mwhittaker / craq_bug Goto Github PK

Hello Michael,

Cool corner case for CRAQ GC.
However, isn't this trivially solvable?

Assuming that each replica simply keeps the latest confirmed version (along with the most recent pending version).
Then a reading replica that receives a delayed message from the tail (i.e., its "confirmed" version is higher) it simply returns the higher locally-confirmed version rather than the one specified in the message?

Thanks,
-Antonis