Shop Talk: 2020-08-24

The Recording

The Panelists

  • Kevin Feasel
  • Mala Mahadevan
  • Tracy Boggiano

Notes: Questions and Topics

Working with ORMs

Our primary topic for the evening was working with ORMs. We’re data platform specialists, so the feeling is generally fairly negative toward ORMs, but I tried to give the “pro” side a reasonable airing, as there certainly are valuable uses for them.

One of the key things about ORMs to consider is that there are two major varieties: lightweight ORMs (also known as micro-ORMs) and their heavyweight bretheren. Two examples of micro-ORMs that I personally like are Dapper and FSharp.Data.SqlClient. These are small wrappers around ADO.NET which create simple objects from T-SQL statements and stored procedures. They save the development time of mapping result set outputs to .NET classes / record types without adding a lot of overhead. For that reason, I’m a big fan of using them.

On the other side, we have heavyweight ORMs like Entity Framework and NHibernate. These do a lot more and aim to create a single development experience in C#, so that developers don’t have to think in two languages. They also work well with fluent APIs like LINQ, translating those statements into SQL queries.

As far as performance goes, micro-ORMs are faster in most cases. Products like EF and NHibernate can generate some really nasty SQL and cause performance problems on complicated queries. But if you stick to fairly simple queries—especially simple insert, update, and delete operations—heavyweight ORMs can save you a good bit of time.

Tips for Creating a Presentation

The secondary topic for this evening was tips for creating a presentation. Mala, Tracy, and I have all put together presentations and so have several of the great people in chat. We talked about some ideas on how to get into presenting and shared a few stories of things that go wrong, with the expectation that hey, stuff happens but the presentation still works out in the end.

Mala’s Book Corner

Mala has two book recommendations for us:

Shop Talk: 2020-08-17

The Recording

The Panelists

  • Kevin Feasel
  • Mala Mahadevan
  • Tracy Boggiano

Notes: Questions and Topics

Thoughts on PASS Pro Membership

Mala, Tracy, and I spent a good bit of time talking about PASS’s professional tier membership. It’s pretty common for trade organizations to require member dues—I brought up as examples the ACM and IEEE, and Tracy & Mala came up with other examples like Oracle’s official user groups. In this case, there’s no requirement: the free tier of PASS membership is the same as before.

We talked for a while about the question of whether to become a pro member. The really short version is that if you’re taking a hard look at membership benefits versus the annual cost, it’s not worth it today. But if you have goodwill built up for PASS, it’s worth it, as it helps keep PASS afloat.

Architectural Diagrams

The other key theme of the evening was architectural diagrams. Mala and Raymond both shared the same article on what makes for a good diagram, so that seems like a sign that it’s a good article.

My key judgments on what makes for a good architectural diagram are:

  • The diagram is concise. Show what you need but don’t include a lot of unnecessary detail. For example, if we’re talking about a ETL process in Azure, such a diagram might show a virtual machine pushing data through Azure Data Factory into an Azure Synapse Analytics SQL Pool, and from there into Azure Analysis Services and Power BI. At this high level, including the list of specific VNet settings is unnecessary. Even the set of data flows you’re creating through this process is unnecessary unless there’s a need—for example, if there are two processes and you need to differentiate them.
  • At the appropriate level. Ideally, boxes or shapes in a diagram should be independent units, each of which is necessary to understand for the solution.
  • Built with the audience in mind. We can have multiple diagrams for different people, and diagrams aren’t (or at least shouldn’t be!) the only documentation available.

As far as diagramming tools, I brought up Diagrams.net and Diagrams as Code. We also received recommendations for PowerPoint, Visio, and LucidChart.

Mala’s Book Corner

Mala has two book recommendations for us:

Jobs

Shop Talk: 2020-08-10

The Recording

The Panelists

  • Kevin Feasel
  • Mala Mahadevan
  • Tracy Boggiano
  • Tom Norman

Notes: Questions and Topics

Transaction Modes

The first topic of the evening was around transaction modes, following from a blog post and video I released yesterday.

