elastic: Increase shards_per_node on high-throughput indices from 5 to 8
This is in response to gitlab-com/gl-infra/production#4653 (closed), we want to spread load more evenly across nodes.
Thus, I would like to increase number_of_shards on the busy indices (rails, gitaly, workhorse, sidekiq, and a few more).
Calculations to avoid allocation failures:
- shards: 8
- total shards per node: 3
- hot nodes per zone: 9
- warm nodes per zone: 3
8 shards / 3 shards per node = 2,6 nodes
=> this fits on our 3 warm nodes
We can rollover some of the affected indices after applying this. The expectation is that it will balance CPU utilization across the hot fleet.