💎🏭SQL Reloaded: Microsoft Fabric's Delta Tables in Action - Import data from CSV Files into Delta Table

Microsoft provides a set of labs and exercises that can be used to learn working with data in Fabric, however the real learning comes when one considers an example that introduces something new. As I've downloaded some time ago an archive with several datasets on Sales forecast from the Kaggle website, I tried to import the Features dataset in different ways and see how it goes.

Store,Date,Temperature,Fuel_Price,MarkDown1,MarkDown2,MarkDown3,MarkDown4,MarkDown5,CPI,Unemployment,IsHoliday
1,2010-02-05,42.31,2.572,NA,NA,NA,NA,NA,211.0963582,8.106,FALSE
1,2010-02-12,38.51,2.548,NA,NA,NA,NA,NA,211.2421698,8.106,TRUE
1,2010-02-19,39.93,2.514,NA,NA,NA,NA,NA,211.2891429,8.106,FALSE
1,2010-02-26,46.63,2.561,NA,NA,NA,NA,NA,211.3196429,8.106,FALSE
1,2010-03-05,46.5,2.625,NA,NA,NA,NA,NA,211.3501429,8.106,FALSE

Within an existing lakehouse, one can import the CSV as it is via 'Files/Upload' and once the data is imported, once can navigate to the file and use 'Load to Tables/New Table' to important the data into a managed table. Unfortunately, because some of the numeric fields include also literal values "NA" for the columns with NULLs, their data type is considered as varchar(8000), which is not ideal for calculations:

-- table created via Load to Tables
CREATE TABLE [dbo].[walmart_features](
	[Store] [int] NULL,
	[Date] [date] NULL,
	[Temperature] [float] NULL,
	[Fuel_Price] [float] NULL,
	[MarkDown1] [varchar](8000) NULL,
	[MarkDown2] [varchar](8000) NULL,
	[MarkDown3] [varchar](8000) NULL,
	[MarkDown4] [varchar](8000) NULL,
	[MarkDown5] [varchar](8000) NULL,
	[CPI] [varchar](8000) NULL,
	[Unemployment] [varchar](8000) NULL,
	[IsHoliday] [bit] NULL
) ON [PRIMARY]

This could be fixed by replacing the NA values with an empty value, which I did and used this version for the next steps. 

I tried then using Spark to import the data, though then all the fields are defined as varchar(8000).

-- table created via Spark
CREATE TABLE [dbo].[walmart_features2](
	[Store] [varchar](8000) NULL,
	[Date] [varchar](8000) NULL,
	[Temperature] [varchar](8000) NULL,
	[Fuel_Price] [varchar](8000) NULL,
	[MarkDown1] [varchar](8000) NULL,
	[MarkDown2] [varchar](8000) NULL,
	[MarkDown3] [varchar](8000) NULL,
	[MarkDown4] [varchar](8000) NULL,
	[MarkDown5] [varchar](8000) NULL,
	[CPI] [varchar](8000) NULL,
	[Unemployment] [varchar](8000) NULL,
	[IsHoliday] [varchar](8000) NULL
) ON [PRIMARY]
GO

So, is needed to define the schema explicitly, however I had to import the IsHoliday as string and cast the value explicitly to a Boolean using a second data frame (see alternatives):

from pyspark.sql.types import *
from pyspark.sql.functions import *

#define schema
featuresSchema = StructType([
      StructField("Store", IntegerType())
    , StructField("Date", DateType())
    , StructField("Temperature",  DecimalType(13,2))
    , StructField("Fuel_Price", DecimalType(13,2))
    , StructField("MarkDown1", DecimalType(13,2))
    , StructField("MarkDown2", DecimalType(13,2))
    , StructField("MarkDown3", DecimalType(13,2))
    , StructField("MarkDown4", DecimalType(13,2))
    , StructField("MarkDown5", DecimalType(13,2))
    , StructField("CPI", DecimalType(18,6))
    , StructField("Unemployment", DecimalType(13,2))
    , StructField("IsHoliday", StringType())
])

# Load a file into a dataframe
df = spark.read.load('Files/OpenSource/features2.csv'
    , format='csv'
    , schema = featuresSchema
    , header=True)

# do the conversion for isHoliday
df2 = df.withColumn("IsHoliday", df.IsHoliday.cast(BooleanType()))

# Save the dataframe as a delta table
df2.write.format("delta").saveAsTable("walmart_features3")

Now, table's definition looks much better:

-- table created via Spark with explicit schema
CREATE TABLE [dbo].[walmart_features3](
	[Store] [int] NULL,
	[Date] [date] NULL,
	[Temperature] [decimal](13, 2) NULL,
	[Fuel_Price] [decimal](13, 2) NULL,
	[MarkDown1] [decimal](13, 2) NULL,
	[MarkDown2] [decimal](13, 2) NULL,
	[MarkDown3] [decimal](13, 2) NULL,
	[MarkDown4] [decimal](13, 2) NULL,
	[MarkDown5] [decimal](13, 2) NULL,
	[CPI] [decimal](18, 6) NULL,
	[Unemployment] [decimal](13, 2) NULL,
	[IsHoliday] [bit] NULL
) ON [PRIMARY]
GO

Comments:
(1) I tried to apply the schema change directly on the initial data frame, though the schema didn't change:

df.withColumn("IsHoliday", df.IsHoliday.cast(BooleanType()))

(2) For the third method one could have left the NA in because by the conversion a NULL will be considered. Conversely, it might be needed to check if there are other values that fail the conversion. 

(3) The following warning in the Notebook when running the above code is a hint that something went wrong during the conversion (e.g. decimals were cut): 

"Your file(s) might include corrupted records"

(4) Especially for the transformed values it makes sense to look at the values (at last when the dataset isn't too big):

-- validating the values for the Boolean data field
SELECT IsHoliday
, count(*) NoRecords
FROM dbo.walmart_features3
GROUP BY IsHoliday
ORDER BY 1

(5) The tables can be deleted directly in the lakehouse or via PySpark (observe the catalog.table_name):

#dropping the table
spark.sql('DROP TABLE IF EXISTS UAT.walmart_features3') 

At the beginning probably it makes sense to remove "IF EXISTS" to make sure that the table is available.

(6) For those who run into a similar issue, the SQL Endpoint for the lakehouse is read only, therefore attempting to delete a table via SSMS will result in such an error:

Drop failed for Table 'dbo.walmart_features2'.  (Microsoft.SqlServer.Smo)
The external policy action 'Microsoft.Sql/Sqlservers/Databases/Schemas/Tables/Drop' was denied on the requested resource.
Cannot drop the table 'walmart_features2', because it does not exist or you do not have permission.

Happy coding!