Not surprisingly, most of us use explicit transactions in many cases, particularly higher-risk scenarios. Tracy has a template which she fills in, and Mala follows a similar path to me: begin a transaction if it’s potentially scary or if you’re doing this in application procedures. Tom’s answer: it depends. Most of the time, Tom uses autocommit because he’s in easy mode, but when he kicks it up to dangerous mode or if he needs to wrap multiple tables in a single transaction, he’ll mark them as explicit.

We also got into a rat’s nest on nested transactions and save points. Nested transactions are a lie. Savepoints are mostly a lie.

Maximum Number of Joins in a Query

Mala brought up a topic asking about the maximum number of joins in a query. Tom took a strong stand with “It depends.” Mostly it depends on how big the tables are.

@srutzky had the best answer: 20 to 30, but they must all be to views which have 20-30 joins. I’d add on that the views need to be nested views going at least three or four levels deep.

I don’t think there’s a real answer to the question. I’ve run into cases with 18-20 joins where the query plan just falls apart and taking one of those joins out (even to a minor table where it’s a simple nested loop lookup of a fairly small number of rows) makes the query perform a lot faster. But I’ve also worked with queries with more joins than that which worked quite nicely. At the end of the day, if you are able to generate a stable plan, that’s how many joins you can get away with.

As a bonus, I rant about the phrase “Normalize until it hurts, denormalize until it works.” This isn’t the 1980s; that phrase generally doesn’t apply to OLTP systems anymore and hasn’t for a good decade-plus. If you need to denormalize your tables to get queries to run efficiently, it probably wasn’t really normalized.

Mala’s Book Corner

Mala is back with two book recommendations for us:

A Diversion on Technical Writing

We ended the broadcast with a discussion on the importance of technical writing and some of the difficulties around it. It started on the idea of writing a book, but we ended up focusing on the documentation itself. One thing I want to stress is just how difficult it is to get this documentation right, especially because we tend to take mental shortcuts and expect that others will know the context currently in our heads. I’m really bad at it and have to try hard to remember that the reader needs all of the relevant context. It’s particularly difficult because the reader will go from A to B to C to D, but I might have written it B, D, A, C, such that by the time I get to A, I forget that I needed to explain something to make B make sense.

Raymond also asked where you can store documents. It’s a tough problem and we punt around the problem a bit.

Shop Talk: 2020-08-03

The Recording

The Panelists

  • Kevin Feasel
  • Tom Norman

Notes: Questions and Topics

Tracy the MVP

Congratulations to Tracy Boggiano for finally getting her MVP. It’s been a long time coming and I’m glad that she is getting the appropriate recognition for her community support.

Azure Data Studio Database Projects

Tom gives us his first thoughts on Azure Data Studio database projects. Tom referenced a blog post by Wolfgang Strasser and walked us through his first thoughts.

We then turned this into an extended discussion on the state of Azure Data Studio today.

A Rant on the XEvent Profiler

Tom and I then discussed Profiler. No, not that profiler; the other one. I hate the fact that Microsoft named this the XEvent Profiler because it really muddies the waters. The product itself is fine and is starting to give Extended Events a reasonable UI. But that name…

The biggest problem I have with the name is that it seems to be intentionally confusing, and as long as there are two tools called Profiler, that ambiguity will lead to confusion. “Oh, I heard from <insert name here> that Profiler is bad, so I’ll avoid this thing called XE Profiler. What’s an XE?” It would have been better to name it something different and make it easier for people to say something like “Avoid Profiler and use the SQL Server Performance Tracker instead.”

The product is fine; the name is not.

Licensing

We had several questions around licensing, and I’m bundling them here.

First, if you have questions about SQL Server licensing, Thomas Grohser did a talk for our group last month and he explains it better than I ever will.

We also talked about licensing in tools like Visual Studio Code, which has its own license based on MIT. We talked a bit about which licenses tend to pass muster in legal teams at organizations, as well as some of the ones which don’t.

