I have around 300GB fulldata and daily I’ll get an update on this data around 10GB. Both files are in text format. I would like to update the fulldata based on updates. How do I proceed the situation with Apache spark in a distributed manner.
I have tried to create a JavaRDD with a map function which override a call method and converted that to an Dataset[Row] from the two files. Now I’m planning to do a sparkSQL join queries over the datasets. Is this the right approach, anyone can guide me in this, as this’s my first footstep with apache spark.
How do I achieve the parallel processing here?
Related
i need to handle a big CSV file with around +750.000 rows of data. Each line has around 1000+ characters and ~50 columns, and i am really not sure what's the best (or atleast good and sufficient) way to handle and manipulate this kind of data.
I need to do the following steps:
Compare the values of two Colomns and write the result to a new column (this one seems easy)
Compare values of two lines and do stuff. (e.g delete if one value is duplicated.)
Compare values of two different files.
My Problem is that this is currently done with PHP and/ or Excel and the limits are nearly exceeded + this takes a long time to process and will be no longer possible when the files get even bigger.
I have 3 different possibilities in mind:
Use MySQL, create a table (or two) and do the comparing, adding or deleting part. (I am not really familiar with SQL and would have to learn it, also it should be done automatically so there is the problem that you cant create tables of CSV files )
Use Java creating Objects in ArrayList or Linked Lists and to "the stuff" (to operations would be easy but handling that much data will probably be the problem)
(Is it even possible to save that many files in Java or does it crash / is there a good tool etc.?)
Use Clojure along with MongoDB to add files from CSV to MongoDB and read files using Mongo.
(Name additional possibilities if you have another idea ..)
All in all I am not a Pro in any of these but would like to solve this problem / get some hints or even your opinion.
Thanks in advance
Since in our company we work a lot with huge csv files here are some ideas:
because these files are in our case always exported from some other relational database we always use PostgreSQL, MySQL or golang + SQLite to be able to use simple plain SQL queries which are in these cases most simple and reliable solution
number of rows you describe is quite low from the point of view of all these databases so do not worry
all have native internal solution for import / export of CSV - which works much quicker than anything created manually
for repeated standard checks I use golang + SQLite with :memory: database - this is definitely the quickest solution
MySQL is definitely very good and quick for checks you described but choose of database depends also on how sophisticated analysis you would need to do further - for example MySQL up to 5.7 still does not have window functions which you could need later - so consider using PostgreSQL in some cases too...
I normally use PostgreSQL for this kind of tasks. PostgreSQL COPY allows importing CSV data easily. Then you get a table with your CSV data and the power for SQL (and a reasonable database) to do basically anything you want with the data.
I am pretty sure MySQL have similar capabilities of importing CSV, I just generally prefer PostgreSQL.
I would not use Java for CSV processing. This will be too much code and unless you take care of indices, the processing will not be performant. An SQL database is much better equiped for tabular data processing (should not be a surprize).
I wouldn't use MongoDB, my impression is that it is less powerful in update operations compared to an SQL database. But this is just an opinion, take it with a grain of salt.
You should try Python with the pandas package. On a machine with enough memory (say 16GB) it should be able to handle your CSV files with ease. The main thing is - anyone with some experience with pandas will be able to develop a quick script for you and tell you in a few minutes if your job is doable or not. To get you started:
import pandas
df = pandas.read_csv('filename.csv')
You might need to specify the column type if you get into memory issues.
https://pandas.pydata.org/pandas-docs/stable/generated/pandas.read_csv.html
I'd suggest to use Spark. Even in a standalone machine the performance is incredible. You can use Scala and Python to handle your data. It's flexible and you can do processing that is impossible in Java or relational database.
The other choices are great also, but I'd consider Spark to all analytics needs from now on.
In our application , in one of our microservice we will query the DB , get the result ( 100k rows ) and generate Excel using Apache POI.In couple of other services they also does the same process ( get DB rows and generate excel) . Here Excel generation process is common , IS this right design to separate this excel generation process as separate micorservice and use in all other services ?
The challenge is passing the data ( 100k rows ) between microservices over HTTP .
How can we achieve it ?
I personally never put the export feature as a separate service.
Providing such a table based data, I provide a table view of the data with paging, and also give export function as an octet streamed data without paging limit. Export could be a type of a view.
I've used the Apache POI library for report rendering but only for the small pages and complex shapes previously. POI also provides streaming version of workbook classes such as SXSSFWorkbook.
To be a microservice, it should have a proper reason to be a external system. If the system only provides just export something, negative. It's too simple and overkill. If you're considering to add versioning, permission, distribution, folder zipping, or... storage management, well.. that could be an option.
By the way, exporting such a big data into a file, Excel has max row limit to 1M size so you may hit the limit if your data size grow more.
Why don't use use just a CSV format? Easy to use, Easy to jump, Easy to process.
You need to ask this question as to what define a service. Reading a chunks of data from a while, does this come under a service?
When I think of separating my services I think along multiple lines like what this module needs to do. Who all will be using it, what all dependencies do I have, how I need to scale it up in future and above all. Which business team will be taking care of it. I tend to divide the modules based on the answers I get to these questions.
Here in your case I see this as less of a service and more of a utility function that can be put in a jar and shared across. A new service will be more along a line of say reporting service reading legacy excel files to create reports or migrating service which uses a utility to read excel.
Also there is no final answer you need to keep questioning your design unless you are happy with it.
I am working on one project where I am receiving somewhere around 10 files each containing size of 200GB. My Project requirement is to extract data from each file and do the joining with other files and extract the data.
E.G like I have file 1.txt where I have Account Id and I have file 2.txt, where I have Account Id and Account Name. On the Basis of the Account Id from the first file I need to extract the data from the second file.
This way I need to perform manipulation on each of the 10 files and create final output files.
I am currently doing this in Java which is really time taking process. Approx it is taking 4 to 5 hours.
Can I increase my performance by any change? Is there any technology or any tool or framework which I can integrate with java and increase my performance?
I have used following approaches.
1) Apache Drill- I am performing a join queries
Result : Drill is throwing Drillbit down exception as file size is too much.
2) Apache Beam- I am performing joining on files using parallel processing.
Result : it is giving me Out Of Memory exception at group by feature.
I am reading this data from hadoop.
I would suggest to use Hadoop and Spark because Spark uses in-memory model which is faster than Map Reduce.
Maybe these two links will help you:
https://content.pivotal.io/blog/3-key-capabilities-necessary-for-text-analytics-natural-language-processing-in-the-era-of-big-data
https://community.hortonworks.com/articles/84781/spark-text-analytics-uncovering-data-driven-topics.html
I am new to Hadoop and Mapreduce. I have a requirement to compare two excel files using mapreduce. I have to go to mapreduce as the size of these files will be very big(>10gb). My question is how do I get two different input files from different mapper and compare these two files in Reducer.I have to convert this application into jar and run it in Amazon EMR.I am not able to find suitable tutorial for this in web. Kindly provide me some ideas to do this.
I think that distributed cache would be useful in your situation. I haven't used distributed cache with large files but please explore and let me know if it works for you.
Based on your answers to my comment here is how I think it should be done if you are implementing it using plain Map-reduce -
Create a custom InputFormat that reads the excel spreadsheet.
As part of this you would need RecordReader that reads Excel spreadsheet and outputs Cell location ( A1 for example) as key and its content as Value.
Once the files are read you need to do inner join on two datasets using Key ( which is cell location).
Post join, you can compare the contents of the cell.
It may be helpful if you take a look at Apache Pig or Cascading kind of APIs that abstract plain map-reduce.
Regards,
Amit
Bulk load usually uses map reduce to create a file on HDFS and this file is then assoicated with a region.
If thats the case, can my client create this file (locally) and put it on hdfs. See as we already know what keys are , what values, we can do it locally without loading the server.
Can someone point to an example, how hfile can be created (in any language will be fine)
regards
Nothing actually stops anyone from preparing HFile 'by hands' but doing so you start to depend on HFile compatibility issues. In accordance to this (https://hbase.apache.org/book/arch.bulk.load.html) you just need to put your files to HDFS ('closer' to HBase) and call completebulkload.
Proposed strategy:
- Check HFileOutputFormat2.java file from HBase sources. It is standard MapReduce OutputFormat. What you indeed need as base for this is just sequence of KeyValue elements (or Cell if we speak in term or interfaces).
- You need to free HFileOutputFormat2 from MapReduce. Check for its writer logics for this. You need only this part.
- OK, you need also to build effective solution for Put -> KeyValue stream handling for HFile. First place to look is TotalOrderPartitioner and PutSortReducer.
If you did all steps you have solution that can take sequence of Put (no issue to generate them from any data) and as a result you have local HFile. Looks like this should take up to week to get something pretty working.
I don't go this way because just having good InputFormat and data transforming mapper (which I have long ago) I now can use standard TotalOrderPartitioner and HFileOutputFormat2 INSIDE MapReduce framework to have everything working just using full cluster power. Feel confused by 10G SQL dump loaded in 5 minutes? Not me. You can't beat such speed using single server.
OK, this solution required careful SQL requests design for SQL DB to perform ETL process from. But now it's everyday procedure.