🧭Business Intelligence: Perspectives (Part XI: Ways of Thinking about Data)

Business Intelligence Series

One can observe sometimes the tendency of data professionals to move from a business problem directly to data and data modeling without trying to understand the processes behind the data. One could say that the behavior is driven by the eagerness of exploring the data, though even later there are seldom questions considered about the processes themselves. One can argue that maybe the processes are self-explanatory, though that’s seldom the case. 

Conversely, looking at the datasets available on the web, usually there’s a fact table and the associated dimensions, the data describing only one process. It’s natural to presume that there are data professionals who don’t think much about, or better said in terms of processes. A similar big jump can be observed in blog posts on dashboards and/or reports, bloggers moving from the data directly to the data model. 

In the world of complex systems like Enterprise Resource Planning (ERP) systems thinking in terms of processes is mandatory because a fact table can hold the data for different processes, while processes can span over multiple fact-like tables, and have thus multiple levels of detail. Moreover, processes are broken down into sub-processes and procedures that have a counterpart in the data as well. 

Moreover, within a process there can be multiple perspectives that are usually module or role dependent. A perspective is a role’s orientation to the word for which the data belongs to, and it’s slightly different from what the data professional considers as view, the perspective being a projection over a set of processes within the data, while a view is a projection of the perspectives into the data structure. 

For example, considering the order-to-cash process there are several sub-processes like order fulfillment, invoicing, and payment collection, though there can be several other processes involved like credit management or production and manufacturing. Creating, respectively updating, or canceling an order can be examples of procedures. 

The sales representative, the shop worker and the accountant will have different perspectives projected into the data, focusing on the projection of the data on the modules they work with. Thinking in terms of modules is probably the easiest way to identify the boundaries of the perspectives, though the rules are occasionally more complex than this.

When defining and/or attempting to understand a problem it’s important to understand which perspective needs to be considered. For example, the sales volume can be projected based on Sales orders or on invoiced Sales orders, respectively on the General ledger postings, and the three views can result in different numbers. Moreover, there are partitions within these perspectives based on business rules that determine what to include or exclude from the logic. 

One can define a business rule as a set of conditional logic that constraints some part of the data in the data structures by specifying what is allowed or not, though usually we refer to a special type called selection business rule that determines what data are selected (e.g. open Purchase orders, Products with Inventory, etc.). However, when building the data model we need to consider business rules as well, though we might need to check whether they are enforced as well. 

Moreover, it’s useful to think also in terms of (data) entities and sub-entities, in which the data entity is an abstraction from the physical implementation of database tables. A data entity encapsulates (hides internal details) a business concept and/or perspective into an abstraction (simplified representation) that makes development, integration, and data processing easier. In certain systems like Dynamics 365 is important to think at this level because data entities can simplify data modelling considerably.

Previous Post <<||>>  Next Post

💎SQL Server: Copilot in Azure SQL Database is in Private Preview (SQL Reloaded) [new feature]

Following the news on Copilot and its adoption in the various areas, probably you asked yourself when it will become available in SQL Server. The good news is that the Copilot in Azure SQL Database is already in Private Preview (see [1], [2]). The bad news is that for the early adoption program, Microsoft looks only for customers that can provide feedback based on real-world use cases, and one will need a valid Tenant ID for signing up for the PP.

The new feature comes with two Azure portal experiences:

• Natural language to SQL: within the Azure portal query editor the natural language prompts are translated into SQL;
• Azure Copilot integration: provides customers with self-guided assistance for managing databases and troubleshooting. 

In the Data Exposed webcast with Anna Hoffman [1], the invitee, Joe Sack, showcases the use of both experiences. In what concerns the first experience, besides simple joins, Joe also showcased the use of a PIVOT based on the prompt:

"Show me a pivot summary table that displays the total number of properties sold in each year from 2020 to 2023".

-- output from Copilot
SELECT *
FROM 
(
	SELECT YEAR(SaleDate) AS Year, COUNT(*) AS TotalPropertiesSold
	FROM [PropertySearch].[dbo].[PropertyHistory]
	WHERE YEAR(SaleDate) BETWEEN 2020 AND 2023
	GROUP BY YEAR(SalesDate)
) AS PivotDate
PIVOT
(
	SUM(TotalPropertiesSold)
	FOR Year IN ([2020], [2021], [2022], [2023]
) AS PivotTable;

Even if this is a modest example, the great thing is that it generates the skeleton on which the logic can be built, which can really help beginners, though it can be also useful for more seasoned data professionals as it saves time. Probably, in a first phase this will be the main benefit of Copilot - to generate skeletons or templates on which further logic can be built. Hopefully in time it will do much more than that.

I wonder how complex the prompts can become and how can the logic be structured to create a multistep scenario. The Copilot versions from other areas showed that complex prompts give results, the question is whether Copilot can devise the steps in an optimum manner, much like a seasoned data professional does. 

The feature utilizes for the moment the table and view names, column names, primary key, and foreign key metadata to generate T-SQL code. Probably, it makes sense to also use index and statistics information, query hints and all the arsenal usually needed by data professionals to optimize a query. Conversely, maybe the second experience could be used for optimizing and troubleshooting the query. 

I'd really love to test this feature, though probably I'll need to wait until it becomes publicly available. In the meanwhile, one can play with the GitHub Copilot [3] or install Copilot in Azure Data Studio [4].

References:
[1] Data Exposed (2024) Introducing Copilot in Azure SQL Database (Private Preview) (link)
[2] Azure SQL Blog (2024) Microsoft Copilot in Azure extends capabilities to Azure SQL Database (Private Preview) by Joe Sack (link)
[3] Azure SQL Blog (2023) GitHub Copilot for SQL Developers: Turbocharge your SQL Development, by Subhojit Basak (link)
[4] Microsoft Developer (2023) Copilot is now in Azure Data Studio and this is how it can help you! (link)

🧭Business Intelligence: Why Data Projects Fail to Deliver Real-Life Impact (Part IV: Making It in the Statistics)

Business Intelligence Series

Various sources (e.g., [1], [2], [3]) advance the failure rates for data projects somewhere between 70% and 85%, rates which are a bit higher than the failure of standard projects estimated at 60-75% but not by much. This means that only 2-3 out of 10 projects will succeed and that’s another reason to plan for failure, respectively embrace the failure. 

Unfortunately, the statistics advanced on project failure have no solid fundament and should be regarded with circumspection as long the methodology and information about the population used for the estimates aren’t shared, though they do reflect an important point – many data projects do fail! It would be foolish to think that your project will not fail just because you’re a big company, and you have the best resources, and you have a proven rate of success, and you took all the precautions for the project not to fail.

