Prometheus provides the ability to replicate locally stored data to remote storage. This helps solving the following issues:

• Global query view, when you want running a query over data from multiple Prometheus instances.
• Long-term storage. Local storage size in Prometheus is limited by local disk capacity. Remote storage can provide higher data capacity via clustering and better compression.
• Scalability. Prometheus instance cannot automatically scale to multiple computers, while certain remote storage solutions can.

I'd recommend using single-node VictoriaMetrics for the use case because of the following features:

• It is easy to run - just a single binary with the minimum number of command-line flags to set up.
• It supports PromQL, so it can be used as drop-in replacement for Prometheus in Grafana dashboards.
• It is resource-efficient and provides good compression ratio for data.

Seems to be due to bdc046a - the image tag variables in the docker-compose files rely on the makefile executing to populate them. Windows users don't have Make, so they're running docker-compose up directly, so the vars don't load. docker-compose falls back to using empty strings, resulting in an invalid reference format error as they're trying to pull prom/prometheus: for example.

I'm not sure if you'd rather use .env files to apply default tag values rather than store them in a an include in the makefile, or revert the change, or just ignore Windows users.