On the timeseries write path, under certain conditions we try to re-open an existing bucket instead of creating a new one. Fetching the bucket to reopen is done by sending a rawData query to the buckets namespace via DBDirectClient.

Since no locks are held, It is possible that this races so that the namespace is the system.buckets collection, but it has since been upgraded to viewless. The query will then find no results.

A priori this is harmless (we will conclude that no bucket can be reopened so we will just create a new one) but we should investigate if fixing allows us to increase the coverage for buckets. (SERVER-121471 made all viewless timeseries upgrade/downgrade suites assume non-canonical buckets layout).