NAME¶

SAPHanaSR-ScaleOut_basic_cluster - SAP HANA System Replication scale-out basic cluster configuration.

DESCRIPTION¶

The SAP HANA System Replication scale-out scenario needs a certain basic cluster configuration. Besides this necessary settings, some additional configurations might match specific needs. Adapting a few SAP HANA settings might be beneficial as well.

* CRM Basics

no-quorum-policy = freeze

The crm basic parameter no-quorum-policy defines how the cluster should act in case of quorum loss. With more than two nodes, the cluster must not ignore the quorum loss. For SAPHanaSR-ScaleOut, an odd number of nodes is required. Setting no-quorum-policy to 'freeze' won't allow the partition to shoot any other node when it doesn't have quorum. Cluster will not be able to add and start new resources, but running will stay alive. If the cluster uses disk-less SBD, the no-quorum-policy 'suicide' is required.

default-resource-stickiness = 1000

The crm basic parameter default-resource-stickiness defines the 'stickiness' score a resource gets on the node where it is currently running. This prevents the cluster from moving resources around without an urgent need during a cluster transition. The correct value depends on number of resources, colocation rules and resource groups. Particularly additional groups colocated to the HANA primary master resource can affect cluster decisions. Too high value might prevent not only unwanted but also useful actions. This is because SAPHanaSR uses an internal scoring table for placing the HANA roles on the right nodes.

concurrent-fencing = true

The crm basic parameter concurrent-fencing allows the cluster to fence more than one node at a time. This helps to reduce the time needed for a take over in case a whole data center is lost. If nodes are fenced one by one, the time needed would be equal to the number of nodes * stonith timeout. With concurrent-fencing enabled the time needed is in the range of 2 * stonith timeout, independent of the number of nodes. See also pcmk_action_limit below.

failure-timeout = 86400

The crm basic parameter failure-timeout defines how long failed actions will be kept in the CIB. After that time the failure record will be deleted. The time is measured in seconds. See also migration-threshold below.

migration-threshold = 50

The crm basic parameter migration-threshold defines how many errors on a resource can be detected before this resource will be migrated to another node. See also failure-timeout.

record-pending = false

The op_default record-pending defines, whether the intention of an action upon the resource is recorded in the Cluster Information Base (CIB). Setting this parameter to 'true' allows the user to see pending actions like 'starting' and 'stopping' in crm_mon and Hawk.

* SBD STONITH Basics

pcmk_action_limit = -1

The sbd stonith parameter pcmk_action_limit defines the maximum number of concurrent fencing actions. It allows parallel fencing of multiple nodes. A value of '-1' means virtually unlimited. See also concurrent-fencing above.

pcmk_delay_max = 1s

The sbd stonith parameter pcmk_delay_max defines an upper limit for waiting before a fencing/stonith request will be triggerd. This parameter should prevent the cluster from unwanted double fencing in case of spilt-brain. A value around 30 seconds is required in two-node clusters. It is not needed in usual SAPHanaSR-ScaleOut setups.

* systemd Basics

saphostagent.service enabled
SAP${SID}_${INO}.service enabled

In case systemd-style init is used for the HANA database, the services saphostagent and SAP${SID}_${INO} need to be enabled and running inside the SAP slice. The instance profile Autostart feature needs to be off. The service saptune is highly recommended, see manual page saptune(8).

* pacemaker service dependency to SAP instance service

[Unit]
Wants=SAP${SID}_${INO}.service
After=SAP${SID}_${INO}.service

In case systemd-style init is used for the HANA database, it might be desired to have the SAP instance service stopping after pacemaker at system shutdown. Therefor a drop-in file for the pacemaker service might help. See examples below.

* pacemaker service basics

PCMK_fail_fast = yes

The parameter PCMK_fail_fast in /etc/sysconfig/pacemaker specifies how pacemaker reacts on failures of its subdaemons. Default "no" means to restart failed subdaemons, while "yes" means fencing the node. Setting "yes" might help to avoid undefined situations. See also SAPHanaSR-alert-fencing(8).
Optional, default no.

* SAP HANA Basics

/usr/sap/${SID}/SYS/global/hdb/custom/config/global.ini

[memorymanager]
final_memory_release_shutdown = [ auto | on | off ]
final_memory_release_crash = [ auto | on | off ]

Starting with SAP HANA 2.0 SPS06, the database shutdown can be accelerated by optimizing memory de-allocation. Please refer to SAP documentation before setting this parameters.

/usr/sap/${SID}/SYS/global/hdb/custom/config/daemon.ini

[daemon]
terminationtimeout = [ millisec ]
forcedterminationtimeout = [ millisec ]

The first parameter defines the timeout from sending SIGTERM to finally terminating child processes when HANA is shutting down by the STOP event. Used also as maximal delay in system restart if 'restartterminationtimeout' parameter is not set. The second defines the timeout from sending the SIGTERM to finally terminating child processes when HANA is shutting down by the QUIT event. See also manual page susChkSrv.py(7). Please refer to SAP documentation before setting this parameters.

