This week, Brent, Erik, Jessica, Angie, and Tara discuss SQL service packs, partial restores, breaking replication, backups, as well as their favorite TV shows.
If you prefer to listen to the audio:
Jessica Connors: Let’s start with the service pack question from Jay H. He asks, “When patching applying SQL service pack to passive node in a cluster should one put into pause mode or okay to simply run the service pack?”
Tara Kizer: I’ve never put it into pause mode. I’ve patched hundreds, maybe thousands of servers.
Brent Ozar: Well but do you patch the passive? I mean you patch the passive and there’s no services running on it so you really don’t have to worry about it.
Tara Kizer: Yeah, nope.
Brent Ozar: Yeah, am I going to have to worry about something failing over to it in the middle of the service pack? I’ve never done it either. Talk about a first-world problem there. So I’ve never paused it. I don’t know, if I was probably going to get ambitious, like PowerShell script the whole thing out, I would probably put in some kind of pause there but I’m just not that guy.
Jessica Connors: What’s this about Terabyteasaurus Rex?
Brent Ozar: That is a presentation that I do about very large databases. There’s things that you want to do to first avoid very large databases. Then once you have one, how you do to cope with the aftermath. It’s like having your own little pet dinosaur. You have to plan for things ahead of time and don’t bite off more than you can chew.
Angie Walker: Like a Tamagotchi.
Brent Ozar: Like a Tamagotchi, yes. Only you can’t carry it around in your pocket.
Jessica Connors: Is that a … presentation you’re doing coming up?
Brent Ozar: That one’s actually from our regular training classes. I’m doing it for IDERA. IDERA pays us to periodically do free presentations for the community. We give them like our menu of private training courses and just go, “Here, which one do you want to buy for the community?” They go buy one and we give it out for free to everybody and off we go. So it’s really awesome how we work that with vendors.
Jessica Connors: Nice.
Brent Ozar: Everybody wins.
Jessica Connors: All right. This question is pretty vague from Abdullah. He says, “Hello. We have just received new hardware with 124 cores and 1 terabyte of memory to host many of our SQL instances.”
Tara Kizer: Dang.
Brent Ozar: Big spender.
Erik Darling: Bad news: the only licensed …edition.[Laughter]
Brent Ozar: Wow. One word, virtualization. Virtualization. Normally, I’d never buy a host that big but I just wouldn’t do instance stacking. I wouldn’t run one Windows OS and then a lot of instance stacking. Have you guys ever run multiple instances on a server and what problems have you seen from it?
Erik Darling: Not willingly. I’ve inherited them and it’s always been like, okay, we’re going to really stifle this one which is less important because this other one is just choking it out anyway.
Tara Kizer: I used to have a four node cluster with eleven SQL instances on it.
Brent Ozar: Oh.
Tara Kizer: Yeah, the biggest challenge was installing patches. There was always some node that suddenly had to be rebooted. This was back on before 2008. It was horrible. Even after you rebooted the server it would say, “Oh, that server needs a reboot.” It would take an act, a miracle, for all four nodes to agree it’s now time to patch. I would reboot like 20 times before it would say it. It was horrible.
Jessica Connors: Oh boy. Did you end up running them all? Did you end up consolidating?
Tara Kizer: We ended up upgrading to newer versions where you could install it passively. Back then the SQL instance had to be in the right state on all four instances, or not the SQL instance, but you had to patch all four instances at the same time. So all four had to agree that it was ready to be patched.
Brent Ozar: Miserable.
Tara Kizer: It was horrible.
Brent Ozar: Still even today Windows patching is a giant pain in the rear. Do you want to take down all of the instances at once just in order to patch Windows? Windows patches come out kind of fast and furious. So virtualization adds that little bit layer of lower performance, and I’m not even going to go down the is virtualization slower or not. But worst case scenario, it’s a little slower and you deal with that. But then you get so much better management. Holy smokes. It’s way easier.
Jessica Connors: All right. Let’s talk about partial restores. “Can you recommend a good article on doing partial restores? I want to move my historical data to a separate NDF file so I can restore just the active portion right away so the users can use the database then restore the historical part.”
Tara Kizer: Do we have an article? I mean I know someone who does a session.
Brent Ozar: Who?
Tara Kizer: I’ve seen it twice.
Brent Ozar: Is it available publicly? Is there somewhere, a blog we can link to?
Tara Kizer: I don’t know. I’ve sure she has a blog. Kimberly Tripp has a whole session on it. At patch she’ll bring a USB, all these USBs, and she’ll unplug them to simulate losing a drive and the database stays online because it’s all in memory. It’s really cool but it has to do with the partial restores as well.
