Deployment decisions

[phette23]

Local or GCP. GCP comes with a number of further decisions.

Reuse our ES cluster or start a new one? Invenio will deprecate ES so we have to use OpenSearch, we cannot reuse our ES cluster.

See also #2 cloud storage research.

[phette23]

A budget for piloting and a budget for running in production. Use Google Compute calculator. Compute, egress, storage, database.

What do we need to know?

• pilot storage space
• lite storage space
• full prod storage stage
• can we estimate egress of current VAULT? Apache logs?
• ask on the discord / CalTech, how much CPU etc. are people using?
• are any AICAD schools running Invenio? is Northwestern too big to be comparable? Reach out to both Prism address at NW and Eli emailed AICAD list.

[phette23]

CalTech's readme notes that they use "a m6i.xlarge AWS EC2 instance with Ubuntu 20.04". From AWS' product details this is 4 vCPU and 16 GiB RAM. Based on their docs, they're running the main app, REST API, and celery worker as services on the same machine.

[phette23]

Northwestern's Galter Health Sciences Library uses three nodes with these resources:

App: 2 vCPU, 8GB, 180GB HDD
DB: 2 vCPU, 4GB, 60GB HDD
OpenSearch: 2 vCPU, 6GB, 500 GB HDD

We would use GCP's Cloud SQL instead of the db node here. They're running all the services (nginx, UI app, REST API, celery worker, and redis cache) on the app node.