Voice of the DBA

Author: way0utwest

  • The London Redgate Summit

    In a week I’m heading to London for the Redgate Summit. I enjoyed these last year and had some very interesting conversations with customers, prospects, and a few Redgate fans.

    This year the event is on Mar 11 at the Tottenham Hotspur Stadium, which should be fun given that I’ve been re-watching Ted Lasso (I still need to make a pilgrimage to Richmond…).

    The schedule has a variety of tracks for leadership and tech people. There are even some breakfast sessions if you want to learn more about Oracle or PostgreSQL. I’ve got the keynote session and a panel for leadership on my agenda.

    Check out the video from last year, and register to come this year. I’ll see you in London next week.

  • Why Not Use AI?

    Erin Stellato, a program manager at Microsoft, asked a very interesting question on LinkedIn: “why do you *not* want a Copilot in SSMS?”

    That got me to stop and think a minute. Why don’t we want to use an AI to help us? It’s a good question, given the hype and (maybe) potential of the technology. While it might not help you now, or with your specific thing, it might help others, so are there good reasons not to use AI technologies, like the GenAI LLMs?

    Let me know today. What are you reasons for not using them (apart from cost or restrictions at work)?

    I will say that I rarely use AI tech to get work done, or even in my home life. I just don’t enough places where I think I need help outside of a search. I search a lot, but I haven’t found AI to be faster than searching. Sometimes it is, but sometimes not and the unreliability of the tech bothers me. That being said, I do have a tag where I write about the AI things with which I am experimenting.

    I think for many of us, AI is too amorphous and unclear. We don’t quite know when it helps or doesn’t. We don’t know how to judge the quality of it. It’s also not a habit or integrated into life. Most of the suggestions I see in various places aren’t things I’d ask. The default page in my local mode has these suggestions:

    • tell me a fun fact
    • show me a code snippet of a website header
    • give me ideas for what to do with kids’ art
    • help me study for a college entrance exam

    That last one might be the only question I ask, not for college, but maybe for something like an MS exam. However, is the AI better than a search? Maybe, but I’d have to try it. Right now I’d be tempted to just search for a list of things to study. Maybe the AI helps me find those better, but really I’d want it to quiz me on different things.

    The problem is I don’t trust it to ask good, relevant questions, or necessarily give me the right answers if I asked it to quiz me.

    I guess ultimately is I don’t have enough “I think an AI is better” reasons over the do what I normally do. Without substantial evidence AI is better, I don’t use it much. Do you feel the same way?

    Steve Jones

    Listen to the podcast at Libsyn, Spotify, or iTunes.

    Note, podcasts are only available for a limited time online.

  • T-SQL Tuesday #183: Improving Permission Management

    This is my (late) answer to my own invitation for T-SQL Tuesday #183. I was very busy a few weeks ago when the invite when out (glad it was scheduled) and I never got this done. This post looks at how I managed permissions in the past.

    Large Enterprises

    Many of us would like to think that large enterprises have standards and they’ve learned about best practices. My experiences in 3 of them were that so often large enterprises were small ones that grew, often with lots of tech debt and busy staff. Even when there are limitations to ensuring good security, often we can’t just fix things because we might break something.

    In one large company (5000+ employees) I found that many of the database servers I managed had permissions set in a variety of manners. Often this included lots of individual permissions granted to logins and users, which was a mess.

    Even when AD groups were in place for departments, they weren’t used as logins in SQL Server since not everyone in a department needed to access the database.

    Cleaning Up

    I hated managing permissions by user. When I started and got a ticket to add a new person to a database for access, I watched someone train me by looking up another similar user in the database, scripting out their permissions, and then search/replace to change the user or login name.

    Not a bad solution, but one that doesn’t scale well over time. What I started to implement was to create an AD group (for Windows users) or a database role (for SQL Users) and then add the login/user to this group or role and assign the permissions from my “other” user to the role. I’d also often move the user to the new role and revoke their specific permissions.

    While this seemed like more work at first, this quickly started to scale well as we had a way to add new users to roles that matched the access they needed.

    Slowly over time we moved a lot of access to AD, which allowed us to remove the burden of disabling users in SQL Server. Plus, we could easily see which access a variety of users had by looking at a role rather than checking multiple accounts. Auditors liked that and it helped us pass various audit checks over time.

    Use roles and groups. They’re not hard and they make things cleaner and easier over time.

  • Trust is a Funny Thing

    I caught an article on AI skepticism and there was a point in the article where trust was mentioned. Specifically the reasons that people distrust an AI or tech tool is that it makes a mistake, so they stop using it. A few examples of this were using a writing AI that made a grammar mistake or a GPS routing device that added a wrong detour. In those cases the humans stopped using the assistance of the algorithm because they felt it wasn’t trustworthy.

    What’s fascinating to me is that I had this same conversation with a human the day before. Someone mentioned they were working with a group and they misstated something. After that, the group stopped listening to all this person’s advice, thinking it was all suspect. Essentially one mistake overrides everything else.

    Trust is a funny thing. It takes time to earn and gain trust from others, and it can be lost quickly with an action. Trust is a reliance, a belief, a confidence in something or someone. The way the GenAI is framed, and the way a system is presented to a user can have a huge impact on the level of trust we give to an GenAI-system. Tell someone the system is trusted or others are using it raises trust. Tell someone to be wary, and they’ll trust the system less.

    I find that GenAIs are unreliable but not completely. They are trustworthy in some ways, but not in others. There are a few variations in the article (and in other places) that note that GenAIs are helpful, but we need to review their work. Essentially trust but verify, which isn’t a bad policy for any interaction with humans or AI systems. We learn to trust our co-workers, but that takes time. Some we might never trust and always review their work, while others we accept what they give us with minimal review.

    It’s interesting to see such a wide variety of responses to the usefulness of GenAI systems. Some people find them amazing, some find them useless, some use them daily for small things, some never bother to submit a prompt. I find myself cautiously using them more and more, especially my local models, trying out different prompts in different areas. One thing I got out of the article is I ought to have more conversations with the GenAI, much as I would with a co-worker. I’ll try to do that more in the future and see how it goes…

    In the meantime, how much do you trust AIs, other computer systems, or coworkers?

    Steve Jones

    Listen to the podcast at Libsyn, Spotify, or iTunes.

    Note, podcasts are only available for a limited time online.