Brent Ozar: I bet if you go to hit Google and you do like “partial restore site:SQLSkills.com” I bet she’s got blog posts on it too because it’s one of her more famous demos. We talk about it in our training classes but I was just sitting there thinking, I don’t think we have a single public post on it.
Erik Darling: I was working on something similar when I was messing around with some foreign key stuff but SQL outsmarted me so I didn’t end up writing the post.
Tara Kizer: Erik will take care of that in the next 15 minutes though.[Laughter]
Brent Ozar: Erik “the blogger” Darling.
Erik Darling: Oh, get out of here.
Brent Ozar: There’s Bob Pusateri, @SQLBob on Twitter. Bob Pusateri has an article too on his site to help move your stuff into a historical file group. He has really nice scripts that help you rebuild objects onto other file groups which is way trickier than it looks like. There’s things like off-row data that don’t move by default.
Erik Darling: If you’re on a version prior to 2012, a lot of that stuff is still offline, especially for large varchar and varchar types.
Jessica Connors: All right. A question from John. He says, “When looking at statistics properties, it gives the date and time when statistics were last updated. How can I tell if that statistic’s object was last updated using a full scan or a sample scan and what that sample scan value was?”
Erik Darling: I know all of this.
Jessica Connors: Do you?
Erik Darling: Off the top of my head. So if you run DBCC SHOW_STATISTICS on the table and index that you’re interested in, you can either run the whole command or you can run it with stat header which will just give you the top row of output. There will be two columns in the stat header part. One will be rows and one will be rows sampled. If rows equal rows sampled, then you did a full scan. If rows sampled is less than rows, then it used a percentage. You can also hit a function sys.dm_db_stats_properties. If you cross apply that with sys.stats, you can pass an object ID and the stats ID and that will also tell you rows sampled versus rows in the table. So you can figure out all that stuff there. If you want to calculate a percentage, just put a calculation in your queries. That’s the way to tell.
Jessica Connors: Thanks, Brent Ozar.[Laughter]
Why are you still Brent Ozar?
Brent Ozar: My alter ego. He’s logging in first to start the webcast because I’m shlumpy, lazy, and I don’t show up on time. So god bless him.
Erik Darling: Brent is taking cold drugs and hanging around.
Brent Ozar: Yes.
Jessica Connors: Got ya. Okay, question from Cameron. He says, “If you want to purposely break replication…”
Brent Ozar: What?
Jessica Connors: Why would you purposely break replication?
Brent Ozar: Is he trying to drop the microphone on his way out the door?
Angie Walker: Should we continue answering this question?[Laughter]
Jessica Connors: “Is it better to unsubscribe from the slave database or should you delete the publication from the master?”
Brent Ozar: Oh, he wants to do a restore.
Tara Kizer: I haven’t used that term slave database, I assume he’s referring to the subscriber for a restore.
Brent Ozar: Subscriber.
Tara Kizer: I just right-click on the publication and say drop it and it takes care of everything.
Brent Ozar: Like it’s hot.
Tara Kizer: But no, you don’t delete the publication from the… I’m confused by the terms that he’s using.
Brent Ozar: I bet, so one thing I would say is I would do Tara’s approach rather than trying to remove any subscriber’s setups because what happens if you leave the publisher’s setup and somebody doesn’t do the restore? Like I’m assuming you’re trying to restore the publisher, not the subscriber. You can leave a hosed up replication setup behind. So as long as there’s only one subscriber, just delete it at the publisher.
Tara Kizer: If you end up with the hose situation, you can run sp_removedbpublisher, something like that. It’s remove something. Remove db something. That will just clean up anything that was left behind.
Brent Ozar: That’s how you know somebody has worked with replication before.
Tara Kizer: Yes.
Brent Ozar: She’s like looking up the rest.
Tara Kizer: Bailed over to the DR site and forgot to drop replication beforehand and it like orphaned at that point. It’s like, what is that command?
Brent Ozar: Oh god.
Jessica Connors: Tara has the best stories.
Tara Kizer: Lost lots of sleep.
Brent Ozar: God bless.
Jessica Connors: Yeah, you could have your own podcast. Yeah, that’s what he says, he says it’s just to do a restore. He’s not in there just for the sake of breaking stuff now that we know his first and last name.
Tara Kizer: I know how to break replication too. He meant drop it.
Jessica Connors: All right. Let’s talk about stack dumps. Chris has a question. He says, “My SQL error logs show a stacked up. Total server memory is 16GBs, Mac memory is setup just above 13GBs, LPIM is enabled. Available memory went to zero just before the crash according to my monitoring software. I’m thinking I should lower the max memory or disable the LPIM. What do you think?”
