Making productive retro of a feature that went over estimation

Our team got into the situation in which a feature that was planned to take 2-3 days (estimated as 5 days, because a less experienced developer was working on it) took almost 2 months to build. The feature was in python - previously the developer didn't have experience with it, but it was as fairly simple: mapping an input to external API and mapping the response back.

How would you advise us to do a productive retrospection of the events that happened without turning it into a blame war?