Warning for sites upgrading to 3.1.0 or higher
XRootD releases 3.1.0 or higher use extended attributes to record special meta-data information. If your file system does not allow extended attribute (see note 2) and you are using multiple partitions, all new file creation requests will place the file data in the name space partition. This will likely fill up this partition and make space recovery much more difficult unless you are using extended architecture (XA) space partitions (see note 3).
The only way we have discovered to make sure that your file systems support extended attributes is to specify that you want support in /etc/fstab by addning the user_xattr option as follows:
/dev/partition /mount fstype defaults,user_xattr 1 2
Make the appropriate substitutions for partition (e.g hda3), /mount, and fstype (e.g. ext3, ext4, xfs, etc). After you make the change your can remount the file system as root by issuing:
mount -o remount /mount
If you are running with multiple partitions and use the oss.cache directive without the xa option, (e.g. oss.cache public /data1) you are running with non-XA spaces. Non-XA spaces severely limit how space is managed by xrootd. You should strongly consider either migrating to an XA-space or adding the oss.runmodeold directive to your configuration file in order to run in backward compatibility mode. Failure to do so before restarting a new version of xrootd will make future conversions more difficult.
Please read the “Migrating to Extended Attribute Name and Data Spaces” guide: http://www.xrootd.org/doc/prod/frm_migr.htm#_Toc284350518 before you run servers using release 3.1.0 or higher and if any of the following are true:
- Your configuration file contains an oss.cache directive,
- You have specified the migrate, purge, or stage option on any all.export directive, or
- You have pinned or memory mapped any file.
If you do use a non-XA space we strongly recommend that you convert to an XA space as soon as possible. While not specifically noted in the migration guide, you must change all oss.cache directives to oss.space directives after you do a space conversion.