Tara Kizer: You need to figure out what is using the memory but I don’t think your 2.5 is enough for the OS. The 2.5 I mean.
Brent Ozar: Yeah, when Tara says find out what’s using the memory, it’s not SQL Server, or at least not the engine. It could be other things in SQL Server like integration services or analysis services or whatever but you set max memory, you set lock pages, and memory turned on. So that’s cool. That amount is locked but now, and people will often say, “I should turn on lock pages and memory. That way I don’t have to worry if something else needs RAM.” Hell yeah you do. You just did. You just suffered a crash because of it. SQL Server couldn’t back down on the amount of memory it needed. So now your fun journey begins to go troubleshoot what is using that extra memory. What would you guys use in order to find out what tools or what apps are using the extra memory?
Tara Kizer: Well I had low memory on my laptop on Monday during a client session. So after I was finally able to investigate it after the call in either the app log or the system log, it told me what the top three processes that were using the memory. It was two SQL server instances and WebEx; those were the top three. I’m not too sure if that would be seen if your actual server crashed though, but maybe. There might be low memory alerts in there leading up to the crash.
Brent Ozar: You’re right on the thing in saying lock pages and memory, should I maybe turn it off. I would while I’m doing the investigating. Just leave it off just to prevent—because this other app is probably going to fire up again before you have the chance to fix it.
Jessica Connors: All right. So let’s talk about what to do when your backups fail. Fred has a question. He says, “Checkdb is successful and our backups always complete successfully but trying to restore the backup gives an error that the backup failed, the data is invalid. Any thoughts on where to look.” He’s running SQL 2008 R2 Enterprise. Not using any Enterprise-only features. Four backups later we usually get a good backup that we can restore from.
Erik Darling: My first question is are you running backups with checksum enabled and do you have page verification turned on for that database? Because you could at least narrow down where the issue is happening. So if you have page verification turned on, SQL will start writing checksums to your pages to make sure that nothing wonky happens to them on disk. Then if you run backups to the checksums, SQL will check those checksums as it does the full backup. So you at least have something verifying there that it’s not SQL and that it’s something with your disk media or like when you’re transferring files over that’s happening. The only time I’ve ever seen that happen was when I was restoring a backup on a dev server and it turned out that one of the disks on the dev server was actually suffering from some malady, some lergy. So that was the issue on that. So I would absolutely verify that it’s not something happening on the primary sequences and then my investigation would be on whatever hardware I have on the instance I’m trying to restore it to.
Brent Ozar: I’ve seen it when you write to crappy storage for your backups, like the backup reports writes successfully but then the data is trash when you go through to read it. But I would like to say, just like the great singer Meatloaf, two out of three ain’t bad. Four out of five successful backups, that’s not such a bad number. 80 percent, that’s a passing score. You probably didn’t need one out of five. It’s probably not that big of a deal. But yeah, I would just try immediately after the backup finishes, try doing a restore with verify only. Either from the same SQL Server or from another SQL Server and that will at least tell you if the backup’s file is hosed.
Jessica Connors: All right. Back to replication.
Brent Ozar: How come nobody ever says, “What do you guys think about flowers? What’s your favorite kind of chocolate?”
Erik Darling: How much do you like [inaudible: Casa Playa]?
Brent Ozar: It’s dreamy.
Jessica Connors: Let’s see. Question from John. He says, “Is it possible to mirror via transactional replication a SQL 2008 R2 database to a SQL 2016 database?”
Brent Ozar: Wow. I bet it would be.
Tara Kizer: I think so since 2016 goes all the way down to 2005 compatibility level.
Brent Ozar: Yeah, I bet you could.
Erik Darling: The only thing I’ve ever seen stand in the way is backwards stuff.
Tara Kizer: Mirroring and transactional replication, replication doesn’t really care about versions. Mirroring does.
Brent Ozar: Yeah, he should be fine.
Tara Kizer: Either way, it should be fine.
Jessica Connors: Can you just upgrade both to 2016?
Brent Ozar: I bet he’s so tired of running with his 2008 R2 box and he’s like just trying to give his users something that has nice, new functionality on 2016. That’s probably what it is. He’s like, “Here, go query over here. It’s really nice and fun.” Maybe he’s got nice nonclustered column store indexes on the table over there, make his queries fast. Maybe that’s what it is.
Jessica Connors: Kanye or Wanye West.
Brent: Wanye. [Laughter] Oh, Wayne you are never going to live that down, Wanye.
