timeout sideeffect is lost when leader dies

Question

timeout sideeffect is lost when leader dies

RoadRunnr opened this issue 4 years ago · comments

The timeout sideeffect is only processed in the Erlang node that is the leader at the time of issuing the sideeffect. If the leader node dies between returning the timeout sideeffect and the actual timeout moment, the the timeout command is not executed.

IMHO the new leader should execute the timeout in this case.

Karl Nilsson · Answer 1 · Tue Jan 05 2021 22:53:01 GMT+0800 (China Standard Time)

You need to re-issue the timer and monitor effects in the optional `state_enter(leader` callback.

On Tue, 5 Jan 2021 at 14:46, Andreas Schultz ***@***.***> wrote: The timeout sideeffect is only processed in the Erlang node that is the leader at the time of issuing the sideeffect. If the leader node dies between returning the timeout sideeffect and the actual timeout moment, the the timeout command is not executed. IMHO the new leader should execute the timeout in this case. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#207>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJAHFBWP5GYISNLEMF3JALSYMQ4VANCNFSM4VVJNJ4A> .

-- *Karl Nilsson*

Michael Klishin · Answer 2 · Tue Jan 05 2021 22:59:57 GMT+0800 (China Standard Time)

This means the new leader would have to have to be aware of all side effects previously issued by the original leader. That would be a massive can of worms to open. More idempotent, "recovering" state machines is a much better idea where you control everything that's going on.