Hi there!
What little_fish said.
Specifically: When the system finds duplicates in the fingerprint database, the blocks that it suspects to be duplicate (SHA1 algorythm can actually generate collisions against a single occurrence, collisions are highly unlikely, but, for instance, when providing for a hot database, potentially bilions of blocks could change simultaneously, so the chance that the SHA1 algorythm will make a mistake (collision) is low, but still real. So, to make absolutely completely sure, the system will compare the 8 KB blocks to eachother byte against byte. (1000 comparisons of 8 bits each) The chance of making a mistake during this comparison is so close to zero, that we can safely use this method to deduplicate data. This is to make absolutely sure that the system won't delete or flush data that is unique. So... no. Deduplication and compression do not delete valid data in AFA's. Not even in hybrid systems, like OceanStor V3, or V5 (dedupe works the same there)
@little_fish: I see a lot of your posts on the site. I like them a lot :)