So, a couple of hours after I wrote the blog post last night saying how everything was going full steam ahead, it all blew up. This morning, many bits of the system which were working just fine are failing. This points to the database, which is at the heart of everything, and all indications are that it broke at about midnight.
I had a poke around, and eventually found the BurstBalance metric. In the top right graph, it’s the orange one that dives into the ground and bounces up.
What it seems to be is that if you overuse your database (in particular, the database’s disk) , you eat into your overuse credits, i.e. the burst balance. And at midnight I ran out of burst balance so the database stopped responding.
Well, that’s something I learned today. At least now I know to watch this when the system is under proper load. It’s also a good indication of when it’s time to fork out for a bigger database.