Dynamic movement of the zFS owner

For zFS read/write sysplex-aware file systems, an important aspect of performance is which system is the zFS owner. The zFS owner is the system that does metadata updates to the file system. zFS automatically moves the zFS owner among zFS systems, based on the amount of activity at the zFS owner from each system. The frequency of the dynamic ownership movement can vary, depending on the level of zFS. On z/OS® V1R13 and later systems, ownership moves less often than on systems running previous levels of z/OS. In addition, if an aggregate is read/write sysplex-aware and zFS-owned on a z/OS V1R13 or later system, zFS does not dynamically move ownership to a V1R12 system, even if it is running sysplex=filesys or sysplex=on.)

File requests do not fail as a result of dynamic aggregate movement. New requests are suspended until the aggregate is moved and then requests are allowed to complete. The system produces the following messages (for example):

Source system
22.19.12 DCEIMGVN IOEZ00548I Requesting that DCEIMGVM takeover aggregate PLEX.JMS.AGGR006.LDS0006 LDS0006 
(requests: local 2, new owner 1202 total 1204 

Target system
22.19.12 DCEIMGVM IOEZ00388I Aggregate takeover being attempted for aggregate PLEX.JMS.AGGR006.LDS0006 
22.19.12 DCEIMGVM IOEZ00044I Aggregate PLEX.JMS.AGGR006.LDS0006 attached successfully.

In message IOEZ00548I, local requests is the number of requests on the source system during the measurement period. New owner requests is the number of requests from the target system during the measurement period. Total requests is the total number of requests from all systems during the measurement period. (Total requests can be greater than the sum of the local requests and the new owner requests). This information is provided to aid in problem determination.

For zFS sysplex-aware file systems, zFS aggregate movement is essentially independent of z/OS UNIX ownership movement (except for the cases discussed later in this section). When z/OS UNIX ownership movement occurs because of the MOUNT AUTOMOVE specification (for example, AUTOMOVE or AUTOMOVE(INCLUDE,SY1,SY2) or AUTOMOVE(EXCLUDE,SY1,SY2)), the z/OS UNIX ownership movement is as expected. Because z/OS UNIX sends requests directly to the local zFS, the z/OS UNIX ownership movement does not change the way that the zFS aggregate is accessed. z/OS UNIX ownership movement between zFS sysplex-aware file systems that have local mounts does not change how the file system is accessed.

There are several cases where the AUTOMOVE option of z/OS UNIX does change file system access:

NOAUTOMOVE: When this option is used, z/OS UNIX makes the file system unavailable (unowned). This causes any file access to be denied by z/OS UNIX.
UNMOUNT: When this option is used, z/OS UNIX unmounts the file system (across the sysplex). This causes the file system to be unmounted and any access occurs on the underlying file system.

Guideline: Mount system-specific zFS file systems with UNMOUNT instead of NOAUTOMOVE.

One way to think of the relationship between z/OS UNIX ownership movement and zFS aggregate ownership movement is:

z/OS UNIX controls whether there is any access at all
zFS ownership controls which system updates the metadata.

If a zFS read/write file system is non-sysplex aware, then z/OS UNIX controls movement of zFS read/write mounted file systems as in prior releases for a shared file system environment and the z/OS UNIX owner and the zFS owner are always the same.