Map Sync Time about dhara HOT 6 OPEN

from dhara.

There won't be any risk to journal integrity -- it just means that your sync() will occasionally be a no-op. If this is acceptable, do you even need to sync? Every 2**log2_ppc sector writes from FatFS will result in an implicit sync anyway.

from dhara.

Thankyou for confirming! With FatFS, we have found that a f_write will write the data to flash, but it will not update the file size, so if a file is not specifically closed or f_sync command is used and we have a system crash, the file will exist, but will be size zero and the data is not recoverable. http://elm-chan.org/fsw/ff/doc/sync.html

from dhara.

Sorry, that was probably ambiguous of me. What I meant to suggest was continuing to call f_sync as you are, but dropping the call to dhara_map_sync. This way the FatFS metadata will continue to be written out at regular intervals, and you will persist at regular intervals too due to the implicit sync that occurs every 15 sector writes (with a 2kB page size).

The downside is that not every call to f_sync will guarantee to persist, but that's already the case if you are sometimes abandoning dhara_map_sync.

from dhara.

Dropping the call to dhara_map_sync is an interesting idea. We'll try what we have for now, and review your suggestion in a future iteration. I sincerely appreciate your input and responsiveness on this issue.

from dhara.

from dhara.