This includes an optimization to the get_bundle_version_files_cached method, which is used very often when loading blockstore data; it was previously being cached only in a process-local cache (lru_cache). My hunch is that in production, with many appservers and LMS workers and frequent deployments and a large number of bundles, the process-local cache is not being hit very often.

I also increased the MAX_BLOCKSTORE_CACHE_DELAY from 60s to 300s; this reduces the frequency with which we check if either (A) an external system modified the blockstore bundle and/or (B) we have a cache invalidation bug somewhere. I am increasing it because that check is more expensive than I thought (calling blockstore API to ascertain latest version of a particular bundle), and I haven't seen any cache invalidation errors that this would help to work around. (Plus, increasing this will make such bugs more obvious.)