FetchIntoObjectPool is not operating in a snapshot and writes directly into the repository
FetchIntoObjectPool
seems to not be running properly in a transaction. We see racy access to the repository as transactions fail to snapshot the repository due to missing files and folders. This happens when FetchIntoObjectPool
has been concurrently running. This RPC doesn't have its transaction started in the middleware and has custom handling implemented in !6775 (merged). Given this is the only RPC that seems to be exhibiting this problem, the custom handling likely has an issue.