I also talked about why I hate Oracle and the exact amount of my Oracle razzing which is real versus me being a troll.

Mala’s Book Corner, by Kevin

In this week’s edition of Mala’s Book Corner Kevin’s Book Hovel, I recommended Spark in Action 2nd Edition by Jean-Georges Perrin. Jean-Georges is local to us in the Triangle area and published a great book on Spark. The examples are a lot better than what I’ve seen in other Spark books and training materials, so if you’re interested in learning about Spark, get this book.

Shop Talk: 2020-07-27

The Recording

The Panelists

  • Kevin Feasel
  • Mala Mahadevan
  • Tracy Boggiano
  • Tom Norman

Notes: Questions and Topics

PASS Summit Speakers

I wanted to call out that there are nine separate North Carolinian speakers for PASS Summit this year.

On Shop Talk, Tracy Boggiano, Tom Norman, and I will all be speaking at PASS Summit. Other speakers include friends of the show Chris Voss and Jared Poche.

Angela Henry heads up the group from the Triad area.

Melissa Coates (whose pre-con is on a different day from mine, so you can see both of ours!), Brad Llewellyn, and Ben DeBow round things out from the Charlotte area.

Dealing with XML and JSON

@iconpro555555 asked a question at the very end of last week’s Shop Talk around handling structured data—that is, XML and JSON—in SQL Server.

SQL Server has a long history of dealing with XML data, including an XML data type, XML-specific indexes, and XPath query notation. It’s notorious among DBAs as “Something we don’t want,” and Tracy points out the biggest problem: once you start to get a decent amount of data, everything falls apart.

In SQL Server 2016, we got JSON support through the OPENJSON, FOR JSON, and JSONVALUE commands. There is no separate JSON datatype or direct indexing, though if you want to, you could create a persistent computed column using JSONVALUE to extract a specific value from the JSON and index that persistent computed column.

I don’t necessarily mind XML or JSON data in the database and it can be the best solution. But one pre-condition for it being the best solution is that you shouldn’t be shredding that data regularly. If you do, build out a proper, normalized data model and take advantage of what relational databases do best.

Storing and Using Secrets in Powershell

The next topic was around secrets in Powershell. This was a question from the very beginning of Shop Talk’s history, but because Rob Sewell recently blogged about this, I wanted to bring it up.

A Rant on Database Diagramming Tools

My official Rant of the Night was around how weak the database diagramming tool is in SQL Server Management Studio. The thing barely works at all, requires you to create database objects (meaning that developers with read-only permissions can’t even create these diagrams for themselves), provides only a modicum of relevant information, and are a royal pain to disentangle if you have more than just a couple of tables.

Tom defended the honor of SSMS database diagrams, while folks in chat recommended tools like DBeaver and DBDiagram. I brought up Mermaid, as I’ve just started getting into it for presentations and it allows for code-centric modeling, meaning that you write Python code to represent your model. I could see that tying in well with queries against system tables to get tables and columns, foreign keys, and the like. I also mentioned sqlDBm in an off-hand fashion because I couldn’t remember the name of it. I’ve used it before to reasonably good effect and the free version works pretty well if you have one fairly stable database. Visio database reverse engineering is excellent as well when your goal is to understand an existing, stable design.

Shop Talk: 2020-07-20

The Recording

The Panelists

  • Kevin Feasel
  • Tracy Boggiano

Notes: Questions and Topics

Azure and Lego Pieces

I got things going with a rant about how things in Azure feel like Lego pieces which don’t quite fit together. The first five hours of an Azure project generally involves me enjoying how well all of these services fit together. Then, you find a spot where things don’t quite work and then you spend 15 hours trying to bludgeon the thing into submission.

My example was around using Azure Synapse Analytics. I used a Spark pool to train a model, and the notebook worked great, saving my model to Data Lake Storage Gen2 and everything.

Then, I wanted to deploy the model using Azure ML. Now, Azure ML lets you deploy PySpark models, so that’s no problem. The problem was, the model is in HDFS (really Data Lake but let’s not split hairs too much given that Blob Storage is an implementation of WebHDFS) but Azure ML won’t let you use Data Lake Storage for the model location—it needs to be local to the machine. Which means local to the driver. Which is the opposite of what you want with Apache Spark.

