As stated in FabricPool, if blocks in S3 object storage got ransom read, then the block will be brought back to the performance tier, sequential read would not bring the block back, and will read directly from S3.
So, this comes to my question, should a RMAN restoration be a sequential or random read, from the SSD aggregate with attached S3 object? suppose blocks are already in S3.
As my understanding to SSD, the write could write to any new pages without data on, so, initially, all pages are new/blank, so these write/read should be sequential, and would not reheat blocks in S3. Further, Write would not overwrite or delete if there are available new/ blank pages. When all blocks are full, then it would seek for what page can be deleted. Based on this process, after some time, the later writes can scattered to any place in the SSD drive, then later write/read will become random, and then read to blocks in S3 will be reheat, and brought back to the local tier.
If my understanding is not correct, restorations should be all sequential reads as we all believed, then there would be no reheat or back to the local tier. The question is why only about 20-30% of backups have been in S3? Majority data should already exceeded minimum cooling days(3 days), and should be tiered. This is the thing I couldn't figure out.