Managing Serviceguard Extension for SAP Version B.05.10, December 2012

The SGeSAP packaging of the ERS Instance provides startup and shutdown routines, failure
detection, split-brain prevention and quorum services to the mechanism. SGeSAP also delivers an
EMS (HP-UX Event Monitoring Service) that implements a cluster resource called
/applications/sap/enqor/<SID>ers<INSTNR> for each Replicated Enqueue in the cluster.
Monitoring requests can be created to regularly poll the status of each Replicated Enqueue.
The EMS monitor can be used to define a resource in the Serviceguard packages. This implements
a follow-and-push behavior for the two packages that include enqueue and its replication. As a
result, an automatism will make sure that enqueue and its replication server are never started on
the same node initially. Enqueue will not invalidate the replication accidentally by starting on a
non-replication node while replication is active elsewhere. It is possible to move the package with
the replication server to any free node in a multi-node cluster without a requirement to reconfigure
the enqueue package failover policy.
During failover of enqueue, its replication will be located dynamically and the enqueue restarts
on the currently active replication node. Enqueue synchronizes with the local replication server.
As a next step, the package with the replication service shuts down automatically and restarts on
a healthy node, if available. In case of a failover in a multi-node environment, this implements a
self-healing capability for the replication function. Enqueue will failover to just any node from the
list of statically configured hosts if no replication package is running.
Two replication instances are required if Enqueue Replication Services are to be used for both the
JAVA stack and the ABAP stack. From this approach, several configuration options derive. In most
cases, it is the best practice to create separate packages for ASCS, SCS, and the two ERS instances.
It is also supported to combine the replication instances within one SGeSAP package. To combine
ASCS and SCS in one package, but only if the two ERS instances are combined in another package
is not supported. To combine ASCS and SCS in one package and keep the two ERS instances in
two separate packages is not supported. Otherwise, situations can arise in which a failover of the
combined ASCS/SCS package is not possible. Finally, ASCS cannot be combined with its ERS
instance (AREP) in the same package. For the same reason, SCS cannot be combined with its ERS
instance (REP).
The sgesap/sapinstance module can be used to cluster Enqueue Replication Instances.
Furthermore, SGeSAP offers the legacy package types rep and arep to implement enqueue
replication packages for JAVA and ABAP.
SAP offers two possibilities to configure Enqueue Replication Servers:
1. SAP self-controlled using High Availability polling.
2. Completely High Availability failover solution controlled.
SGeSAP provides a completely High Availability failover solution and controlled implementation
that avoids costly polling data exchange between SAP and the High Availability cluster software.
There are several SAP profile parameters that are related to the self-controlled approach. Most of
these parameters have names that start with the string enque/enrep/hafunc_. They will not
have any effect in SGeSAP clusters.
Dedicated NFS Packages
Small clusters with only a few SGeSAP packages usually provide HA NFS by combining the HA
NFS toolkit package functionality with the SGeSAP packages that contain a database component.
The HA NFS toolkit is a separate product with a set of configuration and control files that must be
customized for the SGeSAP environment. It needs to be obtained separately.
HA NFS is delivered in a distributed fashion with each database package serving its own filesystems.
By consolidating this into one package, all NFS serving capabilities can be removed from the
database packages. In complex, consolidated environments with several SGeSAP packages, it is
of significant help to use one dedicated HA NFS package instead of blending this into existing
packages.
14 Designing SGeSAP Cluster Scenarios