/hana/shared/${SID}/global/hdb/custom/config/nameserver.ini

[landscape]
master = node1:31nr1
worker = node1 node2
active_master = node1:31nr1
roles_node1 = worker
roles_node2 = worker

For two-node scale-out HANA without standby nodes, this entry is needed at both sites that are managed by the Linux cluster. The HANA has to be stopped before the files can be edited. Do not copy the file between nodes. It might be necessary to un-register and re-register the secondary to make the change effective on the secondary site. In the example, node1 should be the master namerserver´s hostname, node2 the worker´s hostname, nr the instance number. Crucial is to define only one master namerserver per site, but no candidates. Please check SAP HANA documentation for details.

EXAMPLES¶

* crm basic configuration

Below is an example crm basic configuration for SAPHanaSR-ScaleOut. Shown are specific parameters which are needed. Some general parameters are left out.
This example has been taken from SLE-HA 11 SP4 with disk-based SBD:

The following example is for SLE-HA 15 SP1 with disk-based SBD:

* crm SBD stonith configuration

To complete the SBD setup, it is necessary to activate SBD as STONITH/fencing mechanism in the CIB. The SBD is normally used for SAPHanaSR-ScaleOut instead of any other fencing/stonith mechanism. Example for a basic disk-based SBD resource:

* crm simple IP address resource configuration

Let the Linux cluster manage one IP address and move that address along with the HANA primary master nameserver.

* crm protective IP address resource configuration

Let the Linux cluster manage one IP address. The resource should tolerate short absence of (virtual) network cards. If a monitor and the resulting start fails, the node should get fenced. This protects against risk of HANA dual-primary. A colocation constraint between this IP address resource and the HANA primary master nameserver is needed as well, see respective examples.

This configuration might be used in public cloud environments where virtual NICs might be expected to disappear temporarily. Of course, it would be better to not let NICs disappear in production. I.e. by adding redundancy to the network or by setting the cluster into maintenance.

* crm IP address for active/active read-enabled resource configuration

Let the Linux cluster manage an additional IP address and move that address along with the HANA secondary master nameserver.
Note: This example works for two-node HANA scale-out.

* crm grouped IP address resource configuration

Let the Linux cluster manage one IP address and move that address along with the HANA primary master nameserver. An auxiliary resource is needed for specific public cloud purpose.

* check how resource stickiness affects promotion scoring

SAPHanaSR uses an internal scoring table. The promotion scores for HANA primary and secondary master are in a certain range. The scores used by the Linux cluster should be in the same range.

* clean up SDB stonith resource after write failure

In rare cases the SBD stonith resource failes writing to the block device. After the root cause has been found and fixed, the failure message can be cleaned.

* check saphostagent and show SAP instances

Basic check for the saphostagent.

* check systemd services for the HANA database

In case systemd-style init is used for the HANA database, the services can be checked. Example SID is HA1, instance number is 10.

* show pacemaker service drop-in file

In case systemd-style init is used for the HANA database, it might be desired to have the SAP instance service stopping after pacemaker at system shutdown. A drop-in file might help. Example SID is S07, instance number is 00.

* check for pacemaker dependency to SAP instance service

Example SID is S07, instance number is 00.

BUGS¶

In case of any problem, please use your favourite SAP support process to open a request for the component BC-OP-LNX-SUSE. Please report any other feedback and suggestions to feedback@suse.com.

SEE ALSO¶

ocf_suse_SAPHanaTopology(7) , ocf_suse_SAPHanaController(7) , ocf_suse_SAPHanaFilesystem(7) , ocf_heartbeat_IPaddr2(7) , sbd(8) , stonith_sbd(7) , stonith_admin(8) , crm_no_quorum_policy(7) , crm(8) , crm_simulate(8) , SAPHanaSR-ScaleOut(7) , SAPHanaSR-showAttr(7) , corosync.conf(5) , votequorum(5) , nfs(5) , mount(8) , systemctl(1) , systemd-cgls(1) ,systemd-analyze(1) , systemd-delta(1) , ha_related_suse_tids(7) , ha_related_sap_notes(7) ,
https://documentation.suse.com/sbp/all/?context=sles-sap ,
https://documentation.suse.com/sles-sap/ ,
https://www.suse.com/support/kb/ ,
https://www.clusterlabs.org

AUTHORS¶

F.Herschel, L.Pinne.

COPYRIGHT¶

(c) 2018 SUSE Linux GmbH, Germany.
(c) 2019-2024 SUSE LLC
For details see the GNU General Public License at http://www.gnu.org/licenses/gpl.html