Fork me on GitHub

Anyone who tried to visit any of my sites last night, or send me email, probably noticed that I dropped off the face of the earth. I take my uptime pretty seriously - even with virtually no budget, about 10 minutes of UPS time, and everything hosted out of my basement. I’ve only had about a 6-hour block of downtime in the past 21 months, aside from that nothing externally visible except a few sub-3-minute hiccups. And the 6 hours was due to a major power outage that quickly overwhelmed my UPS. Without a generator, I can’t really blame anyone for that. Anyway, last night into this morning I had partial to full outages for about 12 hours. By far the most I’ve had since I literally ripped apart my entire infrastructure, moved it from shelves to a rack, and put it back together on the fly.

I’d like to apologize to the few people whose web sites or other services I host out of my house. On one hand, I think those few people are getting their money’s worth from their free hosting :) On the other hand I know how frustrating it can be, especially when you can’t even resolve DNS and my email is down. Rest assured that this situation deeply bothers me, and I’m already hard at work on plans to at least keep people (both the people I do hosting for and their visitors) informed if something like this should happen in the future.

The Story:

Well, yesterday afternoon as I was at my father’s house about an hour away, I started to get a slew of Nagios notification SMSes (well, email to SMS). They started at 14:22 and stopped after a few minutes, but since I was driving, I didn’t check them. When I got home, I found a scenario I hadn’t really anticipated - the TV worked fine, but I had virtually no connectivity on my cable Internet line. I got home around 16:22 and had spotty-at-best connectivity. I could get some DNS in and out, but it was pretty much impossible to load a web page on my desktop. I was getting Nagios alerts out in bursts, and the postfix queue was pretty full. A cursory inspection showed both routers (mine and Optimum’s) online with link, the Modem had link and was blinking away, but I was passing < 100 Kbps of data. My DOCSIS-MIB checks on the cable modem were spitting back all sorts of bad values. Not good. I checked coax connections, rebooted the modem and router, and went outside to inspect the aerials and the splitters on the outside of the house. Nothing visibly wrong, and no positive change after the reboot. Now the modem wasn’t showing link at all, and I couldn’t even ping its LAN IP.

I called Optimum at 17:12 and went through the initial troubleshooting with the technical support guy. I told him I’d already power-cycled the modem, and gave him a rundown of the status lights. He confirmed that they couldn’t even see the modem on the WAN side, and would have to send a tech out. The big plus to Optimum Business is that, despite it being after 5 PM on a Sunday night, I was given an ETA of 2-4 hours. After about 15 minutes, I power-cycled the modem again, and was able to get link. I was seeing some data pushed through the routers, but only about 50Kbps. I called Optimum back, spoke with another tech, and was told that they couldn’t even get diagnostics back from the modem, and were seeing 94% packet loss on ping. Time to wait for the tech.

The field tech arrived at 17:53. Utterly amazing… about 30 minutes after I got off the phone with tech support. I don’t know if they keep their better techs waiting around for business customers, but this guy - Jason - was one of the most knowledgeable and experienced that I’ve ever met. He poked around the modem a bit, re-did some of the shoddy work that the original installers left, and then climbed the pole to figure out what was going on. About half an hour later, he came back with the bad news. His test scope wouldn’t even lock on to the 609MHz carrier used for the cable modem, so there was something definitely wrong, and it was past the pole in front of the house. He told me they’d need to escalate the problem to the outside plant engineers, but since I was a business customer, I could expect some update or fix in 6-18 hours. He left around 18:30. Well, I was bummed, but I used the time to get other stuff done and start planning for at least minimal DR plans for the future.

According to my off-site Nagios, I at least got some mail out and SSH in from 19:55 to 20:44, and then had another total loss of connectivity. Everything came back around 02:20 today, meaning a full 12 hours of downtime.

Analysis and Future Plans:

Well for the foreseeable future I’m just working my day job and probably not doing much (paid) consulting, so purchasing a backup connection is out of the question - especially since FiOS charges almost twice what Optimum does for static IP service. There’s really no way I could’ve prevented this outage, and it turns out that the problem wasn’t even on my property, so it’s not anything I could have fixed myself (or prevented by convincing Optimum to let me purchase a spare modem to keep on hand). Once again, for something that isn’t directly money-making for me, it’s not really worth it to try and get hosting as a backup, since I’ve got all sorts of complex postfix configurations, BIND master/slave replication, etc. Within my budget, I can’t really say there’s anything I could do to solve this problem, or to get even half of my services back up. My offsite Nagios is behind a dynamic residential cable connection, so that won’t really fix any problems either.

My plan for the short-term is to find a static IP somewhere that I can run a box behind, add it as a NS record, and at the minimum setup a caching Postfix server, a catch-all Apache server with a “we know about it, we’re coming back soon” page, and hacked BIND zone files that point everything at this one box (albeit with a low TTL).

If anyone out there happens to read this, any comments on how to deal with a total loss of connectivity on a budget of, say, $15/month above the cost of my Optimum connection??


Comments

comments powered by Disqus