And this morning’s lesson is…

Let’s say that around 8pm you noticed that your linode is suffering from a lack of memory. And so you decide that the Mailman processes have gotten bloated and need to be restarted. So you do an /etc/init.d/mailman restart. And let’s further say that as you’re heading off to bed 2 hours later, you still haven’t gotten any messages from any of the mailing lists on that server.

Do you

  1. Assume that everybody suddenly got real quiet, and head off to bed without a second thought? or
  2. Assume that the restart didn’t actually restart, and send a test message to one of the list-request addresses, and when it doesn’t come back, do another mailman restart?

Because last night, I did the first one, and didn’t do the second one until this morning. Which is why on these graphs you’ll see no activity for 11 hours, and then suddenly a big spike.

Sorry, people.

In other news, today I’m wearing my Enemy Combatant t-shirt to mark the death of American democracy. Well, it was nice while it lasted.

It’s here!

My “new” used 1U server is here. It’s a 1U VA Linux 1220 with two 1GHz Pentium IIIs, and 1Gb of RAM.
VA Linux 1220VA Linux 1220 (2)

It only has a 20Gb IDE drive, but I have a 250Gb IDE drive on order from NewEgg which should be here soon. (It also has a built in SCSI controller, if I ever decide I have more money than brains.) The built in CD-ROM doesn’t seem to work – the BIOS recognizes it, but it won’t boot from it. And after I installed using another CD-ROM (which doesn’t fit in the case properly), it won’t mount drives in the built-in CD-ROM. I’ve emailed the vendor asking if he can send me a working CD-ROM – I won’t write his feedback until that’s resolved. Another weird thing, it won’t boot if I have my KVM plugged into the USB port. But that’s not a great hardship.

It’s also really, really noisy. Can’t wait to send it off to a nice rack space somewhere.

I’ve got a Debian 3.1 base system installed on it already. Now to get Xen installed.

Is “Redirect permanent” (301) really that hard to parse?

I just moved my blog from http://xcski.com/blogs/pt/ to https://blog.xcski.com/. I set up a “Redirect permanent” on the old location, so that anybody or anything using the old location will get automatically redirected to the new location. And for a lot of places that used to hit my blog at the old site (like rss feeders) seem to have handled the transition painlessly. For instance, NetNewsWireLite, which I use on my laptop, updated the subscription information and doesn’t appear to hit the old URL at all any more. Same with the LiveJouranl syndication.

But for some reason, a lot of them haven’t handled it correctly. For one thing, my spam load is down to almost nothing. Ok, that’s a good thing. But I also notice that some RSS feeders, like various “Planet” web sites, see a redirection from http://xcski.com/blogs/pt/feed to https://blog.xcski.com/feed, and feel like they can ignore the fact that the hostname part of the URL is different, and try to fetch the redirection at http://xcski.com/feed, which of course isn’t a valid URL. It can’t be that they have an old DNS cache entry for blog.xcski.com – it didn’t exist until a few days ago. It must be that they’re trying to be too clever, or not clever enough.

Colocation

I’ve decided to go for it.

I’m currently paying about $30 a month for a Virtual Private Server at linode.com. For that $30 a month I get the equivalent of 150Mb of RAM, 4Gb of disk, and a static IP that isn’t on any black lists. I’m guaranteed at least the equivalent of a 150MHz, but up to the maximum of the actual CPU on the machine, which is dual 3.2GHz Xeons.

The applications I run on there, particularly the Waypoint Database Generator, are severely limited by the memory limitations. I don’t come close to using all the bandwidth, and I’m not usually CPU bound. But as soon as that generator starts the memory swapping, it would use up all my io_tokens, and then the application would get i/o bound. The situation was totally untenable until somebody on linode started running a MySQL server and allowed me to use it. But I don’t like relying on this service that might go away at any time.

I’ve discovered that for about a hundred dollars a month, I can put a 1U server on a local colocation service. By going the colocation route, I can put in as much memory and disk space as I want. I still get the other advantages of the linode, such as a static IP and better bandwidth to the rest of the internet. The extra disk and memory means that I can run my own database server without getting swap and i/o bound. It also means I can move some of the disk hungry stuff like my Gallery server and this blog to it.

Yeah, it’s more expensive, but what I’m planning to do is set up the system with Xen. Then I can divide the box up into multiple virtual machines and sell shares. I’m hoping I can sell one share to start, and maybe another later on.

I’ve been perusing eBay, and it appears that $300 or so should get me a machine with 2x1GHz processors, 1Gb of RAM, and no hard drive. Another $125 or so for a 250Gb drive for the machine and another 250Gb drive to keep at home as a spare. Since the colocation site is here in town, I don’t need the ultra-reliability of the latest servers, RAID arrays, redundant power supplies, etc.

Ok, so I have to buy this machine, build it, and start setting up Xen. Then get my applications working on it. I’m also going to move this blog to “blog.xcski.com” and the gallery to “gallery.xcski.com” to facilitate moving them to the colo.

Maybe this will make Gravatars suck less

This morning, gravatar.com is throwing a wobbly, which it does with depressing regularity. When that happens, the front page of my blog loads fine, but anytime you click on an entry with comments it takes forever. I was wondering if there was a better solution, and a quick google reveals the presense of a “Gravatar2” plugin.

This one caches gravatars for all the people who’ve commented on your blog, and instead of going off to consult gravatar.com when a user views a post, it does it out of a cron job that runs in the background. This means the users don’t get slow page loads when gravatar.com is responding slowly. I guess I’ll have to wait and see if it actually loads and displays them properly when that site is responding again