Wondering how data inconsistencies were handled, since you were caching data locally in the application. I believe there would be many instances of the application each can potentially have different version of same timeline. If user requests are distributed randomly among available set of application instances, it could be a problem right, I might see different data with each refresh?
I don't remember the exact details, this feature was a small part of a larger service. The timelines serving infra had multiple teams involved. Im pretty sure the ttl for objects in the application cache was short, you still need to even be popular enough to be cached in the first place, the timeline would need to be receiving a lot of new tweets. But you are right there would be some inconsistency from how I remember it working.
Wondering how data inconsistencies were handled, since you were caching data locally in the application. I believe there would be many instances of the application each can potentially have different version of same timeline. If user requests are distributed randomly among available set of application instances, it could be a problem right, I might see different data with each refresh?
I don't remember the exact details, this feature was a small part of a larger service. The timelines serving infra had multiple teams involved. Im pretty sure the ttl for objects in the application cache was short, you still need to even be popular enough to be cached in the first place, the timeline would need to be receiving a lot of new tweets. But you are right there would be some inconsistency from how I remember it working.