I identified a race condition when N multiple processes attempt to read the same file at the same time, for the first time.
P_1 -> catfs notices file is missing from cache, begins caching
P_2 ... P_N -> notices cached file is present
P_2 ... P_N -> while reading cached file, file is truncated because loading process from P1 is incomplete.
My solution was to wrap the fd's in a Arc<RwLock> and all of the downstream consequences.
Currently has the downside, that P_2 - P_N block for the entirety of the time it takes to cache the entire file. It should be possible to allow readers up to what has successfully cached.
I identified a race condition when N multiple processes attempt to read the same file at the same time, for the first time.
P_1 -> catfs notices file is missing from cache, begins caching
P_2 ... P_N -> notices cached file is present
P_2 ... P_N -> while reading cached file, file is truncated because loading process from P1 is incomplete.
My solution was to wrap the fd's in a Arc<RwLock> and all of the downstream consequences.
Currently has the downside, that P_2 - P_N block for the entirety of the time it takes to cache the entire file. It should be possible to allow readers up to what has successfully cached.