Usually at the end of a project the team meets together to document the lessons learned in the hope that the next projects will benefit from them. The team did learn something, though as the practice shows even if the team managed to avoid some issues, other issues will impact the next similar project, leading to similar variances. One can summarize this as "on the average the impact of new issues and avoided known issues tends to zero out" or "on average, the plusses and minuses balance each other across projects". It’s probably a question of focus – if organizations focus too much on certain aspects, other aspects are ignored and/or unseen. 

So, your first data project will more likely fail. The question is: what do you do about it? It’s important to be aware of why projects and data projects fail, though starting to consider and monitor each possible issue can prove to be ineffective. One can, however, create a risk register from the list and estimate the rates for each of the potential failures, respectively focus on only the top 3-5 which have the highest risk. Of course, one should reevaluate the estimates on a regular basis though that’s Risk Management 101. 

Besides this, one should focus on how the team can make the project succeed. When adopting a technology, methodology or set of processes, it’s recommended to start with a proof-of-concept (PoC). To make the PoC a helpful experience it’s probably important to start with a topic that’s not too big to handle, but that also involves some complexity that would allow the organization to evaluate the targeted set of tools and technologies. It can also be a topic for which other organizations have made important progress, respectively succeed. The temptation is big to approach the most stringent issues in the organization, respectively to build something big that can have an enormous impact for the organization. Jumping too soon into such topics can just increase the chances of failure. 

One can also formulate the goals, objectives and further requirements in a form that allows the organization to build upon them even if the project fails. A PoC is about learning, building a foundation, doing the groundwork, exploring, mapping the unknown, and identifying what's still missing to make progress, respectively closing the full circle. A PoC is less about overachievement and a big impact, which can happen, though is a consequence of the good work done in the PoC. 

The bottom line, no matter whether you succeed or fail, once you start a project, you’ll still make it in the statistics! More important is what you’ve learnt after the first data project, respectively how you can use the respective knowledge in further projects to make a difference!

Previous Post <<||>> Next Post

References:
[1] Harvard Business Review (2023) Keep Your AI Projects on Track, by Iavor Bojinov (link)
[2] Cognilytica (2023) The Shocking Truth: 70-80% of AI Projects Fail! (link)
[3] VentureBeat (2019) Why do 87% of data science projects never make it into production? (link)

🧭Business Intelligence: Why Data Projects Fail to Deliver Real-Life Impact (Part III: Failure through the Looking Glass)

Business Intelligence Series

There’s a huge volume of material available on project failure – resources that document why individual projects failed, why in general projects fail, why project members, managers and/or executives think projects fail, and there seems to be no other more rewarding activity at the end of a project than to theorize why a project failed, the topic culminating occasionally with the blaming game. Success may generate applause, though it's failure that attracts and stirs the most waves (irony, disapproval, and other similar behavior) and everybody seems to be an expert after the consumed endeavor. 

The mere definition of a project failure – not fulfilling project’s objectives within the set budget and timeframe - is a misnomer because budgets and timelines are estimated based on the information available at the beginning of the project, the amount of uncertainty for many projects being considerable, and data projects are no exceptions from it. The higher the uncertainty the less probable are the two estimates. Even simple projects can reveal uncertainty especially when the broader context of the projects is considered. 

Even if it’s not a common practice, one way to cope with uncertainty is to add a tolerance for the estimates, though even this practice probably will not always accommodate the full extent of the unknown as the tolerances are usually small. The general expectation is to have an accurate and precise landing, which for big or exploratory projects is seldom possible!

Moreover, the assumptions under which the estimates hold are easily invalidated in praxis – resources’ availability, first time right, executive’s support to set priorities, requirements’ quality, technologies’ maturity, etc. If one looks beyond the reasons why projects fail in general, quite often the issues are more organizational than technological, the lack of knowledge and experience being some of the factors. 

Conversely, many projects will not get approved if the estimates don’t look positive, and therefore people are pressured in one way or another to make the numbers fit the expectations. Some projects, given their importance, need to be done even if the numbers don’t look good or can’t be quantified correctly. Other projects represent people’s subsistence on the job, respectively people's self-occupation to create motion, though they can occasionally have also a positive impact for the organizations. These kinds of aspects almost never make it in statistics or surveys. Neither do the big issues people are afraid to talk about. Where to consider that in the light of politics and office’s grapevine the facts get distorted!

Data projects reflect all the symptoms of failure projects have in general, though when words like AI, Statistics or Machine Learning are used, the chances for failure are even higher given that the respective fields require a higher level of expertise, the appropriate use of technologies and adherence to the scientific process for the results to be valid. If projects can benefit from general recipes, respectively established procedures and methods, their range of applicability decreases when the mentioned areas are involved. 

Many data projects have an exploratory nature – seeing what’s possible - and therefore a considerable percentage will not reach production. Moreover, even those that reach that far might arrive to be stopped or discarded sooner or later if they don’t deliver the expected value, and probably many of the models created in the process are biased, irrelevant, or incorrectly apply the theory. Where to add that the mere use of tools and algorithms is not Data Science or Data Analysis. 

The challenge for many data projects is to identify which Project Management (PM) best practices to consider. Following all or no practices at all just increases the risks of failure!

Previous Post <<||>> Next Post

🧭Business Intelligence: Why Data Projects Fail to Deliver Real-Life Impact (Part II: There's Value in Failure)

Business Intelligence Series

"Results are nothing; the energies which produce them

and which again spring from them are everything."

(Wilhelm von Humboldt,  "On Language", 1836)

When the data is not available and is needed on a continuous basis then usually the solution is to redesign the processes and make sure the data becomes available at the needed quality level. Redesign involves additional costs for the business; therefore, it might be tempting to cancel or postpone data projects, at least until they become feasible, though they’re seldom feasible. 

Just because there’s a set of data, this doesn’t mean that there is important knowledge to be extracted from it, respectively that the investment is feasible. There’s however value in building experience in the internal resources, in identifying the challenges and the opportunities, in identifying what needs to be changed for harnessing the data. Unfortunately, organizations expect that somebody else will do the work for them instead of doing the jump by themselves, and this approach more likely will fail. It’s like expecting to get enlightened after a few theoretical sessions with a guru than walking the path by oneself. 

This is reflected also in organizations’ readiness to do the required endeavors for making the jump on the maturity scale. If organizations can’t approach such topics systematically and address the assumptions, opportunities, and risks adequately, respectively to manage the various aspects, it’s hard to believe that their data journey will be positive. 

A data journey shouldn’t be about politics even if some minds need to be changed in the process, at management as well as at lower level. If the leadership doesn’t recognize the importance of becoming an enabler for such initiatives, then the organization probably deserves to keep the status quo. The drive for change should come from the leadership even if we talk about data culture, data strategy, decision-making, or any critical aspect.

