RSS

Tag Archives: DPM

DPM… and “Why are my SQL Transaction Logs filling up?” (the answer… really)


Here at Avanade we run a lot of pre-release software from Microsoft, so we can work the bugs out, so to speak, and thus be able to get better solutions to the customers we’ve got.

One of the software packages that we use is the System Center Suite – including, in this case, the Data Protection Manager part.  It’s now in production, released product for some time, and we now run all our backups with it.

The thing that is totally weird for me as a DBA is that all of the backup jobs I had on the boxes got disabled, and DPM took over. It was like driving a car with an automatic transmission for the first time. “What do you mean I don’t have to do backups?”

If you have things set up properly (which is a deeper subject than this blog post is meant for) – it’s close to hands off – which, come to think of it, is still weird.

So, over the time we’ve had it – I’ve found that there are several things to be aware of in the implementation of it.  This is not an all-inclusive list of surprises, this is just a couple of the things that I found out – after researching them myself and finding a lot of bad information out there.

The information below is from my own experience, from conversations with Microsoft, and from research I’ve done.  That said, my goal is to help keep you out of the DPM weeds by helping you

  1. Understand how DPM handles SQL backups in the varying recovery models you can have (full, simple, etc.)
  2. Understand where it needs drive space. (this can be an absolutely evil ‘gotcha’ if you’re not careful)
  3. Edge cases.
  4. How items 2 and 3 can intertwine and get you into deep doo-doo and what you want to do to stay out of it.

So, ready?

Oh – assumptions:

You’ve got DPM installed, and for the most part, configured.  It’s working, but you have transaction log drives filling up on some of your servers, and it’s not really clear why.

Wanna know why?

Here’s the answer:

It’s because the UI is very unclear, because the documentation is unclear, (there was a hint of it on page 83) and because the things that would be obvious to a DBA simply aren’t mentioned.

So, having said that – let’s explain a little.

After a few years of running it, and we flog it to within an inch of its life, I’ve come to – if not make friends with it, then at least I’ll give it grudging respect.

But you have to know what you’re doing.

So first off: Settings.

You organize your backups in DPM in what are called protection groups.

Think of it as a glorified schedule, with lots of properties you can adjust.

Now when you’re creating the protection group, you can create it to back up file servers, Exchange servers, and SQL servers.  We’ll talk about SQL servers only here.

So when you back up a SQL box, you might have some databases (let’s say the system ones, Master, Model, and MSDB) in simple recovery mode, and the user databases in full recovery.

What you’d do is create two protection groups.

One for the system databases, for simple recovery.

And one for the user databases, for full recovery.

And this is where the first gotcha comes into play.

See, when you create that protection group, you’re going to come across a tab in the creation of it that gives you this choice… It’s like the matrix… Blue pill? Red pill?

And, when we were setting it up, we scoured the documentation to try to figure out how to set it up for databases that were in full recovery mode as well as doing mirroring.

And we kept running into problems with full transaction logs.

It turned out I wasn’t alone in this.

I googled the problem…

I binged the problem..

Heck, I even duckduckgo’ed the problem.

Everyone, it seemed, had the same question.

And the answers were surprisingly varied.

And most, honestly, were wrong.  (Note: that’s not to sound arrogant, this was a tough one to crack, so a lot of folks were coming up with creative ways to try to work around this issue)

Some folks were doing what we’d done initially *just* to keep the systems running. (manual backup, flip to simple recovery, shrink the logs, flip back to full recovery) – yes, we absolutely shredded the recovery chain, just shredded it – but the data in that application was very, very transient, so keeping it up and functioning was more important than keeping the data forever.

So while we were frantically – okay, diligently – searching for answers, managing the problem, we were also looking for a cure to the problem, because there was no possible way this could be an enterprise level application if it was behaving so badly… Right? There had to be some mistake, some setting we (and everyone else in those searches above) weren’t seeing, and it finally ended up in a support call.

My suspicion was that the transaction logs weren’t being backed up at all, even though that’s what we thought we were setting.

I’d been around software enough to know that clicking on an insignificant little button could wreak catastrophic results if that’s not the button you meant to push.

And this was one of them.

See, the databases that were in full recovery (namely those in that app that had the mirroring) were our problem children.  Databases in simple recovery weren’t.

