Instead of working, I blog.

For those of you who follow my blog, you know that I've used LiteSpeed for many years now. And one of the things that's always bugged me is the lousy way they run their upgrades. To upgrade LS typically consists of having to do some version of manually deleting the current repository tables and then running the install program. And this is after uninstalling your previous version manually as well. And if you want to save your data in your repo tables you have to create new ones and copy the data over, or just rename the old tables. And it still isn't that easy because of the relationships. If you just rename the tables you have to rename the relationships too or the new install will fail. And you have to do this for every box you own. What a pain in the ass!!

Well, I have a problem with LS not backing up one of my DBs even though it's been working just fine for almost 2yrs. So I just upgraded from v.4.7 to the latest 4.8 (yes, I know v.5 is coming out soon, but I can't wait). And let me tell you this... it looks like they finally got these upgrade problems fixed. My upgrade went smooth. I didn't have to manually uninstall the old version, I didn't have to play my usual shell game with the repo tables, and I didn't have to burp the service. Life is good.

Now, I'll have to see if it actually fixed my problem. But I'm happy to be able to report something good about LS for the first time in a long time. They've mostly been adding features, and very slowly at that, and haven't been really concerned with our actual problems. Maybe this is a turning point.

Anyway, good job LS guys.

One thing that's as varied as ways to code is how to manage the code itself. Or I guess I should say architect instead of manage, but it all comes down to management.

The 2 major ways are to componentize and to not. And by componentize, I mean taking all the individual components and turning them into small independent chunks that can be called from many sources. A good example is taking a date calculator routine in your SPs and turning it into a function so the SPs just call it instead.
So those are the 2 ways. You either put it inline with the code or you make it into a component and call it from all the modules you need.

Of course, I'm mainly talking about SQL code here, but it really applies to any kind of code I suppose.

It's easy to see the advantages of developing a component solution. This is the DB equivalent of COM, right? But what's not easy to see is what it does for you (or to you) support-wise. What this solution can do is put you in a new form of .dll hell from the old days.

Let's take a simple example...
Let's say that you have sn SP in prod that's giving you problems and you want to track down where the degraded performance is coming from. So you open up the SP and start looking at the code. You see that a major component of it is an external SP call. OK, so you open up that SP. Now you see that that SP calls 3 other SPs. And each one of them is calling more SPs. And so on and so on. This is the definition of spaghetti code isn't it?

I'm actually in the middle of doing this very thing right now. I wanted to stop and blog about it while I was still pissed off. I've been doing this for an hour now and I'm not any closer to a solution than I was when I started. So you can see that while going the COM route can be helpful, you can take it too far. And that's really what happens, isn't it? Devs take a good idea and drive it into the ground until it's so hard to manage it's not even worth having it anymore.

I've been trying to come up with some guidelines for this and it's tough. But roughly, I like to say that if several SPs use the same code, or are calculating the same thing, it's ok to pull out into COM. The same goes with code that gets changed a lot. If you've got code that changes often and is used in a lot of SPs, by all means put it into it's own SP so you only have one place to change it.

Anyway, I guess I'll get back to my plate of spaghetti now.

This post is dedicated to all those field DBAs who like to call up the prod DBAs and tell them how busy the server is based on the number of spids or short-term blocks returned by sp_who(2).

See this is the hidden cost of generic logins that have too many rights. Everyone on the planet can read BOL and try to interpret the results. And whatever they mark in their heads as being the sign of a server being too busy is what they're going to call you with. Our users have 2 criteria for a busy server. The number of spids (active or inactive), and how many short-term blocks they see.

Of course I used to try to explain to them that you could bring the server down with a single spid so the number doesn't matter... and that blocks are fine as long as they don't persist. Since I've been here for quite a while though, and none of them have gotten the hint yet, I usually just thank them for letting us know and that we'll get right on it.

One of my favorite analogies used to be that judging the server on the amount of spids is like loading your car up with people and declaring that traffic is really bad today. Somehow that still didn't get the point across.

It's almost never fun to lose a DBA, but it's a fact of life. People leave jobs, and sometimes jobs leave people. This is another reason why it's really important to not tax your DBAs too heavily. If you've got 2 DBAs and they're both working like dogs, what happens when you lose one? I'll tell you what happens... deadlines start to slip, backups start failing and don't get looked into, maint starts getting behind, security gets relaxed, etc. You want your DBA to have time to do his job and be able to pick up some slack when you lose someone. And it could be quite a while before you get a replacement. Good DBAs are really hard to find and you don't want someone to just warm the seat.
I talked about this recently in my IW blog. Of course, this really only goes for production DBAs, right? I mean, you can work your devs as much as you want. They'll never get a call in the middle of the night because the server's down or because a package failed. So again, DBAs are insurance policies. We're kinda like a clustered server. You don't use the inactive node. It just sits there waiting for something to happen to the primary node. It seems a terrible waste and managers hate spending that money for a box that just sits there. And while DBAs aren't quite that useless, we really should be used in the right way. So it really is just like a multi-node cluster. You never run the primaries at full capacity because one day something will happen and one box will have to take on its workload and one of the downed nodes. So if they're all running at 100%, they can't failover and resume work. So you run them at 50%... give or take, right?

So again, let your prod people do their prod jobs and don't put them on too many actual projects. Afterall, that's what prod means.

And yeah, we're losing our other DBA so I'm all alone now. We'll see how it turns out.

OK, I'm also changing this in the original post, but in case you don't think to go back and check.
When I sent my admin password solution to a buddy at MS, he tried it and it didn't work. After a fairly short discovery, I discovered that it worked for me because I had a drive mapped to the admin share on my DC. So PSExec was using that IPC to connect. So unless you're lucky enough to have something like that, you prob won't be able to use this solution. However, it's a good backdoor so it may be a good idea just to setup a mapped share on your box anyway just in case something happens and you get shut out of your DC.

On that note. This is a really excellent reason why you never want to run SQL under an admin acount and especially not under a domain admin acct. I've seen that more than I care to and it always comes out bad eventually. It only takes a very basic knowledge of SQL to discover that someone with regular user rights can setup a SQL job to promote their user acct to admin because the job will run under the context of the agent acct at the OS level. And if you're running your services under an admin acct they'll have full rights to do whatever they like. And if you're running it under a domain admin acct, then they can promote themselves to domain admin pretty easily by running a .vbs or prob even powershell. Pretty much any scripting language will prob do, huh...
And that goes for running SQL on a DC under local system. That's the same as domain admin as far as the DC is concerned. So be smart and run your SQL accts with non-admin accts and just remove that from the equation. There are enough ways for internal and external hackers to elevate rights. Don't help them.