Hadoop DataLake Implementation Part 9

Load Stocks Fact Table

Using Sqoop we will load the stocks data, initial/base as well as incremental dataset from MySQL to HDFS.

sqoop job --meta-connect "jdbc:hsqldb:hsql://sandbox-hdp.hortonworks.com:16001/sqoop" \
--create jb_stg_stocks \
-- import \
--bindir ./ \
--driver com.mysql.jdbc.Driver \
--connect jdbc:mysql://sandbox-hdp.hortonworks.com:3306/sales \
--username root \
--password-file /user/edw_user/sales/.password \
--table stocks \
--fetch-size 1000 \
--as-textfile \
--fields-terminated-by '|' \
--target-dir /user/edw_user/sales/staging/stocks \
--incremental append \
--check-column id \
--split-by id \
--num-mappers 2

Finally execute the sqoop job to load the data from source to HDFS.

sqoop job --meta-connect "jdbc:hsqldb:hsql://sandbox-hdp.hortonworks.com:16001/sqoop" --list
sqoop job --meta-connect "jdbc:hsqldb:hsql://sandbox-hdp.hortonworks.com:16001/sqoop" --show jb_stg_stocks
sqoop job --meta-connect "jdbc:hsqldb:hsql://sandbox-hdp.hortonworks.com:16001/sqoop" --exec jb_stg_stocks

Now we will define a hive external table for the Sales staging data as well as final Hive managed ORC dimension table. Connect to Beeline CLI using edw_user as username and password as hadoop. We will connect to hive schema ‘sales_analytics’.

One time setup

beeline
!connect jdbc:hive2://sandbox-hdp.hortonworks.com:10000/sales_analytics edw_user

CREATE EXTERNAL TABLE IF NOT EXISTS ext_stocks (
id INT,
showroom_id INT,
product_id INT,
quantity INT,
stock_date DATE,
update_date TIMESTAMP,
create_date TIMESTAMP
)
ROW FORMAT DELIMITED
FIELDS TERMINATED BY '|'
STORED AS TEXTFILE
LOCATION '/user/edw_user/sales/staging/stocks';

SELECT * FROM ext_stocks LIMIT 10;

Now we need to create an intermediate table to load the transformed data from staging in order to replace the product natural keys with product surrogate keys.

CREATE TABLE IF NOT EXISTS stg_stocks (
id INT,
showroom_id INT,
product_key INT,
quantity INT,
update_date TIMESTAMP,
create_date TIMESTAMP,
stock_date DATE
)
STORED AS ORC 
TBLPROPERTIES ("orc.compress"="SNAPPY");

Next we will define the Hive managed ORC fact table.

CREATE TABLE IF NOT EXISTS fact_stocks (
id INT,
showroom_id INT,
product_key INT,
quantity INT,
update_date TIMESTAMP,
create_date TIMESTAMP
)
PARTITIONED BY (stock_date DATE)
STORED AS ORC 
TBLPROPERTIES ("orc.compress"="SNAPPY");

!quit

Now we will define a Pig Script to replace the Product natural key with Surrogate key and loading stocks data to final fact table. This script will be used later in oozie workflow manager to schedule the load.

Initial/Delta Setup

vi /home/edw_user/sampledata/transform_stocks.pig

ext_stocks = LOAD 'sales_analytics.ext_stocks' USING org.apache.hive.hcatalog.pig.HCatLoader();
dim_product = LOAD 'sales_analytics.dim_product' USING org.apache.hive.hcatalog.pig.HCatLoader();
active_product = FILTER dim_product BY (active_flag == 'Y');
stocks_product = JOIN ext_stocks BY (product_id), active_product BY (id) PARALLEL 2;
stg_stocks = FOREACH stocks_product GENERATE ext_stocks::id AS id, showroom_id AS showroom_id, prod_key AS product_key, quantity AS quantity, ext_stocks::update_date AS update_date, ext_stocks::create_date AS create_date, stock_date AS stock_date;
STORE stg_stocks INTO 'sales_analytics.stg_stocks' USING org.apache.hive.hcatalog.pig.HCatStorer();
quit;

hdfs dfs -put /home/edw_user/sampledata/transform_stocks.pig /user/edw_user/sales/scripts

Execute the Pig Script to trigger the initial data transformation & loading.

pig -x tez -useHCatalog -f "/home/edw_user/sampledata/transform_stocks.pig"

Finally we load data from staging sales table to final fact table. This script will also be used later by oozie.

vi /home/edw_user/sampledata/load_stocks.hql

set hive.execution.engine=tez;
set hive.optimize.sort.dynamic.partition=true;
set hive.exec.reducers.max=31;

SELECT * FROM stg_stocks LIMIT 10;

INSERT INTO TABLE fact_stocks PARTITION(stock_date) 
SELECT id, showroom_id, product_key, quantity, create_date, update_date, stock_date
FROM stg_stocks DISTRIBUTE BY stock_date;

ANALYZE TABLE fact_stocks PARTITION(stock_date) COMPUTE STATISTICS FOR COLUMNS;

hdfs dfs -put /home/edw_user/sampledata/load_stocks.hql /user/edw_user/sales/scripts

Execute the script to trigger the initial data load.

beeline -u jdbc:hive2://sandbox-hdp.hortonworks.com:10000/sales_analytics -n edw_user -p hadoop -d org.apache.hive.jdbc.HiveDriver -f "/home/edw_user/sampledata/load_stocks.hql"

Let us quickly check the data loaded.

beeline
!connect jdbc:hive2://sandbox-hdp.hortonworks.com:10000/sales_analytics edw_user

SELECT * FROM stg_stocks LIMIT 10;
SELECT * FROM fact_stocks LIMIT 10;
SELECT fact_stocks.* FROM dim_date, fact_stocks where day_date=stock_date and day_of_week_number=7;

select p.make, sum(s.quantity) as qunatity
from fact_stocks s, dim_product p
where s.product_key = p.prod_key
group by p.make;

!quit

Finally copy & archive the stocks datafiles.

vi /home/edw_user/sampledata/archive_stocks.sh

hdfs dfs -mkdir /user/edw_user/sales/archive/stocks/`date +%Y%m%d`
hdfs dfs -mv /user/edw_user/sales/staging/stocks/* /user/edw_user/sales/archive/stocks/`date +%Y%m%d`

hdfs dfs -put /home/edw_user/sampledata/archive_stocks.sh /user/edw_user/sales/scripts
sh /home/edw_user/sampledata/archive_stocks.sh

exit

In the next article we will use Oozie Workflow Manager to define workflows to orchestrate the daily data loading mechanism.