The end result was that I could get the pieces to fit, but it involved a bit too much welding, filing, and sanding for my taste.

Dealing with a Production Outage

Tracy tells a story of teamwork during an outage. I’d link to it, but unfortunately she’s dealing with a blog outage.

How is the DBA Role Evolving?

Mala couldn’t make it tonight but she did give us a great topic: how is the DBA role evolving?

I took it three ways, noting that this is focused on “production DBAs” rather than database developers:

  1. Is the DBA role evolving? My short answer is, I can argue either way. On the “yes” side, DBAs are expected to have more development skills than they had before. We want them to know about source control, Powershell, and DevOps. They also, in cloud environments, have fewer knobs to turn and lose ultimate control over the machine. But on the “no” side, the key set of DBA roles hasn’t really changed that much over the past several decades. Think about taking backups and testing restoration, ensuring high availability and/or disaster recovery, preventing or correcting corruption, tuning performance, and monitoring systems. The means have changed but the motivations and goals are the same.
  2. Can a DBA from 1985, 1995, or 2005 be teleported into today’s world and survive? Assume that there’s some time for learning about technological changes. Well, take a look at DBAs around you. Many of them got their starts in the 1990s, so the answer is probably “yes.” Now, the person from 1985 or 1995 would need to unlearn quite a bit and learn quite a bit, but I don’t think this is some crazy sci-fi future scenario where a DBA from then gets dropped in and has absolutely no clue how to be a DBA today.
  3. Will there be DBAs 10 years from now? 20 years from now? I think the answer is absolutely yes, because the things which define a DBA’s role don’t come about because DBAs wormed their way into organizations. They are critical for the business: we have data, data which the company finds valuable. As a result, we need to make sure that the data is available, that it won’t disappear tomorrow, that it won’t become corrupt tomorrow, and that business users will be able to get to it in a reasonable amount of time. There will be people who specialize in this administration of the bases of data as long as people at businesses care about data.

Shop Talk: 2020-07-13

The Recording

The Panelists

  • Kevin Feasel
  • Tracy Boggiano
  • Tom Norman

Notes: Questions and Topics

Last night was the “Let’s get a copyright strike from the TV” edition of Shop Talk.

Migrating from Azure VMs to Azure SQL Database

Tom started us off with a discussion of how simple it was to migrate from SQL Server on a virtual machine hosted in Azure to running Azure SQL Database via Availability Groups. The one problem he ran into was around things like logins and SQL Agent jobs not going over to the AG secondaries because system databases can’t be part of an Availability Group. This feature was announced for SQL Server 2019, but had to be pulled, and as of CU5 at least, is not in the product.

Goodbye, Azure Notebooks

I reported the sad news that Azure Notebooks is going away on October 9th. There are several alternatives available. For training people, Azure Labs is probably the best option. For personal work, Visual Studio Code and Azure Data Studio offer notebook options, though I’d probably just use Jupyter Notebooks off of Anaconda and be done with it.

Visual Studio Codespaces does look pretty good as well and the pricing isn’t too bad. But none of these have what I really appreciated about Azure Notebooks: it being free.

Mala’s Book Corner, Hosted by Kevin

Without Mala here to join us, I decided to take on the mantle of book cornerist, where I recommended Learn Azure in a Month of Lunches by Iain Foulds. I’m going through this now, as it just released—I got my copy about a week ago. It looks like Microsoft is offering a free copy of the e-book as well if you provide them a bit of info.

Staying at a Job

My last question of the night was, how long should you stay at a job? In this case, I broke it down into two categories: what’s the minimum number of years you should stay at a job to avoid the “job-hopper” label, and what are the key conditions for leaving a job?

