Don’t call it a streak, but it’s been almost five days since MySQL crashed. I made a change late last week that masks the problem appropriately. I still don’t know what causes the problem, and probably never will, but as long as the server stays up, I’m in a better place. I’ve been fighting this for over a year.
I run DDoE on a 1GB virtual server at Digital Ocean. The heart of the problem is that a memory usage spike causes the Ubuntu kernel to shut down the MySQL service and it never gets started again. Memory management is one of the things a kernel does, so it’s not strange that services get kicked out of RAM when there’s a spike. MySQL will try to restart itself, but it’s not able to because it doesn’t have enough memory. I don’t know what’s causing the memory spike. I do know that the only services of consequence that are running are MySQL and a bunch of Apache2 services. Apache2 is the web server that directs the traffic.
I don’t recall where I read it, but someone on the internet said that it’s likely Apache2 is spawning too many processes. I don’t know how to verify that. I followed the advice on ServerFault on how to tune Apache on Ubuntu. When I shut down Apache, I could see that I was using about 660MB of memory, leaving 340MB free. I was surprised how much memory it was using without Apache, which I assumed would be the biggest component. Using the
command, I could see that an Apache2 process was using 30MB on average. That means if I limit my Apache2 processes to 10, I should not run out of memory. I changed the MaxRequestWorkers parameter from 150 to 10, and so far it’s worked.
There must be a cost to that change. I imagine that when the memory spikes, for whatever reason, and Apache isn’t allowed to spawn more processes, that the user gets an error and is unable to reach the site. As bad as that is, it’s better than what was happening – MySQL would shut down and nobody could reach the site until I got up the next morning and rebooted.
Now that we’re seemingly stable, maybe I’ll make some posts. I’ve been working in SQL Server a lot lately and have a lot to say about it.