I dont think we should chunk persist in the same feature flag as signaling errors. They should be handled at 100% as if theyre the only thing that matters and both required to exist for either to exist
Acking a message on receive instead of on completion of computation is insanity and it should be obvious not to do so. A very important part of this is making sure that messages are idempotent (as you are just saying while I am writing this comment 😅). I am baffled that there seem to be major systems out there that auto-ack on receive. What a bad idea. Isn't it blatantly obvious that this is asking for trouble?
Great to see this topic being discussed honestly. SO much BS about this. I feel that exactly once is helped by the Services being designed to be idempotent. Services can be idempotent if they are build build using CQRS where mutations can always happen again without side effects. It would be useful if a video about how to design services was made ???
Asking for application-level idempotence is just irresponsibly pushing the responsibility of idempotence from the message broker to the application. Irrespective of what is enforcing idempotence, it requires deduplication of message IDs somewhere. This requires a lot of storage, increased latency, and decreased throughput. To me it is synonymous with exactly-once processing. Of course it makes sense to record the ID only after the ID has been acked by the application.
You are hilarious dude ha. You should actually talk shit about products. This is why products get better. TH-camrs and influential content educators are the reason they remain motivated to 1) be free and 2) still be competitive.
Good take. Impressive non-use of "baton handoff", "hot potato", or "tag-you're-it" metaphors. 🙂
Found this video through messing around with Bento. 👏good job 👏
Also would be good to explain that NAS Jetstream does have partitioning, as well as traffic shaping ?
I'd love to do some deep dives into NATS and other systems, maybe one day
I dont think we should chunk persist in the same feature flag as signaling errors. They should be handled at 100% as if theyre the only thing that matters and both required to exist for either to exist
Acking a message on receive instead of on completion of computation is insanity and it should be obvious not to do so. A very important part of this is making sure that messages are idempotent (as you are just saying while I am writing this comment 😅). I am baffled that there seem to be major systems out there that auto-ack on receive. What a bad idea. Isn't it blatantly obvious that this is asking for trouble?
To be fair I can imagine having a good time managing lossy systems as long as there are no consequences (for me)
Great to see this topic being discussed honestly. SO much BS about this. I feel that exactly once is helped by the Services being designed to be idempotent. Services can be idempotent if they are build build using CQRS where mutations can always happen again without side effects.
It would be useful if a video about how to design services was made ???
Asking for application-level idempotence is just irresponsibly pushing the responsibility of idempotence from the message broker to the application. Irrespective of what is enforcing idempotence, it requires deduplication of message IDs somewhere. This requires a lot of storage, increased latency, and decreased throughput. To me it is synonymous with exactly-once processing. Of course it makes sense to record the ID only after the ID has been acked by the application.
You are hilarious dude ha. You should actually talk shit about products. This is why products get better. TH-camrs and influential content educators are the reason they remain motivated to 1) be free and 2) still be competitive.