Operating system cluster size is generally 4K, and InnoDB page size is generally 16K, then it is possible that 16K pages are not stored in a continuous cluster, so that the scanning software will not scan out such a page. To solve this problem, we decided to add half-page scanning function to the software.
In the first scan, all the page headers are correctly characterized, but the page with incorrect tail features is saved in a separate file and a non-full page information list file is generated.
The second scan (when developing a specialized program), searches for the lower half of the possible partial page based on a list of non-full page information, saves it in another file, and generates a list of page information for the lower half.
Third, do a special program to assemble the full page, verify the correct integrity of the page, and save the correct page in the file.
The first scanning program for the third generation of the file scanning, the page again by index_id classification extracted.
According to this principle, it is theoretically possible to scan a page only if it exists in only two disk fragments.
MySQL InnoDB data Salvage (iii) INNODB cluster non-connected page Scan Extraction (PLAN)