You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Lately, EC2 spot prices have been fluctuating quite a bit which has caused the lustre and ZFS build farms to become unstable. Currently the solution is to manually choose a new zone and restart the buildbot master. The alternate solution is to stop using spot instances.
We can potentially make EC2 slaves a bit smarter and have them get running averages of the price per zone, then try each zone that is below our max bid price. We can even have the EC2 slaves fall back to using on demand instances if our max bid price can't be met or when no suitable zone exists.
The text was updated successfully, but these errors were encountered:
@behlendorf had another extension to this. In an offline talk, he mentioned that the issue could also be the region we are attempting to start instances in. In our case, US-West appears to be oversubscribed compared to other regions. We could potentially allow EC2 slaves to pick which region to start in and if we want to restrict the region, we can provide a region list in the constructor.
Lately, EC2 spot prices have been fluctuating quite a bit which has caused the lustre and ZFS build farms to become unstable. Currently the solution is to manually choose a new zone and restart the buildbot master. The alternate solution is to stop using spot instances.
We can potentially make EC2 slaves a bit smarter and have them get running averages of the price per zone, then try each zone that is below our max bid price. We can even have the EC2 slaves fall back to using on demand instances if our max bid price can't be met or when no suitable zone exists.
The text was updated successfully, but these errors were encountered: