We are paying more for IOPS on slow magnetic storage than the price diff. Should save us about 30% on ssd bill for these nodes(while being faster). We seem to break even(at a slight loss) on builders. Can switch to ssd ebs there as we figure out how to further reduce iops and then switch back.
note even on testers with can experiment with writeout values and possibly switch back to magnetic(though instance storage seems to make more sense as a long term solution)
I made a snapshot for /dev/sda1 on tst-linux64-ec2-001, created a gp2 type volume in the same AZ, detached the old one and attached the new one to tst-linux64-ec2-001. Let's see how it behaves at: https://secure.pub.build.mozilla.org/builddata/reports/slave_health/slave.html?name=tst-linux64-ec2-001 I also changed moz-state tag to not-ready to avoid automatic shutdowns.
So far it took 2 jobs, both worked fine. I'm going to work it over the weekend. I don't want to switch to SSD today for all instances since I won't be near computer on the weekend. :)
http://hg.mozilla.org/build/cloud-tools/rev/ac4154a9d7d6 Soon all of tst spot instances will be using gp2 volume type.
No issues with these today, closing.