My advice for new Sys Admins

dustda18 · 2012-02-22T21:10:32.881335+00:00

Rebooting is never a solution, but that doesn't mean you shouldn't reboot your servers, anyway.

You want to make sure that your kernel's all patched and up to date? Do you want to make sure that your service will survive spontaneous reboots? Then reboot some of your servers occasionally. Just because it's UNIX doesn't mean that the service isn't built on a house of cards.

badr · 2012-02-22T00:52:41.725953-07:00

8uzahawt1qz28jyv

letmeitellyou · 2012-02-22T06:29:50.119839+00:00

I have to say, I haven't studied math since I graduated undergrad about three or four years ago (and to be honest, I haven't even taken probability while studying math), but I found this lecture really easy to follow and understand, and there's a hell of a lot of motivation that makes it easy to accept why we'd want to go with measure theory.

Thanks so much!

rockhartel · 2 days

Yeah, I agree with lonejeeper - there's a lot there that you should know cold, and it's not long before Wednesday.

Now, assuming you're fine with completely immersing yourself in study material for a day or so, here's what you should probably study:

DNS: why is it significant? How does it work? What happens when I visit google.com and my immediate DNS resolver doesn't have the record cached?
IP: what's contained in the headers? What role does this layer have in the networking stack?
TCP/UDP: what's contained in the headers? What role do these protocols have? What's the steps in establishing a connection with another machine with both protocols?
General networking: try to go as deep as possible when it comes to describing a packet in flight between two geographically dispersed locations. What failure modes can you expect?
Filesystem knowledge - what's a journal? What happens when I 'save' a file?
Windows: GPO's and other ways of managing fleets of machines would be good to know.

Good luck. Don't feign knowledge, admit when you don't know something, and intelligently guess behavior when possible. Write common scenarios and how you'd fix the problem.

MikeCharlieGolf · 2 days

Yep, was on my laptop downstairs. Once they came in, I knew we would be trapped, so I got a ton of smoking in before they started filming. :p

Yeah, wasn't impressed at all by it.

Were you the guy that had only "15 dollars of clothing" btw?

tiff_seattle · 3 days

Getting out of math class. Was walking through the door when the quake happened. Did that whole 'stretch self across doorway' thing. Felt like a boss for remembering that bit.

envinyater · 3 days

I don't know. It really depends on the employer, IMO.

In my personal experience, many UNIX shops (including places like Google, Amazon, and Facebook) do not give a damn about certificates - in fact, lots of these larger internet centric firms figure that the best measure of success and competency in a role are intense eight hour gauntlets and very technical phone screens (I'm not sure if I agree completely with this method of interviewing, but it seems to have worked for them so far, so I figure it's the best thing we have in our deck at the moment.) Other interview styles I've seen that have been pretty effective were take home programming assignments and technical screens that requires ingenuity in order to even contact the hiring manager (think about the puzzle that ITA Software posted all over Boston that required knowledge of QR codes, a ton of knowledge of HTTP requests, and simple ingenuity).

What matters more is the ability to troubleshoot, the ability to program, deep knowledge of system internals, and being creative on the spot. In opposition to skimmedit, I find that a person's interest in the field is easily judged with these sorts of interviews - someone who is not interested in the job will falter very easily.

On the other hand, I've seen many job postings in enterprise environments that want certificates, especially in companies whose core competency isn't in computing. In this case, certificates are a 'good enough' measure of competency without drilling too hard. Of course, you run the risk of running into a 'paper tiger', so this can be a bit risky for the organization, too.

TL;DR: DEPENDS ON THE ORG.

akashani · 6 days

You know, there's the upvote and downvote arrows for a reason.

John Allspaw is also pretty awesome - AFAIK, he's either a principal or a director of ops in Etsy. Generally speaking, he knows what he's talking about, and what he has to say is generally really good advice.

Powelly_ · 7 days

Don't get too stuck with the whole RH thing - yes, it's a bit different than Ubuntu (how?), but not in the ways that matter too much (except for the init systems, maybe.) Learn core UNIX tools instead, and you should be fine.

chrisarchitect · 8 days

I'm not sure if I agree 100% with this at all - I'd like to think that in healthy organizations, this is something that happens anyway, and that if you want to get healthy, you want to encourage this sort of cooperation.

