Yesterday's issue with Calculated fields was a combination of two independent factors: 1) We separated the job execution service that is responsible for calculated fields recalculation into a process separate from HTTP workers. Due to a problem in configuration, an existing firewall rule blocked the traffic between backend nodes and Gremlin servers. The problem was resolved after about 4 hours. Data was fixed on the next field recalculation, and during the automatic nightly one (2 AM UTC) at the latest. 2) A regression in our codebase led to wiping the fields when Gremlin server returned an error.