Open andrewhowdencom opened 4 years ago
Also boot → healthy should be an operation in tracing.
Audit logging; write "interesting" events to a log and keep for some time. Also, as a requirement from audit logging, be able to reconstruct the state from those audit logs.
(Note: This might be simpler to architect just by making the system event driven, or at least have an event as an intermediary state that can be both persisted and is used for internal logic).
Services should have a max (and, logically, a min) bound they're expected to operate at. At limit, drop reqs.
Easy to auto-scale, means fast boot. Boot → available < 100ms I think is reasonable.