An organization will always need to find the balance between time, scope, cost, and quality, and this applies to operations, tactics, and strategies as well as to projects.  There are hard limits and lot of uncertainty associated with data projects and the tasks involved, limits reflected in cost and time estimations (which frankly are just expert’s rough guesses that can change for the worst in the light of new information). Therefore, especially in data projects one needs to be able to compromise, to change scope and timelines as seems fit, and why not, to cancel the projects if the objectives aren’t feasible anymore, respectively if compromises can’t be reached.

An organization must be able to take the risks and invest in failure, otherwise the opportunities for growth don’t change. Being able to split a roadmap into small iterative steps that allow besides breaking down the complexity and making progress to evaluate the progress and the knowledge resulted, respectively incorporate the feedback and knowledge in the next steps, can prove to be what organizations lack in coping with the high uncertainty. Instead, organizations seem to be fascinated by the big bang, thinking that technology can automatically fill the organizational gaps.

Doing the same thing repeatedly and expecting different results is called insanity. Unfortunately, this is what organizations and service providers do in what concerns Project Management in general and data projects in particular. Building something without a foundation, without making sure that the employees have the skillset, maturity and culture to manage the data-related tasks, challenges and opportunities is pure insanity!

Bottom line, harnessing the data requires a certain maturity and it starts with recognizing and pursuing opportunities, setting goals, following roadmaps, learning to fail and getting value from failure, respectively controlling the failure. Growth or instant enlightenment without a fair amount of sweat is possible, though that’s an exception for few in sight!

Previous Post <<||>> Next Post

🧭Business Intelligence: Why Data Projects Fail to Deliver Real-Life Impact (Part I: First Thoughts)

Business Intelligence Series

A data project has a set of assumptions and requirements that must be met, otherwise the project has a high chance of failing. It starts with a clear idea of the goals and objectives, and they need to be achievable and feasible, with the involvement of key stakeholders and the executive without which it’s impossible to change the organization’s data culture. Ideally, there should also be a business strategy, respectively a data strategy available to understand the driving forces and the broader requirements. 

An organization’s readiness is important not only in what concerns the data but also the things revolving around the data - processes, systems, decision-making, requirements management, project management, etc. One of the challenges is that the systems and processes available can’t be used as they are for answering important business questions, and many of such questions are quite basic, though unavailability or poor quality of data makes this challenging if not impossible. 

Thus, when starting a data project an organization must be ready to change some of its processes to address a project’s needs, and thus the project can become more expensive as changes need to be made to the systems. For many organizations the best time to have done this was when they implemented the system, respectively the integration(s) between systems. Any changes made after that come in theory with higher costs derived from systems and processes’ redesign.

Many projects start big and data projects are no exception to this. Some of them build a costly infrastructure without first analyzing the feasibility of the investment, or at least whether the data can form a basis for answering the targeted questions. On one side one can torture any dataset and some knowledge will be obtained from it (aka data will confess), though few datasets can produce valuable insights, and this is where probably many data projects oversell their potential. Conversely, some initiatives are worth pursuing even only for the sake of the exposure and experience the employees get. However, trying to build something big only through the perspective of one project can easily become a disaster. 

When building a data infrastructure, the project needs to be an initiative given the transformative potential such an endeavor can have for the organization, and the different aspects must be managed accordingly. It starts with the management of stakeholders’ expectations, with building a data strategy, respectively with addressing the opportunities and risks associated with the broader context.

Organizations recognize that they aren’t capable of planning and executing such a project or initiative, and they search for a partner to lead the way. Becoming overnight such a partner is more than a challenge as a good understanding of the industry and the business is needed. Some service providers have such knowledge, at least in theory, though the leap from knowledge to results can prove to be a challenge even for experienced service providers. 

Many projects follow the pattern: the service provider comes, analyzes the requirements, builds something wonderful, the solution is used for some time and then the business realizes that the result is not what was intended. The causes are multiple and usually form a complex network of causality, though probably the most important aspect is that customers don’t have the in-house technical resources to evaluate the feasibility of requirements, solutions, respectively of the results. Even if organizations involve the best key users, are needed also good data professionals or similar resources who can become the bond between the business and the services provider. Without such an intermediary the disconnect between the business and the service provider can grow with all the implications. 

Previous Post <<||>> Next Post

🏭🗒️Microsoft Fabric: Data Governance [Notes]

Disclaimer: This is work in progress intended to consolidate information from various sources for learning purposes. For the latest information please consult the documentation (see the links below)! 

