Am 09.09.2013, 23:01 Uhr, schrieb TylsBurt <a475...@drdrb.com>:
I am incredulous that nobody wants to know *why* the main target of the project (file atomic transactions) cannot be achieved..
Some file operations on some file systems/OS can be atomic (or even transactional) - but a lot of them are not accesible from the java class library (for example you cannot open a file with O_EXCL).
Real (to other components visible) transactions require special file systems (for example TxF NTFS on Windows) which have their own APIs. Using pure java the most transacitional semantics could be implemented with networked filesystems (speaking NFS, iSCSI or DRDB protocols come to mind).
But on local filesystems, even for native code having a portable crash safe io library is next to impossible (especially if you want to also address common OS and Disk problems liek write barriers, reordering and lying devices).
You can of course simulate some multi operation transactional properties (especially if all clients cooperate with the same schema). This is what XADisk and other (locking) schemes expect. This might be enough for some needs, but it is not what others expect (and most likely does not survive all failure modes).
If you only need a transactional storage in a cooperative environment, it would be best to use one of the existing RDB MS.
BTW: the discussion thread seems to be 3 years old. I was actually not aware of the state of commons transaction.file. The project homepage list it as "Dormant", the SVN links exist but dont work. Is there maybe an option to have the site to link to "archive" Download and SVN sites (removing binaries from the regular dl section)?
Gruss Bernd --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@commons.apache.org For additional commands, e-mail: dev-h...@commons.apache.org
