Ozone Manager records writes first to its cache during any write operation and this cache is flushed at regular intervals by the DoubleBuffer daemon.
A filesystem is expected to be strongly consistent and hence reads that follow these writes will hit the cache and the cache has to return the correct results.
The results returned by the cache should be exactly as if the results are returned by the table. i.e there shouldn't be any difference in logic while processing the reads all from cache vs table.
Such inconsistencies can cause intermittent failures in reads as reading from cache results one result while from the table another.
This jira aims to solve such cases.