3
1.6K
I'm particularly interested in how the rollback plan is designed. If we only discover after launch that certain edge cases weren't covered, can we quickly restore the previous state? Also, are the scope of the canary release and the monitoring metrics sensitive enough? After all, even a one-minute outage in an aviation system has a significant impact.
This kind of change may have a bigger impact on the habits of frontline operators than the technology itself.
The trickiest part of system changes usually isn't the code itself, but compatibility with legacy processes and permission structures. It's only truly solid once that part works end to end.