It made me wonder why.

And one day, I was on the phone about another DPM issue, (for another post) and I asked the question, “So what exactly is happening if I click on this button versus that one? Because my suspicion is that the tlog is not getting backed up at all.”

And then I asked the more crucial question for me, and likely for you who are reading this:  “What code is being executed behind these two options?”

And the fellow over at Microsoft researched it for me and came back with this:

“In DPM, when we setup synchronization for SQL protection, those Syncs are log backups. DPM achieves that by running the following SQL Query

BACKUP LOG [<SQL_DB>] TO DISK = N'<Database_location_folder>\DPM_SQL_PROTECT\<Server_Name>\<Instance>\<DB_Name>_log.ldf\Backup\Current.log'

Recovery Points are Express Full backups. If we setup a group to run sync just before recovery points, DPM will create a snapshot from the replica and then create a Full Database backup (sync).

BACKUP DATABASE [<SQL_DB>] TO VIRTUAL_DEVICE='<device_name>'

WITH SNAPSHOT,BUFFERCOUNT=1,BLOCKSIZE=1024

In this case we will never have log backup thus no log truncation should be expected.”

What does this mean?

It means that if you have a database in full recovery, you will want to put it in a protection group that is set to schedule the backup every X minutes/hours like this:

In DPM, click on “Protection” tab (lower left), then find the protection group.

Right click on it and choose ‘Modify’ as below.

GEEQL_DPM_Modify_Protection_Group

Expand the protection group and pick the server you’re trying to set up backups for there – you’ll do some configuring, and you’ll click next a few times, but below is the deceptively simple thing you have to watch…  This dialogue box below – which will have the protection group name up at the very top (where I’ve got ‘GROUP NAME HERE’ stenciled in) can bite you in the heinie if you’re not careful.  So given what I’ve written, and from looking at this and reading what I wrote above – can you tell whether this is backing up databases in full or simple recovery mode?

GEEQL_DPM_Backup_Full_Recovery

See how it’s backing up every 1 hour(s) up there?

That means the code it’s running in the background is this:

BACKUP LOG [<SQL_DB>] TO DISK = N'<Database_location_folder>\DPM_SQL_PROTECT\<Server_Name>\<Instance>\<DB_Name>_log.ldf\Backup\Current.log'

We’ll get into more detail in a bit, but this means you won’t have full transaction logs.  This is the setting you want for the protection group you’ve got to backup your databases in full recovery mode (and the ones that are mirrored or in Availability Groups). The other option you have is to back up “Just before a recovery point” – which, if you’re thinking in terms of SQL and transaction logs, really doesn’t make a lot of sense.  We went through the documentation at one point, and I think we were right around 83 pages in before it gave an indication of what it *might* be doing here – but even so it wasn’t clear, but now we know.  So what you’d want to have in this protection group would be a bunch of databases in full recovery mode.  You might want to create different protection groups for different servers, or different schedules, that’s all up to you… The crux is, if it’s a database in full recovery mode, this is how you want to set it up, by backing up every X minutes/hours… Making sense?

Okay, let’s take a look at the other option…

GEEQL_DPM_Backup_Simple_Recovery

If you have a database in simple recovery, you’ll want to put it in a protection group that does backups just before the recovery point.  And that’s what the screenshot above does.    When you click on that radio button, the code it runs in the background if you’re backing up SQL databases, is this:

BACKUP DATABASE [<SQL_DB>] TO VIRTUAL_DEVICE='<device_name>'

WITH SNAPSHOT,BUFFERCOUNT=1,BLOCKSIZE=1024

And you should be set.

You can change the frequency of the express full backups by clicking on the ‘modify’ button in the dialogue above, and you’ll have quite a few options there.

Understand, you have several different buckets to put your databases in.

  1. Simple recovery (see above)
  2. Full recovery (see above)
  3. Whatever frequency you need for your systems (from above)
  4. Whatever schedule you need for your systems (from above)

Believe it or not, that’s it.

Put the right things in the right place, and DPM is close to a ‘set it and forget it’ kind of a deal.

