Gitaly server-side backups vs Gitaly snapshots with WAL
This issue is to discuss the evolution of repository backups and how we want to position the upcoming options to customers.
We have three new capabilities the Gitaly team are working on which impact the way customers can take repository backups:
The server-side incremental backups are pretty straightforward. They will push backups directly to object storage from the Gitaly nodes. This is a good improvement over the current incremental backup solution since it doesn't require intermediary storage and processing.
When RAFT and WAL arrive, it is my understanding that it will be possible to use disk snapshots together with WAL. Many cloud vendors offer scalable solutions for disk snapshots. The WAL files bridge the gap between the current time and the last backup snapshot allowing these transactions to be played back on top of the restored backup snapshot to reduce the RPO.
My aim in creating this issue is to discuss the pros and cons of each approach. Are there specific scenarios/use cases where one is more suitable than the other?
cc @jcaigitlab @proglottis @mkozono @brodock @mjwood @juan-silva @andrashorvath