-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Jan Lehnardt wrote: > It helps in case of a server failure. Compaction can start where it left of > after a > restart. I don't know the exact frequency, I just go on assuming it is the > same > 1s we have for delayed_commits. In case of compaction we could probably > do them less frequent.
They should be significantly less frequent (eg every 5 minutes). The issue is that the original database file and the .compact of it are in the same directory by default. With the 1 second fsync you can hear the drive heads seek in the original database and then they have to seek to fsync the .compact file then seek back to the original database. The net result is a very slow read and write speed. This is even worse on Linux because fsyncs on the ext family of filesystems turn into a sync so all outstanding data has to be written to disk. Roger -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.9 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iEYEARECAAYFAktHapgACgkQmOOfHg372QRpXwCfQT/+cGzKBC6Zk6PStFojUbNX RQ4An1TPaE8BomrUpUH71C9qFUKfEjU/ =g/gJ -----END PGP SIGNATURE-----
