Archive for August 30th, 2006

Cans and Cannots

Even a lot of tech geeks aren’t good at administering servers. Some of the even admit it, and this is a good thing. 🙂 They (and non-geeks) frequently hire other geeks to administer servers for them, and this too is a good thing. What’s not a good thing is that it’s very difficult for someone who isn’t a competent server administrator to tell who and who isn’t one as well.

Thousands of people have published online tuturials and walkthrus for installing, configuring, setting up, and administering all sorts of useful software packages. This is great, except that it allows people who only sort of know what they’re doing to look like they’re more knowledgable than they are.

Often, you can literally cut-and-paste lines from a website into a unix shell, and get the results you hope for. But software packages change, and different linux distributions differ, and, frankly, there’s no substitute for good old knowledge and experience. The result is someone who finds themselves, metaphorically, in up to the elbows, and unable to tell the heart from the gall bladder. It’s easy to follow instructions – all too easy, really, though many people seem to fail even there – but much harder to know what to do when something goes wrong.

I bring this up because I was today brought in as hired help to fix a piece of software screwed up by an extremely expensive service consultant on someone’s server. The really expensive guy had installed something, in the processing unnecessarily chmod’ing and chown’ing several directories of files. Apparently he realized his mistake, because he then tried to chmod and chown them back to what they were supposed to be.

Only problem was, he seems to have been following instructions on the web, yet thought he was smarter than whomever wrote the webpage, because he left out a few bits. Namely, when he discovered that “chown user group file” didn’t work, he should have used “chown user:group file”. Instead, he just used “chown user file”, which didn’t work. So too with chmod; he apparently knew enough to know that chmod usually takes three arguments, and decided that properties “1755” and “6755” were both typos for the more-familiar “755”. A little knowledge is a dangerous thing.

It wasn’t terribly difficult to fix.

There’s nothing wrong with following instructions online, if you trust the source. But it’s fantastically important that you know not only what you’re doing, but also why you’re doing what you’re doing. If you don’t, you shouldn’t be pretending to be a server administrator – and you definately shouldn’t be charging mid three-digit sums to install a single piece of open-source software.

Published in: 'D' for 'Dumb', Geekiness, General | on August 30th, 2006 | Comments Off on Cans and Cannots

Cutting Corners in Webhosting

I’ve postee occasionally about the problems with the webhosting industry as a whole; all the clueless people who don’t know how to run servers but still do anyway, all the false advertising, and all the other problems that plague the business.

Well, the past couple days have demonstrated another important reason why one should be very, very careful about who’s hosting your website.

A lot of people running RedHat Linux have been reporting that a recent software upgrade from the company has flat-out, outright killed BIND on their servers. BIND, for you who don’t know, is the nameserver daemon, which controls the DNS records for domain names.

You need two nameservers for any domain; it’s a requirement. However, far, far too many companies cheat, and make both “nameservers” the exact same physical machine, answering on two different addresses. It saves money, and is a lot easier to administer. But it’s not just stupid, it’s monumentally stupid.

Why? Because when the RedHat bug hit, people on affected servers suddenly found “both” their nameservers down, and their domains unresolvable. That means their websites won’t load, their email won’t be received, and stuff just generally doesn’t work. All because they chose a dumb, shortsighted hosting company.

If their provider had gone with a better, redundant setup, the slave server would still be responding to queries, and in most configurations would still be responding to queries for seven days after the parent nameserver stopped. That gives the hosting company an entire week to fix the problem without services being down. (Things might be a little slower, as the one nameserver wouldn’t be responding, but 100% of services working at 80% speed is better than 0% working at any speed.)

Until this week, if you asked most of the kids using one machine for all their nameservers about it, they’d dismiss it by saying “well, if the server’s down, then the server’s down, and nothing is going to work anyway, so what’s the point?”. Clearly, the point is that if the server is running, but BIND isn’t, it’d still be good if, you know, something worked.

I consider redundant DNS one of the surest signs of a competent hosting provider. Entropic Memes is hosted by Redpin, who have several nameservers in the US, one in Germany, and one in the UK; our DNS servers are the German, and one of the US ones. In almost two years, we’ve never had a problem, with DNS or anything else.

Sometimes, you get what you pay for.

Published in: 'D' for 'Dumb', Geekiness, General | on August 30th, 2006 | Comments Off on Cutting Corners in Webhosting