Last updated: 23-May-2024

[Microsoft Fabric] Data Governance

• {definition}set of capabilities that help organizations to manage, protect, monitor, and improve the discoverability of data, so as to meet data governance (and compliance) requirements and regulations [2]
• several built-in governance features are available to manage and control the data within Fabric (MF)  [1]
• {feature} endorsement [aka content endorsement] 
• {definition} formal process performed by admins to endorse MF items
• {benefit} allows admins to designate specific MF items as trusted and approved for use across the organization [1]
• establishes trust in data assets by promoting and certifying specific MF items [1]
• users know which assets they can trust and rely on for accurate information [1]
• endorsed assets are identified with a badge that indicates they have been reviewed and approved [1]
• {scope} applies to all MF items except dashboards [1]
• {benefit} helps admin manage the overall growth of items across your environment [1]
• {feature} promoting [aka content promoting] 
• {definition} formal process performed by contributors or admins to promote content
• promoted content appears with a Promoted badge in the MF portal [1]
• workspace members with the contributor or admin role can promote content within a workspace [1]
• MF admin can promote content across the organization [1]
• {feature} certification [aka content certification]
• {definition} formal process that involves a review of the content by a designated reviewer and managed by the admin [1]
• can be customized to meet organization’s needs [1]
• users can request item certification from an admin [1]
• via Request certification from the More menu [1]
• the certified content appears with a Certified badge in the Fabric portal [1]
• {benefit} allows organizations to label items considered to be quality items [1]
• an organization can certify items to identify them an as authoritative sources for critical information [1]
• ⇐ all Fabric items except Power BI dashboards can be certified [1]
• {benefit} allows to specify certifiers who are experts in the domain [1]
• domain level settings
• enable or disable certification of items that belong to the domain [1]
• provides a URL to documentation that is relevant to certification in the domain [1]
• {feature} tenant (aka Microsoft Fabric tenant, MF tenant)
• a single instance of Fabric for an organization that is aligned with a Microsoft Entra ID
• can contain any number of workspaces
• {feature} workspaces
• {definition} a collection of items that brings together different functionality in a single environment designed for collaboration
• can be assigned to teams or departments based on governance requirements and data boundaries [2]
• are associated with domains [3]
• ⇐ {benefit} allows to group data into business domains
• all the items in the workspace are then associated with the domain, and they receive a domain attribute as part of their metadata [3]
• ⇐ {benefit} enables a better consumption experience [1]
• {benefit} enables better discoverability and governance [2]
• {feature} domains [Notes]
• {definition} a way of logically grouping together data in an organization that is relevant to a particular area or field [1]
• allows to group data by business domains
• ⇒{benefit} allows business domains to manage their data according to their specific regulations, restrictions, and needs [3]
• {feature} subdomains
  • {definition} a way for fine tuning the logical grouping data under a domain [1]
  • ⇐ subdivisions of a domain
• {feature} labeling
• default labeling, label inheritance, and programmatic labeling, 
• {benefit} help achieve maximal sensitivity label coverage across MF [2]
• once labeled, data remains protected even when it's exported out of MF via supported export paths [2]
• [Purview Audit] compliance admins can monitor activities on sensitivity labels
• {feature|preview} folders
• {definition} a way of logically grouping MF items
• {feature|preview} tags
• {benefit} allow managing Fabric items for enhanced compliance, discoverability, and reuse
• {feature} scanner API
• a set of admin REST APIs 
• {benefit} allows to scan MF items for sensitive data [1]
• can be used to scan both structured and unstructured data [1]
• {concept} metadata scanning
• facilitates governance of data by enabling cataloging and reporting on all the metadata of organization's Fabric items [1]
• it needs to be set up by Admin before metadata scanning can be run [1]
• {concept} data lineage
• {definition} 
• {benefit} allows to track the flow of data through Fabric [1]
• {benefit} allows to see where data comes from, how it's transformed, and where it goes [1]
• {benefit} helps understand the data available in Fabric, and how it's being used [1]
• {concept} Fabric item (aka MF item)
• {definition} a set of capabilities within an experience
• form the building blocks of the Fabric platform
• {type} data warehouse
• {type} data pipeline
• {type} semantic model
• {type} reports
• {type} dashboards
• {type} notebook
• {type} lakehouse
• {type} metric set

Previous Post <<||>>  Next Post

Acronyms:
API - Application Programming Interface
MF - Microsoft Fabric

Resources:
[1] Microsoft Learn (2023) Administer Microsoft Fabric (link)
[2] Microsoft Learn - Fabric (2024) Governance overview and guidance (link)
[3] Microsoft Learn: Fabric (2023) Fabric domains (link)
[4] Establishing Data Mesh architectural pattern with Domains and OneLake on Microsoft Fabric, by Maheswaran Arunachalam (link) 

💎SQL Reloaded: SQL Antipatterns (Part I: JOINs, UNIONs & DISTINCT)

Introduction

SQL antipatterns refer in general to common mistakes made when developing SQL code, though the term can refer also to situations in which even if the code is syntactically and logically correct, it's either suboptimal, unclear or even incorrect. Therefore "mistake" can cover a wide range of scenarios, some that can be ignored, while others need to be addressed accordingly. 

In this post I consider a few antipatterns observed especially in data warehouses (DWHs). Let's look at the below code created to exemplify several scenarios:

-- Products in open orders (initial query)
SELECT DISTINCT ITM.ProductId                                   -- (1) use of DISTINCT
, ITM.ProductNumber
, ITM.Name 
, ITM.Color 
, ITM.Style 
, ITM.Size
FROM Production.Product ITM
    LEFT JOIN (							-- (5) use of JOIN instead of EXISTS
	-- Open Purchase orders 
	SELECT DISTINCT POL.ProductId
	, 'POs' Source                                          -- (7) use columns not needed in output
	FROM Purchasing.PurchaseOrderDetail POL                 
	     LEFT JOIN Purchasing.PurchaseOrderHeader POH       -- (2) use of LEFT JOIN instead of FULL JOIN
		  ON POL.PurchaseOrderID = POH.PurchaseOrderID
	WHERE POH.Status = 1 -- pending 
	UNION					                -- (3) use of UNION
	-- Open Sales orders 
	SELECT DISTINCT SOL.ProductId
	, 'SOs' Source
	FROM Sales.SalesOrderDetail SOL
	    LEFT JOIN Sales.SalesOrderHeader SOH
		  ON SOL.SalesOrderID = SOH.SalesOrderID
	WHERE SOH.Status = 1 -- in process		        -- (4) use of OR instead of IN
	   OR SOH.Status = 2 -- approved
	) DAT
	ON ITM.ProductID = DAT.ProductID
WHERE DAT.ProductID IS NOT NULL 
ORDER BY ITM.ProductNumber			                -- (6) using too many columns in ORDER BY
, ITM.Name 
, ITM.Color 
, ITM.Style 
, ITM.Size

(1) Use of DISTINCT 

DISTINCT is a dirty way to remove the duplicates from a dataset. Sometimes it makes sense to use it to check something fast, though it should be avoided into code intended for a production environment because it can lead to unexpected behavior especially when selecting all the columns using the "*" (SELECT DISTINCT *).

I saw tools and developers adding a DISTINCT in almost each step, independently on whether it was necessary or not. One can thus but wonder whether the DISTINCT was added to fix a bigger issue with the data in the DWH, to remove special duplicates imposed by the logic or just as poor practice. Unfortunately, when it's used frequently, it can become challenging to investigate its use and discover the actual issues from the DWH.

There are several approaches to eliminate DISTINCTs from the code: GROUP BY, ranking functions or upon case also code rewrites.

(2)  Use of LEFT JOIN Instead of FULL JOIN

When refreshing a DWH there can be the case that related data are out of synch. It would be the case of Purchase or Sales orders where the headers and lines are out of synch, headers existing without lines and/or vice versa. A common practice is to use a FULL JOIN and thus to eliminate such exceptions, though there are also entitled uses of a LEFT JOIN. This antipattern resumes however to the cases in which logically should be used a FULL JOIN, though a LEFT JOIN is used instead.

In the above example there are two distinct occurrences of this pattern: the relationship between the header and lines in the inner query, respectively the LEFT JOIN with a NOT NULL constraint in the outer query. The latter use is useful when during testing one wants to see all the Products, though bringing this further into production may rise some eyebrows even if it's not necessarily wrong. Anyway, the database engine should be smart enough to recognize such a scenario. However, for the header vs lines case, upon case the plan generated might be suboptimal. 

One of the best practices when writing SQL queries is to state one's intent clearly in what the logic concerns. Using a LEFT JOIN instead of a FULL JOIN can make people raise questions about the actual need. When the need is not properly documented, some developer may even go and change the joins. There can be for example business cases that are not cover by the current data, but as soon as case appears it will lead to incorrect logic!

