We were seeing infrequent bursts of errors due to running out of resources when talking with our database.
This is now resolved, with the team having increased our capacity to handle these requests. We've also made changes to improve the performance of the feature that created this pressure, and don't expect the problem to reoccur.