How to bring coordination between file system and database? How to bring coordination between file system and database? database database

How to bring coordination between file system and database?


Access to the file system is indeed not transactional. You will need to simulate an all-or-nothing distributed transaction yourself: if the commit in database fails, delete the file on file-system. Inversely, if writing file fails, rollback database transaction (That will be a bit more complicated, but that's a rough sketch).

Note that it can get pretty complicated when a file is updated. You need first to copy it, so that if the database transaction fails after you've overwritten the file you can still restore the old version of the file. Whether you want to do this depends on the level of robustness that is desired.

Try to enforce that all manipulations go through your application (create, write, delete of files). If you can not do this and you can not prevent a file from being accessed directly on the file system (and maybe deleted), I see no other way than to periodically synchronize the database with the file system: check which file was removed and delete the entry in database. You could create a job that runs each X minute for that.

I would also suggest storing a hash (e.g. MD5) of the file in database. Take a bit of time to compute it, but that has been immensely useful for me to detect problems, e.g. if the file is renamed on file system by error but not in database. That also allows to run some integrity check periodically, to verify nothing was screwed.

If this approach is not sufficient (e.g. you want it to be more robust), I see no other way than to store the binary in the database in LOB. Then it will be really transactional and safe.


An old question I know, but for the benefit of other readers:

Depending on your operating systems you may be able to use Transactional TxF

http://msdn.microsoft.com/en-us/magazine/cc163388.aspx


Treat the two events (managing the reference, and managing the file) as a single transaction. If either one fails, back the other one out. Then you should find it hard to get into a situation where the two are not in sync. It's easier to rollback database operations than filesystem operations.