However…

…there are some Gotchas and some fine print.  This is stuff I’ve found, and your mileage may vary – but just be aware of the below:

  • If you put a db that’s in simple recovery into the protection group meant for databases that are in full recovery, you’ll likely get errors with DPM complaining that it can’t backup the log of a database that’s in simple recovery mode. Since you manually put that db (in simple recovery mode) into that protection group (that’s configured to back up databases in full recovery mode), it will be your job to get it out and put it in the right protection group.  That will make the alerts go away.
  • If you put a db that’s in full recovery mode into the protection group meant for simple, you’ll fill up your transaction logs, fill up your disks, and your backups will fail, and you may, depending on a few factors, hork up your database pretty bad… (this is what most people complain about, and that will solve the disk space issue). And, since you (or someone on your team) likely put the db in the wrong protection group, putting it in the right protection group will be the first thing to do… Having enough space on your log drives is critical at this point – because DPM will start to make copies of your transaction logs as part of its backup process, and will need the room (as in, half of your transaction log drive).  More details below.
  • I’ve found a couple of Corollaries to go with this:
    • Corollary 1: DPM creates a folder on your transaction log drive called “DPM_PROTECT” -it stores copies of the transaction logs in there.  Those are the ‘backups’.
      • You have a choice between compressing backups and encrypting them…
      • If you encrypt them, they’re full sized, even if they’re empty.
      • So if you have transaction logs filling up 50% of your t-log drive – guess what’s filling up the other half?  (DPM’s t-log backups).  That DPM_PROTECT folder is a staging folder and is sometimes full, sometimes not (making it devilishly hard to monitor for), but you need to be aware that if that folder fills up half the drive, you’re running very close to disaster, and that’s when you have to start getting creative in your problem solving (see ‘examples’ below)
    • Corollary 2: DPM can be configured to put the DPM_PROTECT folder on a separate drive, which may suit your needs, and is a topic that will have to be discussed in a separate post, but if you run your transaction log drives pretty full, and have cheaper storage available, this might be an option for you to consider.  We don’t have ours architected that way, so it’s an option I’ve not tried.
  • Examples of things that can go very wrong (like the disaster mentioned above)
    • If you are working with a clustered SQL server and are getting alerts because your log file can’t grow, chances are it’s because your transaction log drive (or mountpoint) is full, and it will be full of both transaction logs, and DPM’s staged backups of the transaction logs.To fix this, you will either need to
      • Extend the drive/make it bigger  (assuming that’s an option for you) and then restart your DPM backups.
        • Note: DPM will likely want to run a ‘validation’ at this point, which will take some time.  My recommendation is to let it do that, but there’s a huge “it depends” associated with this one.  Sometimes – depending on how long things have been broken before you were able to get to them, you might find yourself essentially taking that database out of DPM’s protection group and starting over.  It breaks the recovery chain, but can be faster than letting DPM do its validation of what it thinks your latest backup is compared to your existing one (where you likely broke the recovery chain with the manual log backups)… Like I said, it depends..
      • (not advised, but if you’ve run out of options) backup the database once, and backup the log manually repeatedly (via SQL, not DPM, because you’re trying to empty the drive that DPM has filled up) until you can shrink the transaction log so you have space on the drive for DPM to actually stage a copy of the log file for backup.
        • Once you’ve done that, remember, you’ll have fixed one issue but created another, namely, your recovery chain ends in DPM where you started doing manual SQL backups.  Now you have backups in DPM, and a full + a bunch of log backups in SQL.  Make sure you have a full set of backups in case things go wrong.
    • You’re working with a mirrored server or a server that’s part of an availability group and the databases are in the wrong protection group (simple instead of full recovery)…. You’ve got transaction logs filling up from both the replication that’s involved in this, and transaction logs filling up because they’re not being backed up… It gets ugly.  I ran into this (and wrote about our resolution to it here) where we had issues with log file growth, high numbers of virtual log files, and an availability group with multiple geographically dispersed secondaries… It was, um… “fun…” <ahem>

So…  All this to say something very very simple: Pick the right recovery group, know what’s going on behind the curtains that are in front of what DPM is doing behind the scenes, and honestly, you should be good.  If you understand what radio button to select when you’re configuring the protection group, you as a DBA are about 90% of the way there.  Make sure your transaction log file drives are twice as big as you think they should be (or configure DPM to store them elsewhere), because chances are, you’ll be using half of the transaction log drives for the logs themselves, and the other half for temporary storage of the backups of those transaction logs.

