Reevaluate 2k Cloud Native Hybrid pod sizings for consistency and memory reasons
For the 2k Cloud Native Hybrid Reference Architecture we chose at the time to switch to 2 worker pods to reduce pod size for fitting into smaller node sizes.
On reflection though memory usage isn't linear in Puma here. It has a standard base amount that doesn't change so it can't easily just be halved as we did here.
Task is to reevaluate the memory usage for the 2k architecture and explore if we should just switch it to 4 worker pods for consistency and memory reasons.
Conclusion (2022-08-11)
After further testing the following changes are being made to 2k:
- Webservice pods have been switched from 2 worker to 4 worker to align with other Architectures and allow for better Puma shared memory management
- Supporting node pool on GCP upped to
n1-standard-2
to align with AWS and give more room for system pods / optional monitoring - Increase of Sidekiq workers / pods from 2 to 4 (Omnibus + Cloud Native Hybrid) to align more with other Architectures
- Switch Sidekiq node pool from 2
n1-standard-2
to 1n1-standard-4
. This seems like an academic change but the numbers in Kubernetes don't work with a split of 2 as each node runs several small k8s services that just tip each over.
Edited by Grant Young