I Had Downtime Today. Here’s What I’m Doing About It. | Hacker News Reader

I Had Downtime Today. Here’s What I’m Doing About It.

kalzumeus.com

117 pointssoundsop12 comments

Is see two threads here. (1) How do you communicate with affected customers, and (2) Root cause analysis, ruthlessly done, with no BS and no excuses.

On (1), this appears to have been an outage that only a subset would have noticed, so the direct message to affected users looks like a good way to communicate. I've seen other ways of handling that, both 'broadcast' and 'denial', and neither is good.

On (2), the OP's use of root cause and the unambiguous identification of cause/effect/remediation is something that others should emulate.

If you cannot identify cause of failure and build mitigation/detection/prevention for every failure, every time you fail, you are destined to spend the rest of you career reacting to broken crap. Determining root cause of every single failure is lots of work, but so is reacting to broken crap every day and night. That gets old after a while.

Thanks, the praise means a lot to me.

One factor I forgot to mention: prevention can be scheduled, incident response cannot. This is really a quality of life issue for me: if I can pick the hour I do the preventative work in, then I can pick an hour where the work will have minimal impact to the parts of my life which happen away from a computer. (All appearances to the contrary, I do occasionally get away from a computer. Not very much these last few weeks, granted, but soon hopefully quite a bit.)

However, outages requiring my immediate attention will probably happen when I want to be doing something else -- like sleeping, or spending time with family, or wooing a young lady, or eating unhealthy food and singing karaoke songs with the guys. Obviously, its in my best interest to minimize that, even more than it is generically in my interest to maximize my productivity.

I've noticed you haven't usually written much about your personal life. But I'm curious, I'll be moving to Tokushima this August, do you live anywhere near there?

I live in the general vicinity of Nagoya. Tokushima is approximately 3.5 hours from Nagoya by the fastest available public transportation, on another island.

Writing up details and actions after an outage like this is a good practice for improving the perception of a service. The compliment to taking re-active action like this (e.g. informing users and make underlying tech improvements) is to try show uptime in a passive way that the users register. My old boss - before I had a startup - would often use the analogy of the London Underground, which for a long time was perceived as a poor service, until they introduced regular status updates on all lines in each station. This made people register all the times that the service was working for them (and many others, by showing the status of all lines), so that when there was a problem they saw it in context as a small proportion compared to 'uptime'. Not sure how you could apply that in this example, but it might resonate with more 'infrastructure' like services like web hosts.

I am sympathetic to the overall thrust of this, but posted on my blog (which vanishingly few customers read) rather than my main site for a reason. If your toaster can't toast, you don't want to hear why your toaster can't toast, you just want it to toast toast. This describes the relationship of almost of my users with their computers, the Internet, and Bingo Card Creator.

While I want to apologize to users who got delayed from getting back to their lives because their toaster was on the fritz, I don't want to tell anybody else that toasters sometimes can't toast toast. It needlessly complicates their relationship with their toaster: they have no relationship with their toaster, and that is how they bloody like it.

Yes that's a good point - I think in most cases you're probably right and simplicity (e.g. attempting to not show irrelevant information) is more important than the telling them that their toaster is toasting.

When I visited London and saw notices with "There is currently a Good Service on All Lines", my immediate reaction was "why is this news?" By announcing their successes it seemed to underscore their users' low expectations of the system.

We recently wrote a blog post about monitoring DJ workers with Webmin that you may find helpful: http://blog.activeinterview.com/post/2010/01/28/use-webmin-t...

Thanks for the suggestion and the blog post. I'll look into whether adding Webmin to the server makes sense. ("Every additional moving part is a future source of failure." -- popular maxim locally.)

Another useful thing to monitor is the size of the queue and the age of the oldest important queue entry. This can catch both 'queue consumers are not running' and 'queue consumers are running but not consuming/doing useful things/running fast enough'.

God really is just a horrible way to manage processes. Either use monit or bluepill, they are MUCH more reliable.

Reply on news.ycombinator.com