Forum OpenACS Q&A: Response to PostgreSQL and Journaling FS

Collapse
Posted by Don Baccus on
Yes, ext3 will journal data if you ask it to.  This makes it potentially a great choice for general desktop systems where you're using lots of applications that do simple-minded file I/O.  This set-up should be more robust than an ext2 filesystem (though don't make the mistake of thinking you don't need backups!).

But all this extra work done by the filesystem just gets in the way of an RDBMS.

An important point - filesystem journaling can't replace RDBMS REDO/WAL logging.  The write-ahead log is logging changes made to pages in the RDBMS's shared buffer pool that haven't been written to the filesystem. (Oracle uses different terminology but the same technique).

Such changes are naturally invisible to the filesystem, thus to the filesystem journal, even if it is logging data as well as metadata changes (such as is possible with ext3).

So RDBMS ACIDity requires the db server to journal whether or not the filesystem journals data.  You could, I suppose, write each and every change to the shared buffer pool to the filesystem but the performance  implications of doing so are obvious and not pleasant to contemplate.

Hopefully ext2 partitions won't go away entirely.  If journaling filesystems become the only ones available the PG group may be forced to support raw partitions, something they dearly want to avoid.