General availability outage
Incident Report for Buttondown
For around five hours (meaning the early morning of 11/13, Pacific time) Buttondown's availability was heavily degraded.

## What happened?
Around 50-70% of requests timed out. It wasn't _quite_ a complete DDOS, but essentially so.

## Why did this happen?
This is... actually fairly silly, as far as these things go. An old third-party log handler that Buttondown was using shut off access at the logdrain I was using. (This is a totally reasonable thing to do!) _Unfortunately_, that clobbered a huge amount of the requests being served, to the point where all the active dynos on my infrastructure were busy complaining and throwing errors because they couldn't emit logs. The irony of this does not escape me.

## Well, why did it take so long to fix?
I was asleep.

No, really! That's the reason. I've got two thresholds for Buttondown outages:

1. The server is down for a little, which texts me.
2. The server is hard down for all requests, which calls and pages me.

This was an exceptionally long bout of the former, which meant I woke up to like seventy outage texts but no outright pages.

## Why won't this happen again?
First: I've upped (or lowered, depending on how you look at it) the threshold for what constitutes an outage. I made a lot of these alerts two years ago when Buttondown was a fraction of a fraction of its current size; thankfully, things are generally stable, but its still time to be more alert. Any non-trivial breakage of traffic pages little old me.

Second: to fix the _actual_ issue, I'm spending some time this weekend messing around with the logging & error infrastructure Buttondown uses to more gracefully degrade.

Have any questions? Email me:
Posted Nov 13, 2020 - 04:00 PST