Know what your protection groups will do for you… Know the gotchas, and DPM will, strangely enough, be your friend…

Take care out there – and keep your electrons happy.

Advertisements
 
11 Comments

Posted by on June 21, 2016 in Uncategorized

 

Tags: , , , , ,

Thinking outside the Box in Disaster Recovery


I had an interesting ‘aha’ moment awhile back.

I was trying to take a well-earned day off from work, visit my mom on her birthday, and in general, do a little decompressing.

I’d recently taken her portrait (a profession from a former life) and my goal was to give it to her and later take her to an appointment and then lunch.  We were between the appointment and lunch when my phone buzzed.

It was one of my Faithful Minions from the land of Eva Perón, who was in a bit of a bind.

Seems that a server had a database that had gone corrupt.

No, not the Chicago style “vote early, vote often” corruption.

Database level corruption.

It seems that this server had been giving us some trouble with backups (it didn’t want to back certain databases up, depending on what node of the cluster it was on) – and the team responsible for fixing that had been working on it for a few days – with limited success.

Aaaand another one of my Faithful Minions, from the land of Chicken Curry, had tried several things, one of which included detaching the database.

Which worked.

It’s just that once you detach it, you likely want to reattach it.

But trying to reattach a database that’s been munged to some degree is a bit of an issue, as SQL tries to recover the database and chew through the transaction log file prior to bringing it online.

That’s as it should be.

Problem is, this db had some level of corruption that would cause an issue when it got to 71% recovered. Stack dump, the whole nine yards.

It was, in a word (or several) a perfect example of a disaster recovery exercise. Yay! I’d always wanted to have one of my very own…

(well, not really, and not on my day off, and…)

Hmm…

I gave my Faithful Minion from the land of Perón some instructions while sitting in the car in front of the restaurant Mom and I were going to have lunch at, and then tried to enjoy birthday lunch with her, but still had my mind being pulled dripping out of the clam chowder and back to the database while Minion worked on his Google Fu a bit to try to figure the error out. Mom and I finished up lunch and I took her home, where I logged in to see what Minion and I could accomplish.

So the thing is, we use DPM (part of Microsoft’s System Center suite of products) to do backups. That’s Microsoft’s Data Protection Manager, a program – no – a system that really takes some getting used to. There will be another post in this series about it – but one of the big things you have to get used to as a DBA is that you don’t do backups anymore…

You don’t use Maintenance Plans.

You don’t use Ola Hallengren’s or any of the other amazing tools out there to handle backups.

You really don’t do backups any more.

DPM does.

No backup jobs on your servers…

No backup drives…

No…

Backups…

DPM handles them all semi-invisibly…

It is – well… weird.

It took me a long time to wrap my head around it.

But it works.

And that’s where this situation kind of got a little strange.  (Well, strange*r*)

See, DPM creates Recovery Points (its version of a backup) and it will stage the backup on disk either locally on the SQL box or in the DPM system where you have MDF and LDF files created with some sort of wizardry before it’s shipped off to tape.

So, we poked and prodded, and did all sorts of technical things, until my next set of Minions from The Pearl of the Orient came online – and we tried to find the backup of the database from DPM.  This took a good while.

Longer than we were expecting, actually.

While they were looking and I was troubleshooting, much time had passed in the land of rain (a bit south of Seattle) and lunch was long, long past.  Mom made some dinner while I was working on my laptop, sitting in my dad’s old chair, while I said goodbye to and thanked a very tired Minion in the land of Perón,   Meanwhile, Minions in the Pearl of the Orient, and Minions (and a Minion in Training) in the land of Curry were all trying to help, trying to troubleshoot, and in the case of Minion in Training, trying to learn by soaking up individual droplets of knowledge from a fire hose of a real live DR exercise.

Which is where it got interesting…

See, with DPM, you can choose a recovery point you want to restore to, and what it will do is simply replace the MDF and LDF files on your SQL box with the ones it’s reconstructed in all its magic.

The good news about that? Once you’ve learned how DPM works, anyone can pretty much restore anything that’s been backed up. That means Exchange servers, file servers, SQL servers, you name it…

It is (and I hesitate to use this word at all) kind of cool.