Jessica Connors: I think that’s a good question. Where is Richie?
Brent Ozar: Oh, he’s in Disneyworld.
Jessica Connors: Of course he is. He’s always getting lost at Disneyworld.
Tara Kizer: Driving home.
Angie Walker: Yeah, he’s on the road.
Jessica Connors: Let’s see here. Question from Jay H. He says, “Last year after applying a couple of particular Windows KB updates issues arose with JBDC TSL connections and had to be removed. Has Microsoft fixed this and can updates now be applied?”
Brent Ozar: I remember seeing Aaron Bertrand blog about this. This is one of those SSL and the connection string I’ve never paid too much attention to but I think Aaron Bertrand blogged about this. Other people are nodding like we vaguely have seen something along these lines.
Erik Darling: Yeah. I’ve just seen some stuff sitting around 2016 with TLS 1.1 and 1.2 having some weirdness bug things.
Brent Ozar: Yeah, we don’t touch it with a ten-foot pole. If you search for SQL Server TLS Aaron Bertrand and Aaron is A-A-R-O-N Bertrand, I bet you you’re going to find a blog post in there where he went into detail on that. Because like anything else with SQL Server updates Aaron Bertrand looks over those with a magnifying glass and a fine-tooth comb.
Erik Darling: In a kilt.
Brent Ozar: In a kilt. So Rusty Householder already replied with the answer, blogs.SQLSentry.com TLS support. I’m going to put that in the resources for everybody there. But yeah, it is a thing that Aaron blogged about.
Jessica Connors: Let’s see. This is the last comment/question. People have been pretty quiet today. From Chris Wood, he says, “Thanks for the help on the blocking.” You helped Chris with blocking?
Brent Ozar: I believe we did last week I think. I think we did.
Jessica Connors: Via Critical Care?
Brent Ozar: Oh no, it was a question about—I remember this. It was a database restore involving Relativity. He was doing a database restore on Relativity and I think we posted the question on Stack Exchange as well. SP who was active showed blocking and we couldn’t figure out which query it was that was doing the blocking. Turned out it was a system SPID that was a doing a full text crawl. So when you finished doing restore, it did a full text crawl and it locked some of the tables in the database. People weren’t allowed to access them. Awesome. Got to love that. People are like wow…
Erik Darling: I could have answered that one.
Brent Ozar: Oh could you, have you had that same problem?
Erik Darling: Yeah, embarrassingly.
Brent Ozar: Unbelievable.
Jessica Connors: Angie is getting a shout out. Were you posting on #SQLhelp?
Angie Walker: Nope. [Laughter] Apparently I have an impersonator. I didn’t think there were any other Angies out there. Oh, Tara, ah. It’s pretty hard to tell us apart, I know.
Brent Ozar: Just one of us cartoons, all our cartoons look like.
Erik Darling: I get mistaken for Jessica all the time.[Laughter]
Erik Darling: They’re like, “Hey we need to buy some stuff.” I’m like…
Brent Ozar: Unsubscribe.
Jessica Connor: Yeah. Let’s see, Brent. Brent’s an awesome name. He says, “Any experience with Experian Correct Address which has to be installed on the database server for SQL CLR. Do you have any experience there?”
Brent Ozar: Oh, I’m vaguely remembering that this calls a web service. That it goes and validates people’s addresses. The way that it does it, whenever you want to like validate someone’s address you call an extended stored procedure, a CLR stored proc and it goes off and it calls a web service. So it’s possible that this is working on one node and not another because of firewall rules or network permissions. Windows Firewall, UAC, I mean, it could be almost anything that involves accessing the interwebs.
Erik Darling: I hate to say it but this is actually something that Master Data Services is good at.
Brent Ozar: Really?
Erik Darling: Yeah. You can do like address lookups and have like post office integration where you can get whatever like you know a post office valid address for a thing is, it can validate that. I don’t know a ton about it because I’ve only ever seen it in action a couple times but that’s actually something that Master Data Services does well which I feel filthy saying.
Brent Ozar: That’s a couple times more than me. I’m guessing it didn’t require CLR then, it was probably just stuff built into the database server?
Erik Darling: Yeah, but I don’t know how it was called so it still might have been CLR but it was integrated with Master Data Services. So it was like SQL friendly. It didn’t need to call out to anything else. It was already like built in somewhere.
Brent Ozar: Doing its thing.
Erik Darling: Yeah.
Jessica Connors: Let’s see. I think we talked about this last week. Nick Johnson, he says, “I found a couple articles that talk about how compatibility level 90 2005 does not work in SQL 2014. You guys have any confirmation on that even though 2014 in fact shows it, it doesn’t work.”
