maxorders >> FROM orders >> ) t >> INNER JOIN users u ON t.user_id = u.id >> AND u.name = 'Mich' >> ORDER BY t.order_id >> """ >> spark.sql(sqltext).show(50) >> if __name__ == '__main__': >> main() >> >> Final outcome displaying orders for user Mich >> >> Doing a join on two temp views >> >> Doing a join on two temp views >> >> +----+--------+-----------------+------+---------+ >> |name|order_id| description|amount|maxorders| >> +----+--------+-----------------+------+---------+ >> |Mich| 50001| Mich's 1st order|101.11| 101.11| >> |Mich| 50002| Mich's 2nd order|102.11| 102.11| >> |Mich| 50003| Mich's 3rd order|103.11| 103.11| >> |Mich| 50004| Mich's 4th order|104.11| 104.11| >> |Mich| 50005| Mich's 5th order|105.11| 105.11| >> |Mich| 50006| Mich's 6th order|106.11| 106.11| >> |Mich| 50007| Mich's 7th order|107.11| 107.11| >> |Mich| 50008| Mich's 8th order|108.11| 108.11| >> |Mich| 50009| Mich's 9th order|109.11| 109.11| >> |Mich| 50010|Mich's 10th order|210.11| 210.11| >> +----+--------+-----------------+------+---------+ >> >> You can start on this. Happy coding >> >> Mich Talebzadeh, >> Lead Solutions Architect/Engineering Lead >> Palantir Technologies Limited >> London >> United Kingdom >> >> >> view my Linkedin profile >> <https://www.linkedin.com/in/mich-talebzadeh-ph-d-5205b2/> >> >> >> https://en.everybodywiki.com/Mich_Talebzadeh >> >> >> >> *Disclaimer:* Use it at your own risk. Any and all responsibility for >> any loss, damage or destruction of data or any other property which may >> arise from relying on this email's technical content is explicitly >> disclaimed. The author will in no case be liable for any monetary damages >> arising from such loss, damage or destruction. >> >> >> >> >> On Tue, 25 Apr 2023 at 18:50, Marco Costantini < >> marco.costant...@rocketfncl.com> wrote: >> >>> Thanks Mich, >>> >>> Great idea. I have done it. Those files are attached. I'm interested to >>> know your thoughts. Let's imagine this same structure, but with huge >>> amounts of data as well. >>> >>> Please and thank you, >>> Marco. >>> >>> On Tue, Apr 25, 2023 at 12:12 PM Mich Talebzadeh < >>> mich.talebza...@gmail.com> wrote: >>> >>>> Hi Marco, >>>> >>>> Let us start simple, >>>> >>>> Provide a csv file of 5 rows for the users table. Each row has a unique >>>> user_id and one or two other columns like fictitious email etc. >>>> >>>> Also for each user_id, provide 10 rows of orders table, meaning that >>>> orders table has 5 x 10 rows for each user_id. >>>> >>>> both as comma separated csv file >>>> >>>> HTH >>>> >>>> Mich Talebzadeh, >>>> Lead Solutions Architect/Engineering Lead >>>> Palantir Technologies Limited >>>> London >>>> United Kingdom >>>> >>>> >>>> view my Linkedin profile >>>> <https://www.linkedin.com/in/mich-talebzadeh-ph-d-5205b2/> >>>> >>>> >>>> https://en.everybodywiki.com/Mich_Talebzadeh >>>> >>>> >>>> >>>> *Disclaimer:* Use it at your own risk. Any and all responsibility for >>>> any loss, damage or destruction of data or any other property which may >>>> arise from relying on this email's technical content is explicitly >>>> disclaimed. The author will in no case be liable for any monetary damages >>>> arising from such loss, damage or destruction. >>>> >>>> >>>> >>>> >>>> On Tue, 25 Apr 2023 at 14:07, Marco Costantini < >>>> marco.costant...@rocketfncl.com> wrote: >>>> >>>>> Thanks Mich, >>>>> I have not but I will certainly read up on this today. >>>>> >>>>> To your point that all of the essential data is in the 'orders' table; >>>>> I agree! That distills the problem nicely. Yet, I still have some >>>>> questions >>>>> on which someone may be able to shed some light. >>>>> >>>>> 1) If my 'orders' table is very large, and will need to be aggregated >>>>> by 'user_id', how will Spark intelligently optimize on that constraint >>>>> (only read data for relevent 'user_id's). Is that something I have to >>>>> instruct Spark to do? >>>>> >>>>> 2) Without #1, even with windowing, am I asking each partition to >>>>> search too much? >>>>> >>>>> Please, if you have any links to documentation I can read on *how* >>>>> Spark works under the hood for these operations, I would appreciate it if >>>>> you give them. Spark has become a pillar on my team and