Data engineering

Question 1

Which 2 file formats store data in column based and row based files

Answer 1

parquet = column-based file format
avro = row-based file format

Question 2

What does this SQL command do?
DBCC PDW_SHOWSPACEUSED(‘[scheme].[table]’)

Answer 2

This command specific to microsoft’s SQL server parallel data warehouse (dedicated SQL pool).

DBCC stand for data console commands
PDW_SHOWSPACEUSED show information about storage used by the table. It returns, amongst others, the rows contained per distributions which can be summed up to get the total row count of the table.

Question 3

There is a Retry option for activities in azure data factory. What does this option specify?

Answer 3

How many retries a certain activity should run before failing the whole pipeline

Question 4

There are 3 types of dimensions in dim/fact modelling. What are the characteristics of type 0,1,2

Answer 4

Type 0 = Static, data gets entered then never changes.
Type 1 = Here you just override the data, so if a record should be updated, you simply just override the original row.
Type 2 = These dimension are classified as slowly changing dimensions and often has multiple records of the same instance to track historical data. They often have a start-date, end-date and isCurrent attribute to help identify the current record. Soft deletion can also happen here.

Question 5

In a Azure Stream Analytics job, you see a high number of Backlogged input events. How can you solve this problem.

Answer 5

Increase the number of streaming units assigned to the job.

Question 6

How many distribution is the on a default dedicated sql pool in azure synapse

Answer 6

60

Question 7

What is the 3 kinds distribution that you can use in a dedicated sql pool?

Answer 7

Round-robin (default)
Hash
Replicated

Question 8

How does these 3 distribution work?

Answer 8

Round-robin: is the same concept as in load-balancing, so it just takes the first row and places it on the first distribution, the second row get put on the second distribution and so on.

hash: just like only hashing takes a input x which can be one or more column values, calculates a hash and place it on some distribution, thing with the same hash will of course be placed on the same distribution. Can however create data skew

replicated: Basically just replicates a table across all notes.

Question 9

If you assume you have a data warehouse under normal conditions (star schema, table sizes, how things are joined together and so on…) Where could we see the 3 kinds of distributions be used

Answer 9

Hash = Large fact tables

Replicated = smaller dimension tables

Round-robin: the staging layer here we just need to get the rows placed somewhere before we transform it.

Question 10

There are several Azure Stream Analytics job in place. We want to ensure that the data stream is segmented into distinct time segments and ensure that events don’t overlap. Which windowing function should we use?

Answer 10

Tumbling window.

Question 11

For a normal General Purpose V2 storage account, what needs to be enabled for it to behave like a data lake gen 2 account

Answer 11

Hierarchical namespace.

Question 12

There are several Azure Stream Analytics job in place. We want to ensure to output events only for points in time when the content of the window actually changes? Which windowing function should we use?

Answer 12

Sliding window.

Question 13

What is data skew

Answer 13

When you use hash distribution and you hash a value which appears often, the data will accumulate on certain distributions and others will barely have any data. This causes sub optimal performance

Question 14

Which service could you use to host your data-warehouse?

Answer 14

Azure Synapse Analytics

Question 15

Which sort of SQL pools are able to work with EXTERNAL TABLES?

Answer 15

Both the serverless and dedicated SQL Server pools can use EXTERNAL TABLES.

Question 16

Are EXTERNAL TABLES persisted or just a logical view on data?

Answer 16

A logical view.

Question 17

What is polybase?

Answer 17

It is the most effective way to move large amount of data from external sources such as csv, json, avro, parquet into the dedicated sql server pool.

Question 18

Based on microsoft own number what is the minimum number of rows you should have per distribution and partition to get good performance?

Answer 18

1 million rows.

Question 19

In Azure data factory if you to store pipeline-run data for 60 days, what should you then configure?

Answer 19

Diagnostic setting

Question 20

What is a heap table

Answer 20

heap tables are tables that does not have a clustered index. Therefore data is not stored in any particular order, however they do not have any extra overhead of indexes which means data insertion is faster as data does not need to be sorted first. This makes them a good option for small or temporary tables.

Question 21

what is type 3 dimensions.

Answer 21

In this approach you add columns instead for rows. so if a named changed in a dimension record, you add a new column named something like “changed name” and the name column might be rewritten as “original name”

Question 22

what is indexes

Answer 22

Indexes in SQL are tools that allow developers to create additional structures within a database to speed up data retrieval operations. They are categorized into rowstore and columnstore indexes. Some Rowstore indexes change psychical order of the data stored in the table (which is row-wise), and other creates a separate structure to facilitate faster search and access to rows. Columnstore indexes, on the other hand, organize data physically in a column-wise format, optimizing for queries that read large volumes of data but only a few columns

Question 23

In a normal relational database system, how is the data normally stored

Answer 23

Row-wise

Question 24

In a data warehouse system, how is the data normally stored

Answer 24

column-wise

Question 25

In azure synapse with a data-warehouse associated (dedicated sql server pool), how can you create a column-wise and a row-wise index?

Answer 25

CLUSTERED INDEX (column_name) can be used to create a rowstore index. CLUSTERED COLUMNSTORE INDEX, because you here split the table up in columns you dont need to provide a input.

Question 26

What is the main difference between CLUSTERED INDEXES and NONCLUSTERED INDEXES?

Answer 26

There can only be one clustered index on a table whereas there can be multiple NONCLUSTERED INDEXES on tables on a table. Both store the data row-wise however CLUSTERED INDEXES change the physical order of the data whereas NONCLUSTERED INDEXES creates separate data structures which points to certain rows so queries can be speed up.

Question 27

When should you consider using heap tables or clustered index?

Answer 27

If your table contains columns of the type varchar(max), nvarchar(max) or varbinary(max): because it would be very costly to store a column that contains this big amount of data compared to the other columns. if you want to create temporary tables if you have a table with less than 60 millions rows.

Question 28

How do you create a heap table in TSQL?

Answer 28

WITH ( HEAP, DISTRIBUTIOIN = ROUND_ROBIN -- could of course be any distribution. )

Question 29

What are table partitions?

Answer 29

Partitions are a tool to speed up certain queries by separating the data into smaller logical(aka. not physical) partitions. This is not the same as distributions, as that only has do with splitting data between distributions(nodes) . Partitions are helpful when you have something that nicely split your data into even size. E.g. dates. Helps in particular with WHERE queries.

Question 30

How does partitions work?

Answer 30

By splitting up the data into logical partitions a distribution can more effectively locate data by using a process called partition elimination. This allows each distribution to narrow down the amount of data it needs to process.

Question 31

What is a windowing function?

Answer 31

A windowing function is a SQL tool that allows you to split rows into different sets and apply aggregations to each set. It work by using the OVER clause, and could be followed with the PARTITION BY keywords. This way you can abstractly create windows for your data.

Question 32

What is Mapping Data flows

Answer 32

This is a feature which resides inside of Azure data factory. It helps your visualise your data transformations and perform them without any code. It runs on Apache spark clusters.

Question 33

What is the derived column activity?

Answer 33

This is a sub feature within Mapping Data Flows which allows you to create new columns based on the data available in the pipeline. If you need a extra column in your fact table which is a product of two other columns values. Then this activity is well suited.

Question 34

What are chached sinks and looks up for in the mapping data flow

Answer 34

Chached sinks can be used to store results from queries or similar, earlier in the pipeline and through lookups be found and used later in the flow.

Question 35

When working in azure factory and your pipelines becomes troublesome time wise. what feature can you use to test your pipelines quicker than just running these pipelines.

Answer 35

the Data flow debug feature can