Ignore multiple completions for awakeables

[slinkydeveloper]

Fix #636

[tillrohrmann]

Thanks for creating this PR @slinkydeveloper. Can you quickly add some more context to this PR or the originating issue which panic you saw and where it originated? I guess it was not in the store_completion method. This would help me validating the change.

[slinkydeveloper]

The panic comes from Codec::write_completion. The originating issue is that we end up invoking completeCallback twice (due to the sdk bug restatedev/sdk-typescript#122), so on the second store we invoke write_completion and panic on the debug assert checking whether the entry is already invoked.

Logically, it seems to me more appropriate that we do this check in the state machine, that is in store_completion, as in terms of "entries" and protocol themselves, that debug assert is correct so ignoring the completion in the codec doesn't seem the right place.

[slinkydeveloper]

I also tested this change with restatedev/e2e#171 and it doesn't panic anymore.

[tillrohrmann]

LGTM. +1 for merging.

[tillrohrmann]

Logically, it seems to me more appropriate that we do this check in the state machine, that is in store_completion, as in terms of "entries" and protocol themselves, that debug assert is correct so ignoring the completion in the codec doesn't seem the right place.

I was confused as well because I couldn't find a check whether the entry is completed or not in the store_completion method but also didn't look into the Codec::write_completion implementation. I guess the debug_assert says that the Codec does not allow the duplicate completion of an entry (even though technically it could allow it). Also because the Codec changes the header flags for some entries.