Erik Darling: You can’t upgrade directly, isn’t that it? Or is that from 2000?
Brent Ozar: I can’t remember either.
Angie Walker: I think in general, you can’t do more than two versions, right? You couldn’t go through 2005 to 2012 or straight to 2014. You’d have to make the hop in between.
Tara Kizer: You can’t restore but the compatibility level is there. So on 2014 you can go all the way down to 2005 compatibility level. Like he’s saying, the option is there but apparently some articles are saying it doesn’t work. I don’t know. I see it.
Brent Ozar: Yeah, I vaguely remember that during the release process, like it worked in SSMS but the ALTER didn’t work, like the ALTER DATABASE didn’t work. I think people thought, “It’s not going to work when it finally releases,” like that they’re going to yank 2005 compat. But I’m pretty sure it still does work because I distribute Stack Overflow in that format too and Doug was asking questions about that this week. It should work fine. I don’t know if it runs in 2016, if 2005 compat mode runs in 2016. No clue. And who the hell would use that? Why would I swear in the middle of a podcast? Who knows. Don’t do that. Don’t do that. By that, I mean 2005 compat mode, not swearing. You should totally swear. You’re a grown person.[Laughter]
Jessica Connors: All right, Abdullah asks or states he’s in progress building a DR site on a multisite topology. Any recommendation for SAN replication?
Erik Darling: The fast one.
Tara Kizer: It’s going to be expensive.
Brent Ozar: And the cheap one.[Laughter]
Brent Ozar: Yeah, you don’t get much of a choice. It’s whatever brand your SAN supports. I think EMC makes stuff as well, like appliances that will do between SANS. But cha-ching.
Erik Darling: Also make sure that you are deleting old snapshots and copies because you can find yourself—depending on like how big these copies are, deleting them could take a very long time so you want to make sure that you have enough room to both copy and have other stuff being deleted off. Because deleting 20 terabytes of SAN replication snapshots is time consuming no matter what.
Jessica Connors: All right, well I guess while we’re on the topic of that, do we like async DB mirroring as a DR strategy?
Tara Kizer: Yes.
Erik Darling: Yes.
Tara Kizer: I do. I used it for years. So when I joined a company three years ago, we were on SQL 2000 using log shipping and only upgraded 2005. We were relieved to get rid of log shipping just due to the amount of work it took to failover to DR site for a lot of servers. It was a lot of work. We were real excited about mirroring. We used asynchronous mirroring between our two sides. There was about 300 miles apart. We could not have done synchronous database mirroring because of the drastic performance degradation that you would have. But async mirroring worked great. We failed over to the DR site regularly, two, three times a year. Ran production out of it for a few weeks then failed back with it. It works great. All you have to do is when you want to failover, set it to synchronous mode, let it catch up, then do your failover and set it back to async.
Jessica Connors: There’s a Trump DBA handle on Twitter. Are you guys familiar with this?
Bret Ozar: Yes. I encourage humor in the SQL Server community. It’s not me doing it. That’s somebody else doing it. But I’m always like, if people want to have fun, that’s kind of cool.
Jessica Connors: What about DBA Reactions? Are you still involved with that?
Brent Ozar: I am. I took a brief hiatus from it while other people kept submitting stuff. Other people were submitting so many things. So I just went in and approved them all the time. Then of course I fell back in love with GIFs. I’m like, “Oh, let me go in and look and see because it’s been a while.” Oh my god, there’s so many good GIFs these days. So I started queueing them up again. I think I’ve already gotten them written through most of next week.
Jessica Connors: Oh, so you’re still doing that.
Brent Ozar: Yeah, I love it. I have a disturbing amount of fun with it.
Jessica Connors: They used to get a newsletter, the DBA Reactions.
Brent Ozar: I cut it back to like only once a week. It was going Tuesdays and Thursdays. Now it’s down to either just Tuesday or Thursday, I forget which one because I didn’t want to overwhelm people’s email boxes. There are like 5,000 people signed up to get this in their email box every week.
Jessica Connors: Yeah. I remember the time that somebody called for SQL Critical Care and they’re like “I heard about you guys from DBA Reactions. I love … I’m like, “we made a sale.”[Laughter]
Brent Ozar: Well and at that point they’re going to call us because anybody who’s crazy enough to go, “I like DBA Reactions, they’re my people,” they already know exactly what we’re like. They know exactly how we work.
Jessica Connors: That’s fair. Cool, well you guys are being fairly quiet today so I think we’ll end there.
Brent Ozar: Thanks everybody for hanging out with us. We will see you guys next week.