receptor work distributed protocol

work_protocol.md

Receptor Remote Work Protocol

The diagrams do not capture the current state of the code in Receptor. They are to start a conversation that moves towards desired behavior of Receptor.

Happy Path

Receptor work protocol for work submitted to Bob destined for Alice.

Crash Scenarios

1. The first work unit 123 never makes it to Alice. Upon restart, Bob will resubmit work unit 123 request to Alice.

2. Bob submits work that is already known to Alice. In this case, Alice ignores the work and replies to Bob informing him of this case. Bob then records this for efficiency sake. If Bob were to crash any time after RemoteStarted 123, he would know that work has already been submitted to Alice. Note that even if Bob lost his recording of RemoteStarted 123; it would be fine for him to re-submit work unit 123. The RemoteStarted is a nicety.

Note that the above 2 crash scenarios are really the same scenario; a crash of Bob or Alice occuring before RemoteStarted 123 will result in re-submitting work unit 123 to Alice. This is idempotent behavior.

work_protocol.puml

@startuml happy
Bob->Alice : work unit 123
create "work unit 123"
Alice -> "work unit 123" : work unit 123
Bob->Alice : payload
Alice->"work unit 123" : payload
Bob->Bob: RemoteStarted 123
@enduml

@startuml crash1
Bob ->x Alice : work unit 123
Bob -> Bob : CRASH, RESTART
Bob -> Alice : work unit 123
create "work unit 123"
Alice -> "work unit 123" : work unit 123
Bob->Alice : payload
Alice->"work unit 123" : payload
Bob->Bob: RemoteStarted 123
@enduml

@startuml crash2
Bob->Alice : work unit 123
create "work unit 123"
Alice -> "work unit 123" : work unit 123
Bob->Alice : payload
Alice->"work unit 123" : payload
Bob->Bob: CRASH, RESTART
Bob->Alice: work unit 123
Alice->Bob: work already submited
Bob->Bob: RemoteStarted 123
@enduml