As far as years of service go, Tracy said 2-3 years. Anders mentioned that he starts looking at 2 years just to stay in tune with the market. When I started my career, 5+ years was the expectation, though I did get some pushback from folks in the audience and we settled on 5 years being a midwestern thing. Anyhow, I’d say that the number today is 2 years: we stay at a job that long and it’s “career break-even.”

This ties in with the other side, conditions for leaving. @thedukeny mentioned boredom and the quest for more money, and Anders echoed on the boredom comment. @rporrata mentioned three things: money, family, and (lack of) training. Tom mentioned the company not paying for training and also falling behind on technology. With Tracy, it’s terrible bosses and old systems.

I mentioned one other criterion: moving up. In the Olden Days, you had clearly defined levels where you could move up in a company, from Programmer 1 to Programmer 2 to 3 to 4 to Architect, and so on. Each role had specified tasks and commensurate pay. You could expect some level of growth at a company over the course of a couple decades. But that’s changed radically at most organizations. Now, you have a role: Programmer. In many places, there aren’t even labels for Junior or Senior—you’re just Programmer. This has nice benefits for employees in making it easier to pick up new problems to solve—you aren’t constrained to doing tasks A, B, and C because that’s all a Programmer 1 does. But it also means that your potential for growth is quite limited. There are no new titles to which you can aspire or big salary bumps which go along with it. And companies have this strange habit of assuming that an across-the-board 2% pay increase for employees is fine, regardless of how that person’s market value has changed.

As a result, people leave companies to grow. Each job change is an opportunity to get back up to your market salary level as well as take on more roles and assignments. There’s also much less stigma about people leaving and returning after a few years, as most people recognize that this isn’t the 1950s or the 1980s in terms of working at companies. There are some “lifer” companies out there with a fairly high percentage of extremely long-term employees, but those are rare.

Shop Talk: 2020-07-06

The Recording

The Panelists

  • Kevin Feasel
  • Tracy Boggiano
  • Mala Mahadevan
  • Special Guest Star Jared Poche

For show canon purposes, Tom was captured in a multi-state illegal fireworks ring, smuggling the good stuff.

Notes: Questions and Topics

Last night was the “Let’s get a copyright strike from the TV” edition of Shop Talk.

I Remember Halloween

Mala started us off with a question about Halloween protection. What is Halloween protection? How can we identify this? What are possible solutions?

Answer: Jared gave us a description of Halloween protection, and I followed up with a quick demo. If you want more thorough answers on Halloween protection and solutions to the performance issues it can provide, I recommend Paul White’s series on the topic, Jared’s blog post, and Mala passed along an article from Itzik Ben-Gan.

Upgrading SQL Server via Distributed Availability Groups

Tracy then told us about a project she had undertaken to upgrade SQL Server using Distributed Availability Groups, a feature introduced in SQL Server 2016 and enhanced with direct seeding in 2017. Tracy has used this technique to perform near-instant upgrades from 2016 to 2017, and points out that it can work for upgrading any instance from 2012 and on. Tracy shares a few important tips, such as changing the listener rather than upgrading connection strings, as that obviates the risk of missing some connection string somewhere.

Tracy should have a blog post forthcoming on the topic as well.

The Silliness of Security Compliance

In last night’s Rant of the Evening, I channeled Sean McCown in pointing out that there is nothing inherently wrong with enabling xp_cmdshell. If you’re not a sysadmin, you can’t use xp_cmdshell by default; if you are a sysadmin, you can enable xp_cmdshell whenever you want. So any security check telling you to disable xp_cmdshell is wrong—the correct answer is not to hand out xp_cmdshell rights willy-nilly.

All of this comes in the context of security compliance audits. It was Tracy’s topic and she did a good job airing out some of the sillier ones when I wasn’t cutting her off.

As a point of order, the concept of security is by no means silly. Having people who understand the behaviors (including unintended!) of first- and third-party applications and systems is critical, and I support those people how I can.

Mala’s Book Corner

Mala recommended two books for us this week:

Getting a Job

I wrapped things up with a question: how do you get a new job in adverse conditions? I also tried to tie it into the dot-com collapse, as that was the best analog I had for today’s environment.

Shop Talk: 2020-06-29

The Recording

The Panelists

  • Kevin Feasel
  • Tracy Boggiano
  • Tom Norman

Notes: Questions and Topics

I was a little ill last night when we recorded, so that’s why I made any mistakes I did. It was a 24-hour thing, though, so I’m going to have to come up with a new excuse for why I make so many mistakes.

Listing System Functions

Nadir hit me up via e-mail and wanted to figure out how to list system functions in T-SQL.

Answer: This script will give you the set of system functions available in SQL Server.

SELECT
       o.name,
       o.type,
       o.type_desc,
       m.definition
FROM sys.system_sql_modules m
       INNER JOIN sys.all_objects o ON m.object_id = o.object_id
WHERE
       o.type NOT IN ('P', 'V')
ORDER BY
       o.type ASC,
       o.name ASC;
GO

Interestingly, there are still some functions which don’t show up at all, including metadata functions. As a note, I mentioned @@SPID as a metadata function but that’s not correct—it’s a configuration function.

Preventing Table Changes

Tom dropped the next question on us: he wants to prevent anybody from modifying a particular table, like adding or removing columns. How can we notify or prevent this sort of modification?

Answer: There are a few techniques available. Tom mentioned using SQL Audit and creating a view with schemabinding against that table—that will prevent table modifications so long as the view is shcemabound.

I also mentioned DDL triggers, which you can use to roll back specific operations like modifications to a particular table. You can also use Policy-Based Management.

Tracy calls out preventing SELECT * operations against a table by adding a computed column which divides 1/0 to teach developers not to do that. @srutzky did mention a couple reasons not to do this, though

A Little Bit on Containers

From there, I started to bring up a question that Mala had asked around why you might want to use containers. We talked about using Docker containers for androgogical purposes, though Tom pushed Azure Lab Services, which is another good option. We also covered Docker containers for automated testing, where I think it makes a huge amount of sense.

I mentioned some bits about container insecurity and there are some practices for making these better.

During this, Tom also asked the question, will SQL Server be on Windows 202X, say 5 years from now? My answer is an adamant yes. As long as there is a Windows operating system, I’d expect to see SQL Server on Windows. There are too many features only available on Windows today, and even some new features have are only partially implemented in Linux.

SQL Server Authentication Options

@iconpro5555 asked, what are the different methods for connecting to SQL Server?

Answer: There are two techniques: Active Directory and SQL authentication. Within Active Directory, we can break it out a bit further. On-premises, we have Windows Integrated Authentication. When dealing with Azure Active Directory, you have three ways of connecting: via multi-factor authentication (MFA), by typing in a username and password, or integrated Azure Active Directory.

As a side note, I wish that we had all of those AAD options for on-prem Active Directory.

Apple on ARM, Continued

Tom brought up Apple moving over to ARM processors, especially now that we’ve heard a little bit more. I mentioned rumors that Apple turned away from Intel because the Skylake CPU series was bug-laden garbage. Also, we’re all very happy that AMD is there as a legitimate competitor, and I think that, combined with Apple’s decision, will force Intel to improve.

We talked a little bit about app support, as ARM-based Macs won’t support Boot Camp. But it looks like Microsoft 365 and Adobe Creative Cloud will support it.

Shop Talk: 2020-06-22

The Recording

The Panelists

  • Kevin Feasel
  • Tracy Boggiano
  • Mala Mahadevan
  • Tom Norman

Notes: Questions and Topics

Apple on ARM

@thedukeny starts us off by talking about rumors that Apple is moving away from Intel chips and toward ARM processors for its new line of MacBooks and asked for our thoughts. @rporrata follows up with questions about SQL Server on ARM.

Answer: Answering the second question first, Azure SQL Edge runs on 64-bit ARM processors and x64 processors. I’ve used it in private preview and liked what was available. You can also get into the public preview right now.

