Thread: Interesting comments about fsync on Linux
I was looking for some information on how write barriers interact with software raid and ran across the following kernel thread referenced on LWN. The suggestion is that fsync isn't really safe on Linux as it is currently implented. (The thread was from February 2008, so it probably still applies.) http://lwn.net/Articles/270891/
On Sat, 3 May 2008, Bruno Wolff III wrote: > The suggestion is that fsync isn't really safe on Linux as it is currently > implented. The issue of needing to turn off underlying disk write caches under Linux is documented: http://www.postgresql.org/docs/8.3/interactive/wal-reliability.html As for concerns about individual blocks not being written in order, as long as everything associated with a commit is done when the fsync returns that shouldn't be an issue for the PostgreSQL WAL implementation. It is easier to shoot yourself in the foot than it should be on Linux, but it's possible to avoid that with care. -- * Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD