Here’s something that I’m running into and seeing if anyone else has run into it before.
What I’m seeing is sporadic bursts of write activity from my apache servers. Let me take a step back first.
So the config is like this:
- “small.1” instance with 1.7G memory and 1 EC processor
- Linux (Linux 3.2.18-1.26.6.amzn1.i686 i686 in case you’re interested)
- PHP 5
- PHP APC is installed pointing to a shared file on /tmp which is in memory
What caused my initial exploration is looking at the EBS watcher and I’m seeing around 15 writes/s to my volume. (not pushing the drive all)
So I started to look at iotop to start and I didn’t really see anything. Until all at once one of my httpd processes jumped from 4Kb write to over 5Mb write all over the course of a second.
I looked around to see if it had any open files that would explain it with lsof. Nothing caught my eye that would fit the bill.
Ok… so let’s start up an strace on one of the processes and see if we can catch it in the act.
Watching iotop in one window while the trace ran in another I banged on the site a bit and sure enough the process I was tracing blipped up to 6Mb of write.
I pored through the resulting file and found… nothing.
This isn’t a big problem… the IO is going to cost $0.10 a day, but it’s something I’d like to eliminate it I could. More out of principal than anything.
So, does anyone have any suggestions or ideas?