One week of monitoring show that our instances are largely over-provisioned, at least for the current traffic.
We have downsized the 3 sentries, even with that we can still handle peak of 5 times the current load! In the same intervention we also added stuff to be able to publish memory monitoring on our infrastructure status page. Will be available shortly.
That explain the jump in CPU usage for the 3 sentries:
Track of intervention on bockchain :
- Start : START Resizing servers
- End : END Resizing servers - 3 sentries done - success - no block loss