With great power comes great responsibility
Uncle Ben seemed like a pretty wise dude when when he dropped this particular knowledge bomb on Peter Parker. As sysadmins we should already be aware of the tremendous amount of power that has been placed into our hands. Using tools like SCCM further serve to underline this point and while I think SCCM is an amazing product and has the ability to be a fantastic force multiplier you can also reduce your business’ infrastructure to ashes within hours if you use it wrong. I can think of two such events where an SCCM Administrator has mistakenly done some tremendous damage: In 2014 a Windows 7 deployment re-imaged most of the computers, including their servers at Emory University and another unfortunate event where a contractor managed to accomplish the same thing at the Commonwealth Bank of Australia back in the early 2000s.
There are a few things you can do to enjoy the incredible automation, configuration and standardization benefits of SCCM while reducing your likelihood of an R.G.E.
Dynamic Collection Queries
SCCM is all about performing an action on large groups of computers. Therefore it is absolutely imperative that your Collections ACTUALLY CONTAIN THE THINGS YOU THINK THEY DO. Your Collections need to start large and gradually get smaller using a sort of matryoshka doll scheme based on dynamic queries and limiting Collections. You should double/triple/quadruple check your dynamic queries to make sure they are doing what you think they are doing when you create them. It is wise to review these queries on a regular basis to make sure an underlying change in something like Active Directory OU structure or naming convention hasn’t caused your query to match 2000 objects instead of your intended 200. Finally, I highly recommend spot-checking Collection members of your targeted Collection before deploying anything particular hairy and/or when deploying to a large Collection because no matter how diligent we are, we all make mistakes.
“The bond traders are down! The bond traders are down! Cry and hue! Panic! The CIO is on his way to your boss’s office!” Not what you want to hear at 7:00 AM as you are just starting on your first cup of coffee, huh? You can prevent this by making sure your Maintenance Windows are setup correctly. SCCM will do what you tell it to do and if you tell it to allow the agent to reboot at 11:00AM instead of 11:00PM, that’s what’s going to happen.
I like setting up an entirely separate Collection hierarchy that is used solely for setting Maintenance Windows and include my other Collections as members. This prevents issues where the same Collection is used for both targeting and scheduling. It also reduces Maintenance Window sprawl where machines are members of multiple Collections all with different Maintenance Windows. It’s important to consider that Maintenance Windows are “union-ed” so if you have a client in Collection A with a Maintenance Window of 20:00 – 22:00 and in Collection B with a Maintenance Window of 12:00 – 21:00 that client can reboot anywhere between 12:00 – 22:00. There’s nothing more annoying than a workstation that was left in a forgotten testing Collection with a Maintenance Window spanning the whole business day – especially after the technician was done testing and that workstation was delivered to some Department Director.
I am also a huge fan of the idea of a “Default Maintenance Window” where you have a Maintenance Window that is in the past and non-reoccurring that all SCCM clients are a member of. This means that no matter what happens with a computer’s Collection membership it isn’t just going to randomly reboot if it has updates queued up and its current Maintenance Window policy is inadvertently removed.
Last but not least, and this goes for really anything that is scheduled in SCCM, pay attention to date and time. Watch for AM versus PM, 24-hour time vs. 12-hour time, new day rollover (i.e., 08/20 11:59PM to 08/21 12:00PM) and UTC versus local time.
Required Task Sequences
Of all the things in SCCM this is probably one of the most dangerous. Task Sequences generally involve re-partitioning, re-formatting and re-imaging a computer which has the nice little side effect of removing everything previously on it. You’ll notice that both of those incidents I mentioned at the start of this post were caused by Task Sequences that inadvertently ran on a much larger group of computers than was intended. As a general guideline, I council staff to avoid deploying Task Sequences as Required outside of the Unknown Computers Collection. The potential to nuke your line of business application servers and replace them with Windows 10 is reduced if you have done your fundamentals right in setting up your Collections but I still recommend deploying to small Collections, making your Deployment Available instead of Required (especially if you are testing), restricting who can deploy Task Sequences and password protecting the Task Sequence. I would much rather reboot severs to clear the WinPE environment than recover them from backups.
Automatic Deployment Rules
Anything in SCCM that does stuff automatically deserves some scrutiny. Automatic Deployment Rules are another version of Dynamic Collection Queries. You want to use them and they make your life easier but you need to be sure that they do what you think that they do, especially before they blast out this month’s patches to the All Clients collection instead of the Patch Testing collection. Deployment templates can make it harder to screw up your SUP deployments and once again pay attention to the advertisement and deadline time watching for mistakes with UTC vs. local time or +1 day rollover, the Maintenance Window behavior and which collection you are deploying to. And please, please, please test your SUP groups first before deploying them widely. You too can learn from our mistakes.
Source Files Management and Organization
A messy boat is a dangerous boat. There is a tendency for the source files directory that you are using to store all your installers for Application and Package builds to just descends into chaos over time. This makes it increasingly difficult to figure out what installers are still being used and what stuff was part of some long forgotten test. What’s important here is that you have a standard for file organization and you enforce it with an iron fist.
I like to break things out like this:
It’s a pretty straight forward scheme but you get the idea: Applications – Vendor – Software Title – Version and Bitness – Installer. You may need to add more granularity to your Software Updates Deployment Package folders depending on your available bandwidth and how many updates you are deploying in a single SUP group. We have had good results with grouping them by year but then again we are not an agency with offices all over rural Alaska.
There are a few techniques you can use to prevent yourself from doing something terrible.
Roll-based Access Control
You can think of Security Scopes as the largest possible number of clients a single admin can break. If you have a big enough team, the clever use of RBAC will allow you limit how much damage individual team members can do. For example: You could divide your 12 person SCCM team into three sub-teams and use RBAC to limit each sub-team to only being able to manage 1/3 of your clients. You could take this idea a step further and give your tier-1 help desk the ability to do basic “non-dangerous” actions but still allow them the ability to use SCCM to perform their job. This is pretty context specific but there is a lot you can do with RBAC to limit the potential scope of an Administrator’s actions.
Application Requirements (Global Conditions)
You can use Application Requirements as a basic mechanism to prevent bad things from happening if they are deployed to the wrong Collection inadvertently.
Look at all these nice, clean servers… it would be a shame if someone accidentally deployed the Java JRE to all of them, wouldn’t it? Well, if you put in a Requirement that checks the value of ProductType in the Win32_OperatingSystem WMI class to ensure the client has a workstation operating system then the Application will fail its Requirements check and won’t be installed on those servers.
There’s so much in WMI that you could build some WQL queries that prevent “dangerous” applications from meeting a Requirement of clients outside its intended deployment.
SCCM is a pull-based architecture. An implication of this is once the clients have a bad policy they are going to act on it. The first thing you should do if you discover a policy is stomping on your clients is to try and limit the damage by preventing unaffected clients from pulling it. A simple PowerShell script that stops the IIS App Pools backing your Management Points and Distribution Points will act as a crude but effective kill switch. By having this script prepped and ready to go you can immediately stop the spread of something bad and then focus your efforts on correcting the mistake.
Sane Client Settings
There is a tendency to crank up some of the client-side polling frequencies in smaller SCCM implementations in order to make things “go faster” however another way to look at the polling interval is that this is the the period of time it takes for all of your clients to have received a bad policy and possibly acted on it. If your client policy polling interval is 15 minutes that means in 15 minutes you will have re-imaged all your clients if you really screwed up and deployed a Required Task Sequence to All Systems. The longer the polling frequency, the more time you have to identify a bad policy, stop it and begin rebuilding before it has nuked your whole fleet.
A few simple soft processes can go a long way. If you are deploying out an Application or Updates to your whole fleet, send out a notification to your business leaders. People are generally more forgiving of mistakes when they are notified of significant changes first. Perform a gradual roll-out over a week or two instead of blasting out your Office 365 installation application to all 500 workstations at once. Setting sane scheduling and installation deadlines in your Deployments helps here too.
If you are doing something that could be potentially dangerous, grab a coworker and do pilot/co-pilot for the deployment. You (the pilot) perform the work but you walk your coworker (the co-pilot) through each step and have them verify it. Putting a second pair of eyes on a deployment avoids things like inadvertently clicking the “Allow clients to restart outside of Maintenance Windows” checkbox. Next time you need to do this deployment switch roles – Bam! Instant cross training!
Don’t be in a hurry. Nine times out of ten, the dangerous thing is simple to deploy but the simple settings cannot be wrong. Take your time to do things right and push back when you are given unrealistic schedules or asked to deploy things outside of your roll-out process. In the mountains we like to say, slow is fast and fast is dead. In SCCM I like to say, slow is fast, and fast is fired.
Read-Only Friday is the holiest of days on the Sysadmin calendar. Keep it with reverence and respect.
Consider enabling the High Risk Deployment Setting. If you do this make sure you tune the settings so your admins don’t get alert fatigue and just learn to click next, next, finish or eventually they will click next, next, finish and go “oops”.
I hope this is helpful. If you have other ideas on how not blow up everything with SCCM feel free to comment. I’m always up for learning something new!
Until next time, stay frosty.