Similarly, splitting a piece of logic into two or more steps unnecessarily can create confusion. There can be however also entitled s situations (e.g. query optimization), which ideally should be documented.

(3) Use of UNION

When a UNION is used, the values returned by the first query will be checked against the values of the second query, and thus unnecessary comparisons occur even if they are not needed. This depends also on the business context, which might not be easily to identify from the query (especially when the reviewer doesn't know the business case). 

The misuse of a UNION will not make a big difference when the volume of data is small, though the more data are processed by the query, the higher the impact. 

Besides the proper use of the UNION, there are also situations in which a query rewrite can eliminate the need for a UNION (see the rewritten query below).

(4) use of OR instead of IN

One can occasionally find queries in which a OR was used for 10 to 50 distinct values as in the example above. Even if the database engine generate in both cases the same query plan, it's easier to read and maintain a query that used IN. However, if the number of values go beyond a certain value, other techniques should be used to improve the performance.

The only benefit I see for a OR is to document meaning's values or remove during testing one of the values, especially when the list is a selection coming from the user. Frankly, it's not the appropriate way for documenting logic (even if I'm doing it sometimes in ad-hoc queries).

There's a more extreme scenario in which distinct subqueries are written for each or a set of ORs (e.g. the distinction between open vs closed vs. invoices orders), which can make sense sometimes (e.g. the logic is completely different). Therefore, an antipattern can be dependent also of the context or use case. 

(5) use of JOIN instead of EXISTS

When there are no values returned from the subquery, quite often it makes sense to the EXISTS or not EXISTS operators in the queries (see the rewritten query below). This might not be indicated however for distributed environments like serverless SQL pool in which the distribution of the processing across multiple tasks might benefit when the pieces of the logic distributed don't require heavy reshuffles. 

(6) Using too Many Columns in ORDER BY

The columns specified in an ORDER BY clause need to make sense, otherwise they just add extra burden on the database engine, at least from the perspective of the checks that need to be performed. In the above query, at least the Name doesn't make sense.

It helps also if the columns can use existing indexes, though this depends also on query specifics. 

Another antipattern scenario not exemplified above is the use of ordinals to refer to the columns, which should be avoided in production environments (because the order of the columns can be changed accidentally or even :

-- using ordinals instead of number columns (not recommended)
SELECT ITM.ProductId                                  
, ITM.ProductNumber
, ITM.Name 
, ITM.Color 
, ITM.Style 
, ITM.Size
FROM Production.Product ITM                           
ORDER BY 2, 4, 5, 6

(7) Use Columns Not  Needed in Output

Besides the fact that each column included unnecessarily in the query can increase the size of the data processed (unless the database engine is smart to remove them), there can be also performance issues and/or optimizations involved. For example, if all the other columns are part of a covering index, the database engine might opt for a suboptimal index compared to the case in which the unnecessary columns are removed. 

Conversely, some columns are helpful to troubleshoot the logic (and that's why the Source column was considered) even if they aren't considered in the final output or the logic. It doesn't make sense to bring the query version with the respective fields into production, even if this would mean to have maybe a second version of the query used only for troubleshooting needs. Commenting the unnecessary columns could be a better choice, even if it's not recommended in general as too many such comments can obfuscate the case. 

Rewriting the Query

With the above input the query can be rewritten as follows:

-- Products in open orders (modified query)
SELECT ITM.ProductId                                  
, ITM.ProductNumber
, ITM.Name 
, ITM.Color 
, ITM.Style 
, ITM.Size
FROM Production.Product ITM
WHERE EXISTS (										            
	-- Open Purchase orders 
	SELECT POL.ProductId
	FROM Purchasing.PurchaseOrderDetail POL                 
	     JOIN Purchasing.PurchaseOrderHeader POH      
		  ON POL.PurchaseOrderID = POH.PurchaseOrderID
	WHERE POH.Status = 1 
	  AND ITM.ProductID = POL.ProductID
	)
 OR EXISTS (				                                    
	-- Open Sales orders 
	SELECT SOL.ProductId
	FROM Sales.SalesOrderDetail SOL
	     JOIN Sales.SalesOrderHeader SOH
		  ON SOL.SalesOrderID = SOH.SalesOrderID
	WHERE SOH.Status IN (1, 2)
	  AND ITM.ProductID = SOL.ProductID
	)	                           
ORDER BY ITM.ProductNumber			                           
, ITM.Color 
, ITM.Style 
, ITM.Size

Please note that in general to each rule there are also exceptions which should be considered against one's common sense. When the benefit of addressing an antipattern is neglectable compared with the effort involved and the logic doesn't create any issues, probably it's better to let the code as it. One can still reconsider the antipatterns later with the next refactoring opportunity. 

There are zealous data professionals who treat minor inconveniences (e.g. not using upper case for SQL reserved words, alternate code formatting, alternative writing of words, especially function names, different indentation, the use of "--" for commenting within a query, etc.) as antipatterns. Use your common sense and evaluate the effort against the benefits or risks and, not less important, be patient with others' mistakes!

Happy coding!

🧭Business Intelligence: Perspectives (Part X: The Top 5 Pains of a BI/Analytics Manager)

Business Intelligence Series

1) Business Strategy

A business strategy is supposed to define an organization's mission, vision, values, direction, purpose, goals, objectives, respectively the roadmap, alternatives, capabilities considered to achieve them. All this information is needed by the BI manager to sketch the BI strategy needed to support the business strategy. 

Without them, the BI manager must extrapolate, and one thing is to base one's decisions on a clearly stated and communicated business strategy, and another thing to work with vague declarations full of uncertainty. In the latter sense, it's like attempting to build castles into thin air and expecting to have a solid foundation. It may work as many BI requirements are common across organizations, but it can also become a disaster. 

2) BI/Data Strategy

Organizations usually differentiate between the BI and the data Strategy because different driving forces and needs are involved, even if there are common goals, needs and opportunities that must be considered from both perspectives. When there's no data strategy available, the BI manager is either forced to address thus many data-related topics (e.g. data culture, data quality, metadata management, data governance), or ignore them with all consequences deriving from this. 

A BI strategy is an extension of the business, data and IT strategies into the BI knowledge areas. Unfortunately, few organizations give it the required attention. Besides the fact that the BI strategy breaks down the business strategy from its perspective, it also adds its own goals and objectives which are ideally aligned with the ones from the other strategies. 

3) Data Culture

Data culture is "the collective beliefs, values, behaviors, and practices of an organization’s employees in harnessing the value of data for decision-making, operations, or insight". Therefore, data culture is an enabler which, when the many aspects are addressed adequately, can have a multiplier effect for the BI strategy and its execution. Conversely, when basic data culture assumptions and requirements aren't addressed, the interrelated issues resulting from this can prove to be a barrier for the BI projects, operations and strategy. 

As mentioned before, an organization’s (data) culture is created, managed, nourished, and destroyed through leadership. If the other leaders aren't playing along, each challenge related to data culture and BI will become a concern for the BI manager.

4) Managing Expectations 

A business has great expectations from the investment in its BI infrastructure, especially when the vendors promise competitive advantage, real-time access to data and insights, self-service capabilities, etc. Even if these promises are achievable, they represent a potential that needs to be harnessed and there are several premises that need to be addressed continuously. 

Some BI strategies and/or projects address these expectations from the beginning, though there are many organizations that ignore or don't give them the required importance. Unfortunately, these expectations (re)surface when people start using the infrastructure and this can easily become an acceptance issue. It's the BI manager's responsibility to ensure expectations are managed accordingly.

5) Building the Right BI Architecture

For the BI architecture the main driving forces are the shifts in technologies from single servers to distributed environments, from relational tables and data warehouses to delta tables and delta lakes built with the data mesh's principles and product-orientation in mind, which increase the overall complexity considerably. Vendors and data professionals' vision of how the architectures of the future will look like still has major milestones and challenges to surpass. 

Therefore, organizations are forced to explore the new architectures and the opportunities they bring, however this involves a considerable effort, skilled resources, and more iterations. Conversely, ignoring these trends might prove to be an opportunity lost and eventually duplicated effort on the long term.

Previous Post <<||>> Next Post

📊R Language: Data Transformations (Part I: Temperatures' comparison between F° and C°)

The time series used for weather analysis use either Fahrenheit (F°) or Celsius (C°) for the temperature values. Looking at the A and B plots below that represent the values of the same dataset in F°, respectively C°, there seems to be no difference between the two plots independently on whether one works with F° or C°, however the scales are different. Once one uses the same scale for both values (see C) the plots are distorted according to the formula used for transformation.

Comments:
(1) Typically, it makes sense to adapt the temperature scale to the audience, though on the Web there will be always a mix of audiences (and that's why weather websites allow to choose one of the values). 
(2) Not starting from 0 might show in the end the same trend at same scale, though the behavior can change occasionally. As long as the Y-axis is correctly labeled, this shouldn't be a problem. Conversely, it's better to control the scale and provide the min-max values for the axis accordingly.
(3) When creating such plots, it's important to be aware of the distortion that might be introduced by transformations. For linear transformations of the type a*x+b, the value of the "a" coefficient tells how much the resulting values are stretched or contracted.

I used as exemplification the airquality dataset which contains data for 1973, the temperature being given in F°. Unfortunately, the dataset contains only the day and the month, so the date must be constructed and added to the dataset. For simplification, I've added the calculated temperature in C° as column as well:

#reviewing the data
help("airquality")

#preparing the data
head(airquality)
airquality$date <- with(airquality, as.Date(ISOdate(1973, Month, Day))) #adding the date
airquality$TempC <- with(airquality, (Temp - 32) * 5/9) #adding the temperature in C°
head(airquality)

And, here's the code used to generate the plots:

#Temperatures' comparison between F° and C°
par(mfrow = c(2,2)) #1x2 matrix display

plot(airquality$date, airquality$Temp, ylab="Temperature (F°)", xlab="date", type="l", col="blue", main="A")

plot(airquality$date, airquality$TempC, ylab="Temperature (C°)", xlab="date", type="l", col="brown", main="B")

plot(airquality$date, airquality$Temp, ylab="Temperature (F°) vs (C°)", xlab="date", ylim=c(0,100), type="l", col="blue", main="C")
lines(airquality$date, airquality$TempC, col="brown")

# using inline formula
plot(airquality$date, (airquality$Temp - 32) * 5/9, ylab="(Temp-32)*5/9", xlab="date", ylim=c(0,100), type="l", col="brown", main="D")

mtext("© sql-troubles@blogspot.com @sql_troubles, 2024", side = 1, line = 4, adj = 1, col = "dodgerblue4", cex = .7)
title("Temperatures' comparison between F° and C°", line = -1, outer = TRUE)

In the fourth plot I directly used the formula for transforming the values from F° and C°. If the values based on the formula need to be used repeatedly, it's probably better to add a column to the dataset.

Unfortunately, the standard library has its limitations when creating visualizations. While writing this post I tried to work also with the plotly library, which offers a richer set of tools and can be used to create wonderful visualizations (though it proves also more complex to use). 

install.packages("plotly")
library("plotly")

Here's the code used to plot the below graphic (the points have labels, much like in Power BI):

fig <- plot_ly(airquality, type = 'scatter', mode = 'lines+markers')%>%
  add_trace(x = ~date, y = ~Temp, name = 'Temp (F)')%>%
  add_trace(x = ~date, y = ~TempC, name = 'Temp (C)')%>%
  layout(showlegend = F, title="Temperatures' comparison between K° and C°")

fig

The temperatures via Plotly

Happy coding!

🏭🗒️Microsoft Fabric: Polaris [Notes]

Disclaimer: This is work in progress intended to consolidate information from various sources and may deviate from them. Please consult the sources for the exact content!