OTOH, there's organizations who try to hire 'devops' who just simply don't get that you just cannot hire a 'devops' and have these things magically spring up from nowhere - it's a culture that must exist already, and is absolutely not a job title. And it's fucking hard to establish this culture, too (it's all about the people - if your people aren't really collaborative in nature and are very judgmental, then it can clam people up in the future.)

There was a talk by Theo Schlossnagle a while ago (I want to say it was for the Velocity conference, but I don't remember) which argued that there doesn't need to be more dev in ops - there needs to be more ops in dev (which I agree with 100%. If there's one thing that Amazon gets right, it's forcing everyone to be a systems engineer, which means, you write more robust code, and you tend to write more carefully. God I'd love this in any organization.)

muiduck · 9 days

Any papers by Amazon, Google, Akamai, etc. are all really good resources (the Amazon Dynamo paper is especially good.) They usually go over real world workloads and their rationale for going with certain design decisions.

Anything in the field of distributed computing is a gold mine.

cparedes · 10 days

One other question - do you have any books that you usually refer people to?

I've been mostly brushing up by looking things up in Google and Wikipedia, but wasn't sure if there's anything else I should read. I figure it's too late to try to go through, say, UNIX Network Programming by W. Richard Stevens, but if there was some kind of canonical reference for a lot of these things that I could refer to, it'd save a ton of time in terms of second guessing some of the hits I'm getting in Google.

cparedes · 10 days

Dude, can't thank you enough for the tips. I feel confident on fairly large picture stuff when it comes to, say, bootstrapping a machine with PXE, Debian preseed files, etc., but still need to read up on specifics. I've been doing a lot of reading on CDN's (reading a few papers from Akamai) as well as on Amazon's Dynamo k/v store (figure I ought to learn about it as a good lens into good distributed system design.)

Also, I used to work in SLU anyway, I'm incredibly jealous of the view you guys have. :p

aloha2436 · 10 days

Stick with CS - as a sysadmin who also writes a ton of application code, I really regret not sticking with CS. That, and it's a necessity if you want a chance at the large Internet companies (Amazon, Google.)

It isn't a deal breaker in most cases, though; you don't need CS to be a good sysadmin. It's just something to keep in mind if you want to work at those large companies someday.

cparedes · 10 days

Hard to say. It's with AWS Cloudfront, which, AFAIK, deals with a lot of machines outside of Amazon DC's. (which makes sense, they are a CDN after all.)

realfacts2 · 13 days

To be fair, the intuition counts for a lot more (when do I really need the device? Could I do this on a napkin even if I'm not precise?) than learning how to operate a calculator.

I use tons of crutches on the job too, but I can usually figure out the gist of what's going on without a calculator.

johnnyneverdies · 13 days

np. :)

johnnyneverdies · 14 days

Do you have scp or rsync available? These are way more conducive for automation than using sftp.

Also, if you aren't allergic to Python, there's FTP libraries available for scripting this out.

dec0o · 15 days

Oh, sorry about that.

I mistyped the sentence just as he was about to clear the dog.

dev_random · 15 days

It sounds like you'll want a combination of a few things:

Artifact management. You're looking at things like Nexus for managing these things (these are things like compiled JAR's or vendor JAR's that you need to use as dependencies.)
More modern source control practices. Use feature branches, hotfix branches, etc. liberally.
Configuration management. After the app is deployed, configure the application automatically.
Continuous integration. Try to get yourself to a point where you don't have to worry about diffing things, as your CI tool should be able to do that automatically.

MBuddah · 16 days

I wish they didn't put vehicles behind the credit system - they added quite a bit to the gameplay in T1 and T2. I haven't really seen anyone use vehicles at all in Tribes: Ascend.

paulc3250 · 22 days

Not really - they're usually pretty harmless.

paulc3250 · 23 days

None. Seattle's a pretty safe place in general.

cparedes

TROPHY CASE

you'll need to login or register to do that

create a new account

login