Skip to main content

Getting the Metadata Statistics of a HDFS file

 Below bash command will give you the metadata statistics such as distinct number of columns and   length of a record in a raw file. This will help you on landing phase, to determine if any raw files are corrupted.This will help you especially when there are n number of raw files are in your landing area and you need to do a quick testing of it. (The command is hard coded with .dat extension, please change according to your file type)

  1. Sample Out put of the below command is as follows:
    2. Output will be saved to a metadata file metadata_stat_<unixtimestamp>
    3. If we have more than one value for distinct_no_of_cols for a files, this means one or more records of your file is corrupted or there are some quoted strings in your records which contains delimiter itself as a value
    4. If your file is a fixed length files , distinct_no_lengths columns should have only one value per file. 

command:

metafilename=metadata_stat_$(date +%s%3N).txt;echo file_name\|distinct_no_cols\|distinct_no_lengths>$metafilename;for filename in `hadoop fs -ls /path/to/data/files/ | awk '{print $NF}' | grep .dat$ | tr '\n' ' '`; do echo $filename \|$(hadoop fs -cat $filename | awk -F "," '{print NF}'|sort|uniq|tr "\n" "," | sed 's/\(.*\),/\1 /') \|$(hadoop fs -cat $filename | awk '{print length($0)}'|sort|uniq|tr "\n" "," | sed 's/\(.*\),/\1 /') >> $metafilename;done

Note: If you have delimiters inside a quoted string (which is mentioned in 2nd point) , you can use below command 


metafilename=metadata_stat_$(date +%s%3N).txt;echo file_name\|distinct_no_cols\|distinct_no_lengths>$metafilename;for filename in `hadoop fs -ls /path/to/data/file/files | awk '{print $NF}' | grep .dat$ | tr '\n' ' '`; do echo $filename \|$(hadoop fs -cat $filename | awk -vFPAT='([^,]*)|("[^"]+")' -vOFS=, '{print NF}'|sort|uniq|tr "\n" "," | sed 's/\(.*\),/\1 /') \|$(hadoop fs -cat $filename | awk '{print length($0)}'|sort|uniq|tr "\n" "," | sed 's/\(.*\),/\1 /') >> $metafilename;done



Comments

  1. KathyMarch 31, 2021 at 1:57 AM

    Well written articles like yours renews my faith in today's writers. The article is very informative. Thanks for sharing such beautiful information.
    Best Data Migration tools
    Penetration testing companies USA
    What is Data Lake
    Artificial Intelligence in Banking
    What is Data analytics

    ReplyDelete

Post a Comment

Popular posts from this blog

Understanding spark architecture in Deep with YARN

OVERVIEW Apache spark is a Distributed Computing framework. By distributed it doesn’t imply that it can run only on a cluster. Spark can be configured on our local system also. But Since spark works great in clusters and in real time , it is being implemented in multi node clusters like Hadoop, we will consider a Hadoop cluster for explaining spark here. We can Execute spark on a spark cluster in following ways Interactive clients(scala shell,pyspark etc): Usually used for exploration while coding like python shell Submit a job (using spark submit utility):Always used for submitting a production application Basic Architecture Spark follows a Master/Slave Architecture. That is For every submitted application, it creates a Master Process and multiple slave processes. Master is the Driver and Slaves are the executors. Say If from a client machine, we have submitted a spark job to a cluster. Spark will create a driver process and multiple executors. Similraly...
38 comments
Read more

Different ways of Transposing a Dataframe in Pyspark

When I have started coding on transposing Dataframes, I found below different methods.  I am sharing all those info here. Creation of a test Input Dataframe to be Transposed ds = {'one':[0.3, 1.2, 1.3, 1.5, 1.4, 1.0],'two':[0.6, 1.2, 1.7, 1.5, 1.4, 2.0]} df = sc.parallelize([ (k,) + tuple(v[0:]) for k,v in ds.items()]).toDF() df.show() method 1 (This method involves conversion of spark object to a python object (rdd to list of tuples of entire data)): inp_list=df.rdd.map(tuple).collect() # Creating a list of tuples of rows # Unpacking the list and zipping all tuples together except the header #list(zip(*inp_list))[1:] is having the transposed tuples  and list(zip(*inp_list))[0] is having the header df_transpose=spark.createDataFrame(list(zip(*inp_list))[1:],list(zip(*inp_list))[0])  df_transpose.show() method 2 (In this method only header data we are converting into python list. Rest of the transformations are carried out as...
7 comments
Read more

Spark (With Python) : map() vs mapPartitions()

As a Beginner in spark, many developers will be having confusions over map() and mapPartitions() functions. This article is an attempt to resolve the confusions This blog is for : pyspark (spark with Python) Analysts and all those who are interested in learning pyspark. Pre-requesties: Should have a good knowledge in python as well as should have a basic knowledge of pyspark functions. If you use map() over an rdd , the function called  inside it will run for every record .It means if you have 10M records , function also will be executed 10M times. This is expensive especially when you are dealing with scenarios involving database connections and querying data from data base. Lets say inside map function, we have a function defined where we are connecting to a database and querying from it. Lets say our RDD is having 10M records. Now this function will execute 10M times which means 10M database connections will be created . This  is very expensive. In these kind ...
2 comments
Read more