You need a restore?

All you have to do is open up DPM, find the database you need…

<click>

<typety type>

<find the backup that happened just prior to whatever blew up.>

<click> <click>

<click>

<pick the recovery point you want>

<click… scrollllllllll…. Scrollscrollscroll <click> <click>

Then you wait till DPM says it’s ready.

And tadaa…

That is…

If DPM had backed that database up.

But for that particular database…

On that particular node…

Of that particular cluster…

That database had not been backed up.

By DPM.

For… Let’s just say it was out of SLA, shall we?

It was truly a perfect storm.

The team *had* been working on it – it just hadn’t gotten actually fixed yet, and the crack that this database had slipped through grew wider and wider as it became clear to us what had happened.

It’s the kind of storm any one of us could run into, and this time it was me.

We looked high and low for a DPM backup set, and found one on that was supposed to be on tape.  But, given that this was Thursday where I was and Friday where various minions and co-minions were, the earliest we could get the tape would be by Monday…

More likely Tuesday…

On a scale of one to bad, we were well past bad.  We’d passed Go, had not collected our $200.00 – and…

Well, it was bad.

Several days’ worth of waiting on top of the existing time lost wasn’t really going to cut it.

Then, Co-Minion found that there was a DPM backup on disk – but a day older than the one on tape.

I could have that one in two hours (the time it took to copy the mdf/ldf files over and then put them in the right location.)

Hmmmm. A backup in the hand is worth two in the – um… Tape library?

So we copied it over.

And I attached it…

And now we had a database on the box that worked.

Yay, system was up and running.

But…

it wasn’t the database I wanted.

Like any of you out there, I wanted the backup from just before the database had blown up, and because of all the troubleshooting that had happened before I got involved, some of which actually made the situation quite a bit more challenging, it turned out I couldn’t do anything with the files I had… so reluctantly, I had to go with the older files, simply because I could get them quicker.

And the next day, while trying to fix another issue on that cluster, I got to talking to the guy who runs the SAN and the VM’s, and I explained what all we’d gone through over the course of getting that database back, and he said, “Wait, when did you need it for?”

I told him.

“Oh, I’ve got copies on the SAN of those drives – we could mount it if you need – 12 hours ago, 24 hours ago, whatever…”

He… You…

What?

I realized what he was telling me was that if I’d contacted him (or, frankly, known to contact him at that time of night) – I could have had a much more recent backup sooner instead of spending so many hours trying to fix the existing, corrupt database.

What it meant, was I was thinking inside the various boxes I was dealing with…

Trying to get a *SQL* backup…

Trying to get a *DPM* backup…

When what I needed, frankly, was bits on disk.

I needed MDF and LDF files.

Not only could DPM get me MDF and LDF files, but so could the SAN.

And the SAN backups had them.

I just didn’t know it.

By the time I found out – the old database we’d attached had had data written into it and had been up for a good while. Merging the two would have been far, far more of an issue than it was already (I’d had experience with that once before as an end user) .

So where does this leave us?

It means that if you’ve got a SAN as your storage mechanism that’s backing itself up, you might find yourself thinking outside the box you’re used to thinking of when trying to restore/recover a database.  Go buy the person or people running your SAN a cup of decent coffee and just chat with them, and see if this kind of stuff is possible in your shop.

They might be able to help you out (and save you substantial amounts of time) in ways you quite literally couldn’t have imagined.

In the end, I learned a lot…

My team learned a lot.

My mom learned a lot.

I will be making some changes based on what we learned, with the goal of being able to have some more resiliency in both the hardware resources we have as well as my faithful Minions who worked hard at trying to fix things.

Ultimately, one of the takeaways I got from this was something simple, but also something very profound.

Make sure the stuff you’re doing in your day job continues to work, so you can actually have a day off to spend with your loved ones when you need it.

Oh – and the portrait of mom? She loved it.

It’s here:

The portrait of Mom I wanted to deliver to her.

The portrait of Mom I delivered to her.

 

Take care out there, folks…

Tom

 

 
Leave a comment

Posted by on April 1, 2016 in Uncategorized

 

Tags: , , , , , ,