cephadm: allow services to have dependencies on rbd
Currently features like grafana and prometheus use the local filesystem of the host to store data, which binds the service to that physical machine.
In order to provide HA for these services, complex deployments of multiple instance and issues relating to keeping the different copies of the data in sync have been discussed
Instead this feature proposes a different solution that would allow a service to require an rbd image. The rbd image would encapsulate the service, configuration and data - allowing it to be started on a different host in the event of a host outage. This would reduce the data loss to the time it takes to redeploy and attach the rbd image - and the act of redeployment already would handle the reconfiguration of features like dashboard.
#1 Updated by Sebastian Wagner 10 months ago
My two thoughts:
- Prometheus instances are supposed to scrape targets independent of each other, right?
- if you use an RBD image for backing up the monitoring data, your monitoring data is no longer available in case of a disaster. Which by coincidence is a time when you depend on your monitoring data in order to find out the cause of this disaster.
You're really sure we want to implement this? I'd advice against it.