On the point about Apple, I can see it happening for their lower-end devices, as we’ve seen a cottage industry of ARM-based Chromebooks out there, so we know it’s viable. You wouldn’t use them for heavy gaming, video editing, and the like, but they’d definitely offer battery life improvements over the current generation if you’re a casual user of the product.

Partitioned Tables

Tom dropped the next question on us: he is looking to use table partitioning to improve performance where he only needs recent data. Will this technique help him out?

Answer: Mala chimed in with the best answer: maybe, but probably not. For a detailed answer, check out Kendra Little’s video. For a deep dive on partitioning itself, Andrew Pruski has a great presentation on the topic.

GOTO and the Modern Era

Mala made me defend goto. Her question: when and where would you expect to use goto statements in code? What was Dijkstra’s argument against them based upon?

Answer:  I enjoyed the dive into this topic, as it really tells a story of the history of computer science. If you want, you can easily read Dijkstra’s letter to the editor (and which title Dijkstra really didn’t like; a classic case of the editor getting to choose the title). But temper this with a quotation from Dijkstra a few years later:

Please don’t fall into the trap of believing that I am terribly dogmatical about [the go to statement]. I have the uncomfortable feeling that others are making a religion out of it, as if the conceptual problems of programming could be solved by a single trick, by a simple form of coding discipline!

Donald Knuth had a response a few years after the paper’s release which defended goto in specific circumstances.

Summarizing a couple of themes that I spent a lot of time on in the episode, it’s hard for those of us who never really worked with pre-structured programming languages to understand the debate. At the time Dijkstra was writing, common languages didn’t always have structure components like break, continue, return, switch, case, do, while, and sometimes not even else! In lieu of those structural keywords, programmers needed to use the tools available, and the biggest one was goto. Today, we throw the brakes when we see a single goto statement. But Dijkstra wasn’t really concerned about that; he was concerned about it being the blunt instrument programmers used even if there were better options available.

Nowadays, it seems like the generally accepted exceptions to “don’t use goto” are:

  • switch cases in languages like C#, where you can go to a different case in the switch.
  • Breaking out of deeply nested loops, though in that case the question might be, why are you nested so deeply?
  • Ensuring that all code paths reach a certain destination for cleanup steps in languages without finally. For example, this might involve freeing memory, closing connections, and releasing handles.

The reason I’m not automatically critical of GOTO in SQL Server is that there is no FINALLY block in TRY/CATCH. But then again, it’s really uncommon that you’d need that construct.

Mala’s Book Corner

Mala recommended two books for us this week:

I can second both of these recommendations, having the paper copy of Joe’s book and a PDF of Kalen’s. The link to Kalen’s book lets you download it as a PDF for free.

Second Thoughts on Azure ML

The final thing I’m covering here is some second thoughts on Azure Machine Learning. The brief version of it is as follows.

When it came out, Azure ML felt like SSIS for machine learning. You dragged and dropped items, clicked the mouse a whole bunch of times, and end up with a pretty-looking data flow to build a model. Data scientists tended to complain that they could do most of the work in Azure ML in about 6 lines of R or Python code, and that the visual interface made model comparison really clunky.

Later on, demos were still drag-and-drop, but I remember that instead of dragging and dropping various data cleanup components, they’d drop in an R/Python code block and put in those 6 lines of code. So that was better, but the visual interface was still too constraining and compute was rather expensive once you did the math.

I skipped Azure ML for a while (including the Studio phase but came back to it as the result of a work project and I have to say that it looks a lot better. The integration with Azure Container Instances and Azure Kubernetes Service is pretty nice, model registration competes with MLflow (in that it’s easier to maintain, though not as feature-rich), and I approve of AutoML for at least getting you 80% of the way there (though @thedukeny points out that in many cases, AutoML can do at least as well as a data science team). And pricing isn’t too bad—we have a moderately used web service (called approximately 16K times per day and pushes roughly 4 million rows) and would be out approximately $100 per month for 24/7 utilization of ACI. That’s a fair bit less than we’re paying now.

The bottom line is, if you ignored Azure ML over the past couple of years like I did, I recommend giving it another try to see if it might fit some of your needs.