knowing it in more >>>>> detail is warranted. >>>>> >>>>> Slightly pivoting the subject here; I have tried something. It was a >>>>> suggestion by an AI chat bot and it seemed reasonable. In my main Spark >>>>> script I now have the line: >>>>> >>>>> ``` >>>>> grouped_orders_df = >>>>> orders_df.groupBy('user_id').agg(collect_list(to_json(struct('user_id', >>>>> 'timestamp', 'total', 'description'))).alias('orders')) >>>>> ``` >>>>> (json is ultimately needed) >>>>> >>>>> This actually achieves my goal by putting all of the 'orders' in a >>>>> single Array column. Now my worry is, will this column become too large if >>>>> there are a great many orders. Is there a limit? I have search for >>>>> documentation on such a limit but could not find any. >>>>> >>>>> I truly appreciate your help Mich and team, >>>>> Marco. >>>>> >>>>> >>>>> On Tue, Apr 25, 2023 at 5:40 AM Mich Talebzadeh < >>>>> mich.talebza...@gmail.com> wrote: >>>>> >>>>>> Have you thought of using windowing function >>>>>> <https://sparkbyexamples.com/spark/spark-sql-window-functions/>s to >>>>>> achieve this? >>>>>> >>>>>> Effectively all your information is in the orders table. >>>>>> >>>>>> HTH >>>>>> >>>>>> Mich Talebzadeh, >>>>>> Lead Solutions Architect/Engineering Lead >>>>>> Palantir Technologies Limited >>>>>> London >>>>>> United Kingdom >>>>>> >>>>>> >>>>>> view my Linkedin profile >>>>>> <https://www.linkedin.com/in/mich-talebzadeh-ph-d-5205b2/> >>>>>> >>>>>> >>>>>> https://en.everybodywiki.com/Mich_Talebzadeh >>>>>> >>>>>> >>>>>> >>>>>> *Disclaimer:* Use it at your own risk. Any and all responsibility >>>>>> for any loss, damage or destruction of data or any other property which >>>>>> may >>>>>> arise from relying on this email's technical content is explicitly >>>>>> disclaimed. The author will in no case be liable for any monetary damages >>>>>> arising from such loss, damage or destruction. >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> On Tue, 25 Apr 2023 at 00:15, Marco Costantini < >>>>>> marco.costant...@rocketfncl.com> wrote: >>>>>> >>>>>>> I have two tables: {users, orders}. In this example, let's say that >>>>>>> for each 1 User in the users table, there are 100000 Orders in the >>>>>>> orders >>>>>>> table. >>>>>>> >>>>>>> I have to use pyspark to generate a statement of Orders for each >>>>>>> User. So, a single user will need his/her own list of Orders. >>>>>>> Additionally, >>>>>>> I need to send this statement to the real-world user via email (for >>>>>>> example). >>>>>>> >>>>>>> My first intuition was to apply a DataFrame.foreach() on the users >>>>>>> DataFrame. This way, I can rely on the spark workers to handle the email >>>>>>> sending individually. However, I now do not know the best way to get >>>>>>> each >>>>>>> User's Orders. >>>>>>> >>>>>>> I will soon try the following (pseudo-code): >>>>>>> >>>>>>> ``` >>>>>>> users_df = <my entire users DataFrame> >>>>>>> orders_df = <my entire orders DataFrame> >>>>>>> >>>>>>> #this is poorly named for max understandability in this context >>>>>>> def foreach_function(row): >>>>>>> user_id = row.user_id >>>>>>> user_orders_df = orders_df.select(f'user_id = {user_id}') >>>>>>> >>>>>>> #here, I'd get any User info from 'row' >>>>>>> #then, I'd convert all 'user_orders' to JSON >>>>>>> #then, I'd prepare the email and send it >>>>>>> >>>>>>> users_df.foreach(foreach_function) >>>>>>> ``` >>>>>>> >>>>>>> It is my understanding that if I do my user-specific work in the >>>>>>> foreach function, I will capitalize on Spark's scalability when doing >>>>>>> that >>>>>>> work. However, I am worried of two things: >>>>>>> >>>>>>> If I take all Orders up front... >>>>>>> >>>>>>> Will that work? >>>>>>> Will I be taking too much? Will I be taking Orders on partitions who >>>>>>> won't handle them (different User). >>>>>>> >>>>>>> If I create the orders_df (filtered) within the foreach function... >>>>>>> >>>>>>> Will it work? >>>>>>> Will that be too much IO to DB? >>>>>>> >>>>>>> The question ultimately is: How can I achieve this goal efficiently? >>>>>>> >>>>>>> I have not yet tried anything here. I am doing so as we speak, but >>>>>>> am suffering from choice-paralysis. >>>>>>> >>>>>>> Please and thank you. >>>>>>> >>>>>>
