Services/Sync/Server/Archived/BackendBraindump
From MozillaWiki
- Load Testing
- We need to come up with a definitive "user" model that we can plug into Grinder to use for load testing. Working through Ragavan to see if we already have this available from metrics
- Operational scaling
- Still a lot of work to be done operationally on the new cluster
- Migration ongoing
- Buildbot and daily monitoring scripts need installation
- Scaling this level of operations is likely to present new challenges
- Even simple monitoring tasks, scaled up to tens of millions of users, become non-trivial
- Still a lot of work to be done operationally on the new cluster
- Quotas
- Disk
- Fumbling towards a two-tier plan
- Overall quota
- Daily max write
- This way nobody can get more than 2x quota in (and, at that point, they can't add anything else until fully cleaning up)
- Scaling this is tightly coupled with scaling monitoring
- Fumbling towards a two-tier plan
- Rate Limiting
- None at present
- Would handle some of the more obvious abuse cases
- Need to make sure it doesn't impact initial sync
- Disk
- Support level for external servers
- What level of support are we willing to dedicate resources to for own-host servers?
- How much internal testing for versions we don't run internally
- How many tiers of server available. How many backends?
- Flexibility vs Maintainability
- Future of Minimal server
- New Platform
- Migration seems like a good plan given Operational cross-team support potential
- Opportunity for a ground-up reworking
- Many old assumptions that went into the old version can be questioned. Sharing?
- Generally need comprehensive approach to services beyond sync
- Keep Reg codebase separate?
- Mongolian Auth?