Specifying when structure rebuilds occur after connectivity is lost

The rebuild threshold is a percentage that you specify for the REBUILDPERCENT parameter in the CFRM policy. z/OS® uses the REBUILDPERCENT value to determine whether to initiate a structure rebuild in the case of a loss of connectivity to the coupling facility that contains the structure.

About this task

The percentage is based on the sysplex failure management (SFM) weights of all the systems that have active connections to a structure at the time. (A connection is considered active even if connectivity is lost.) You also specify weights on the SFM policy.

Db2 calculates the total weight of:

  • All systems that have active connections but that lost connectivity to the structure (A)
  • All systems that have active connections to the structure (B)

z/OS then divides A by B and compares the result to the REBUILDPERCENT value. (In the case of multiple connections to a structure from a single z/OSsystem, that system is counted only once. Multiple connections from a single z/OS system exist, for example, when two members reside on the same z/OSsystem.)

Procedure

To control when a rebuild occurs:

Specify REBUILDPERCENT values in your CFRM policy.
Recommendation: If you have high availability requirements, specify a small REBUILDPERCENT value, and allow the system to rebuild the structure. Specify a larger REBUILDPERCENT value only if you prefer to have a member lose the use of a structure rather than temporarily disrupt all of the members that were using the structure so that the structure can be rebuilt.

Example

All z/OS systems in an eight-member Parallel Sysplex® are of equal weight (10), and six of the systems have active connections to the group buffer pool. A equals the total weight of all systems that lose connectivity to the group buffer pool. B equals the total weight of all systems that have active connections to the group buffer pool. If two of the six systems lose connectivity, the value of A is 20, and the value of B is 60.

z/OS determines whether to initiate a rebuild as follows:
  • If (A/B) × 100 is greater than or equal to the REBUILDPERCENT value, Db2 rebuilds the structure.
  • Otherwise, Db2 does not rebuild the structure. Instead, the affected members disconnect from the group buffer pool.

In this example, (20/60) × 100 is equal to 33. If the value of REBUILDPERCENT is 35, z/OS does not initiate a rebuild.

What to do next

You can monitor a performance class 20 trace (IFCIDs 0267 and 0268) to determine how long a rebuild of a structure takes and the reason for the rebuild.

You can also examine the messages that are returned to the console to monitor a rebuild. The reasons for a rebuild can include:
  • Lost connectivity
  • Operator command
  • Duplexing is being established