High availability for pages file storage backend
On GitLab we maintain a single file server for hosting all customer gitlab pages. For our customers this means the following:
- If the file server is redeployed by our cloud provider as it was in https://gitlab.com/gitlab-com/production/issues/415 we will suffer a full pages outage for the time of the reboot.
- If there is a non-recoverable failure on disk there will be customer dataloss between the time of the failure and the last disk snapshot.
I would like to use this issue to explore options for adding high availability to this service. Given that we are serving static files has it been discussed previously to use an object storage backend for this service?