This has been finally resolved by our upstream provider. We will be migrating some core infra to more reliable providers (as we've been doing the last little while) to mitigate this in the future.

We've identified that this is a result of redis timeouts in the ord region. We're working on a fix for this.

We're looking into this