Last updated: 31-Mar-2024

[Microsoft Fabric] Polaris

• {definition} cloud-native analytical query engine over the data lake that follows a stateless micro-service architecture and is designed to execute queries in a scalable, dynamic and fault-tolerant way [1], [2]
• the engine behind the serverless SQL pool [1] and Microsoft Fabric [2]
• petabyte-scale execution [1]
• highly-available micro-service architecture
• data and query processing is packaged into units (aka tasks) [1]
• can be readily moved across compute nodes and re-started at the task level [1]
• can run directly over data in HDFS and in managed transactional stores [1]
• [Azure Synapse] designed initially to execute read-only queries [1]
• ⇐ the architecture behind serverless SQL pool
• uses a completely new scale-out framework based on a distributed SQL Server query engine [1]
• fully compatible with T-SQL
• leverages SQL Server single-node runtime and QO [1]
• [Microsoft Fabric] extended with a complete transaction manager that executes general CRUD transactions [2]
• incl. updates, deletes and bulk loads [2]
• based on [delta tables] and [delta lake]
• the delta lake supports currently only transactions within one table [4]
• ⇐ the architecture behind lakehouses
• {goal} converge DWH and big data workloads [1]
• the query engine scales-out for relational data and heterogeneous datasets stored in DFSs[1]
• needs a clean abstraction over the underlying data type and format, capturing just what’s needed for efficiently parallelizing data processing
• {goal} separate compute and state for cloud-native execution [1]
• all services within a pool are stateless
• data is stored durably in remote storage and is abstracted via data cells [1]
• ⇐ data is naturally decoupled from compute nodes
• the metadata and transactional log state is off-loaded to centralized services [[1]
• multiple compute pools can transactionally access the same logical database [1]
• {goal} cloud-first [2]
• {benefit} leverages elasticity
• transactions need to be resilient to node failures on dynamically changing topologies [2]
•  ⇒ the storage engine disaggregates the source of truth for execution state (including data, metadata and transactional state) from compute nodes [2]
• must ensure disaggregation of metadata and transactional state from compute nodes [2]
• ⇐ to ensure that the life span of a transaction is resilient to changes in the backend compute topology [2]
• ⇐ can change dynamically to take advantage of the elastic nature of the cloud or to handle node failures [2]
• {goal} use optimized native columnar, immutable and open storage format [2]
• uses delta format 
• ⇐ optimized to handle read-heavy workloads with low contention [2] 
• {goal} leverage the full potential of vectorized query processing for SQL [2]
• {goal} support zero-copy data sharing with other services in the lake [2]
• {goal} support read-heavy workloads with low contention [2]
• {goal} support lineage-based features [2]
• by taking advantage of delta table capabilities 
• {goal} provide full SQL SI transactional support [2]
• {benefit} all traditional DWH requirements are met [2]
• incl. multi-table and multi-statement transactions [2]
• ⇐ Polaris is the only system that supports this [2]
• the design is optimized for analytics, specifically read- and insert-intensive workloads [2]
• mixes of transactions are supported as well
• {objective} no cross-component state sharing [2] 
• {principle} encapsulation of state within each component to avoid sharing state across nodes [2]
• SI and the isolation of state across components allows to execute transactions as if they were queries [2]
• ⇒ makes read and write transactions indistinguishable [2]
• ⇒ allows to fully leverage its optimized distributed execution framework [2]
• {objective} support snapshot Isolation (SI) semantics [2]
• implemented over versioned data
• allows reads (R) and writes (W) to proceed concurrently over their own data snapshot 
• R/W never conflict, and W/W of active transactions only conflict if they modify the same data [2] 
• ⇐ all W transactions are serializable, leading to a serial schedule in increasing order of log record IDs [4]
• follows from the commit protocol for write transactions, where only one transaction can write the record with each record ID [4]
• ⇐  R transactions at the snapshot isolation level create no contention
•  ⇒  any number of R transactions can run concurrently [4]
• the immutable data representation in LSTs allows dealing with failures by simply discarding data and metadata files that represent uncommitted changes [2]
• similar to how temporary tables are discarded during query processing failures [2]
• {feature} resize live workloads [1]
• scales resources with the workloads automatically
• {feature} deliver predictable performance at scale [1]
• scales computational resources based on workloads' needs
• {feature} efficiently handle both relational and unstructured data [1]
• {feature} flexible, fine-grained task monitoring
• a task is the finest grain of execution 
• {feature} global resource-aware scheduling
• enables much better resource utilization and concurrency than traditional DWHs
• capable of handling partial query restarts
• maintains a global view of multiple queries
• it is planned to build on this a global view with autonomous workload management features
• {feature} multi-layered data caching model
• leverages 
• SQL Server buffer pools for cashing columnar data
• SSD caching
• the delta table and its log are are immutable, they can be safely cached on cluster nodes [4]
• {feature} tracks data lineage natively
• the transaction log can also be used to audit logging based on the commit Info records [4]
• {feature} versioning
• maintain all versions as data is updated [1]
• {feature} time-travel
• {benefit} allows users query point-in-time snapshots
• {benefit)} allows to roll back erroneous updates to the data.
• {feature} table cloning
• {benefit} allows to create a point-in-time snapshot of the data based on its metadata
• {concept} state 
• allows to drive the end-to-end life cycle of a SQL statement with transactional guarantees and top tier performance [1]
• comprised of 
• cache
• metadata
• transaction logs
• data
• [on-premises architecture] all state is in the compute layer
• relies on small, highly stable and homogenous clusters with dedicated hardware for Tier-1 performance
• {downside} expensive
• {downside} hard to maintain
• {downside} limited scalability
• cluster capacity is bounded by machine sizes because of the fixed topology
• {concept}[stateful architecture] 
• the state of inflight transactions is stored in the compute node and is not hardened into persistent storage until the transaction commits [1]
• ⇒ when a compute node fails, the state of non-committed transactions is lost [1] 
•  ⇒ the in-flight transactions fail as well [1]
• often also couples metadata describing data distributions and mappings to compute nodes [1] 
• ⇒ a compute node effectively owns responsibility for processing a subset of the data [1] 
• its ownership cannot be transferred without a cluster restart [1]
• {downside} resilience to compute node failure and elastic assignment of data to compute are not possible [1]
• {concept} [stateless compute architecture] 
• requires that compute nodes hold no state information [1]
• ⇒ all data, transactional logs and metadata need to be externalized [1]
• {benefit} allows applications to 
• partially restart the execution of queries in the event of compute node failures [1] 
• adapt to online changes of the cluster topology without failing in-flight transactions [1] 
• caches need to be as close to the compute as possible [1] 
• since they can be lazily reconstructed from persisted data they don’t necessarily need to be decoupled from compute [1] 
• the coupling of caches and compute does not make the architecture stateful [1] 
• {concept} [cloud] decoupling of compute and storage
• provides more flexible resource scaling
• the 2 layers can scale up and down independently adapting to user needs [1] 
• customers pay for the compute needed to query a working subset of the data [1] 
• is not the same as decoupling compute and state [1] 
• if any of the remaining state held in compute cannot be reconstructed from external services, then compute remains stateful [1] 

Previous Post <<||>>  Next Post

Acronyms:

ADLS - Azure Data Lake Storage

CRUD - Create, Read, Update, Delete

DCP - distributed computation platform 

DFS - Distributed File System

DWH - data warehouse

HDFS - Hadoop DFS

SI - Semantic Isolation 

SSD - Solid-State Drive

References:
[1] Josep Aguilar-Saborit et al (2020) POLARIS: The Distributed SQL Engine in Azure Synapse, Proceedings of the VLDB Endowment PVLDB 13(12)  (link)
[2] Josep Aguilar-Saborit et al (2024), Extending Polaris to Support Transactions (link)
[3] Advancing Analytics (2021) Azure Synapse Analytics - Polaris Whitepaper Deep-Dive (link)
[4] Michael Armbrust et al (2020) Delta Lake: High-Performance ACID Table Storage over Cloud Object Stores, Proceedings of the VLDB Endowment 13(12) (link)