I’m always looking for a way to save space in SQL Server. From archiving old data to just flat out deleting unused objects, I take great joy in removing superfluous stuff. The less junk in the system, the easier it is to focus on the things that matter.
..and fit it in a 10 kg bag
The biggest useless space eaters are tables that are (supposedly) no longer used. I could script them with data to a file, but what if they’re 100+GB? I could also back them up to another DB and then drop them from the database; that would certainly free up the space in the original DB. What if they’re needed for some process that I was unaware of and we can’t wait for the time to restore/move them back?
My conundrum was this. So, I decided to implement a process that looked at a single DBA-controlled schema and compressed every table created prior to a certain date. I could TRANSFER the superfluous table to that schema, and leave it. At some point in the future a job would come along and compress it.
If the data was needed within X days, then the table could easily be transferred back to the original schema, no harm: no foul. Also, I would save space as tables would be automatically PAGE compressed and could be decompressed if needed. De/Compression is really fast in SQL Server.
It’s Compression Time
So, this super-simple stored procedure was created prCompressCleanupTables (click for github link). It takes the following parameters:
@CompressBeforeDate – A DATETIME variable that accepts how old the table must be before it is compressed (Looks at the created date)
@Schema – Sysname variable that takes the schema name that you want to compress. Keep in mind that this is the same schema for every database, so make sure it’s unique (I use the ‘Cleanup’ schema personally, hence the name).
It skips the following databases by default: master, tempdb, model, msdb, distribution, ReportServer, SSISDB. It will skip any database that is in any state other than ONLINE, too.
Also remember that compression is locked to certain editions of SQL Server, as well as being 2008+ (you really need to upgrade if being 2008 is a limiting factor).
I’m Also A Client
I have this implemented as a job on several servers which checks weekly for new tables to compress in the appropriate databases. It checks for any tables created prior to GETDATE() – 60. I have to say, that it runs very quickly even on large tables.
Management Data Warehouse is a neat tool for collecting data on your instances with regards to performance, disk usage, and anything else related to SQL Server you might want to know. I’ve recently been rolling it out to various servers in test environments to get a good handle on all that it can do, and maybe more importantly, how it’s going to affect performance.
When I was satisfied that it was working the way I wanted in these environments, I started to push it to what you could call QA. Immediately I was hit with litany of errors, all of which said basically the same thing:
Caught error#: 14684, Level: 16, State: 1, in Procedure: sp_syscollector_create_collection_set, Line: 203, with Message: Caught error#: 14262, Level: 16, State: 1, in Procedure: sp_syscollector_create_collection_set, Line: 80, with Message: The specified @schedule_name ('CollectorSchedule_Every_30min') does not exist. (Microsoft SQL Server, Error: 14684)
Um, what? Of course it doesn’t exist, I’m installing it now. Just to verify, I did some poking around, and I found that it was indeed attempting to add a schedule that didn’t exist (as you would expect) to the collection set. After a while my Jr DBA (read: Google) turned up this article on removing MDW which says, in no uncertain terms:
The fact is that, while these schedules may look like they were created by MDW, they are created by default in a default installation of SQL Server 2008 or above. You can delete these easily, if you really want to, and if you are sure you will never enable MDW again. I would still suggest scripting the schedules and saving those scripts to your file system or source control so that you can recover them in the future.
I took a look back on the TEST instances where I was previously successful and found these schedules existed both on instances where MDW had been installed, and in places where it had not. These were part of the defaut SQL Server install and had been manually removed. Whoops.
As far as I know, there isn’t any direct way to script out schedules, so I could port them to the boxes where they had been removed. To get around this, I added them to the collector set jobs and then scripted that. Then I was able to change a few variables (and remove the schedule_uid) so that this:
I ran these add_schedules on the QA boxes, went through the Configure Management Data Warehouse Wizard again, and tada!
For your convenience, I’ve scripted out the four schedules I needed to get it going below. These are not from the most current version of SQL Server (2016 as of this writing), and are not all of the schedules included out of the box, but these let me get MDW running so I could add my own custom schedules for the remainder.
The moral of this story: Don’t remove MDW schedules unless you really mean it, for all time, forever and ever, pinky swear.
The four ‘core’ schedules I needed, for your convenience: